Post on 05-Nov-2018
MPRAMunich Personal RePEc Archive
Lifeline or Means-Testing ElectricUtility Subsidies in Honduras
Quentin Wodon and Mohamed Ishan Ajwad and Corinne
Siaens
World Bank
2003
Online at httpmpraubuni-muenchende15419MPRA Paper No 15419 posted 26 May 2009 0012 UTC
277
8
Lifeline or Means-Testing ElectricUtility Subsidies in Honduras
Quentin Wodon Mohamed Ihsan Ajwadand Corinne Siaens
(c) The International Bank for Reconstruction and Development The World Bank
81 Introduction
Many countries around the world have implemented subsidies for utilityconsumption especially in the case of water and electricity Most subsi-dies take the form of a lifeline or increasing block tariff whereby house-holds that consume less pay less on a unit basis The idea is that house-holds with low consumption levels are likely to be poor and someintervention is warranted to enable them to meet their basic needs (the life-line) at an affordable cost Whether such subsidies are successful at help-ing the poor is not clear as illustrated by the experience of a number ofCentral American and Latin American countries
Goacutemez-Lobo and Contreras (2000) suggested that in Chile and Colom-bia errors of exclusion (poor households not benefiting from the subsidy)and inclusion (nonpoor households benefiting from the subsidy) for waterand electricity subsidies are large (for a review of this and other studiessee Estache Foster and Wodon 2002) In Colombia dwelling and neigh-borhood characteristics are used as proxies for income in a six-tier classi-fication of households Households classified as belonging to the lowerstrata get a percentage reduction in their water and energy bills that isfinanced by a surcharge for households from the upper strata Subsidiza-tion takes place primarily within each utility that is the subsidies arefinanced by higher tariffs on unsubsidized customers but the governmentalso provides public funds to utilities in areas with few high-income house-holds Because eligibility rules are not stringent the subsidy reaches alarge share of the poor that is the rate of errors of exclusion is low but thisleads to large errors of inclusion up to 80 percent of beneficiaries are non-poor households depending on the definition of who is poor
In Chile eligible households receive a subsidy of 20 to 85 percent of thewater bill for the first 15 cubic meters of monthly consumption The target-ing mechanism is based both on regional data on water consumption andtariffs and on household characteristics as measured by a national means-testing system While errors of inclusion for nonpoor households are lowerthan in Colombia the errors of exclusion are much higher at up to 80 per-cent In addition research by Clert and Wodon (2001) suggests that the
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
278
The research was funded through a grant from the Energy Sector Management Assistance Programunder the project Low Income Energy Assistance in Latin America
(c) The International Bank for Reconstruction and Development The World Bank
water subsidyrsquos overall performance in reducing poverty and inequality issubstantially weaker than that of other social programs targeted using thenational means-testing system These other programs include noncontribu-tory pensions for the elderly poor family allowances for poor families withchildren or pregnant women subsidized child care centers and housingsubsidies Given that their impact on poverty and inequality tends to belarger than that of utility subsidies this raises the question of whetherkeeping these subsidies is worthwhile or whether the funds used for thesubsidies should be invested in these other programs
In Guatemala following reform of the electricity sector prices for resi-dential customers increased after 1998 in some case by up to 85 percentover a three-year period To ensure affordability for poor customers thegovernment introduced a social tariff for those consuming less than 500kilowatt hours (kWh) per month In January 2001 the government reducedthe lifeline to 300 kWh per household per month This threshold which issimilar to the threshold in Honduras examined in this chapter is still highbecause an average household consumes only 102 kWh per month Anevaluation by Foster and Araujo (forthcoming) suggests that two-thirds ofbeneficiaries are not poor and because the nonpoor consume more elec-tricity 90 percent of the funds go to the nonpoor
Finally in El Salvador until recently the Electricity-Telephone Invest-ment Fund subsidized residential electricity consumption for householdsconsuming less than 200 kWh per month The subsidy was a flat 75 per-cent reduction on the bill paid by the fund to the electricity company everysix months An additional subsidy came from the payment of the consump-tion tax (13 percent) on that share of the householdrsquos consumption As inother countries this lifeline subsidy had high leakage rates to the nonpoorand it benefited mostly urban households The subsidy was recently termi-nated
In this chapter we assess the targeting performance of a similar subsidyfor electricity implemented in Honduras Honduras is the second largestcountry in Central America and one of the poorest in the Western Hemi-sphere (World Bank 2001) The country has 65 million inhabitants Aspart of its efforts to reduce poverty the government is funding two largesubsidies for basic infrastructure services The first subsidy is for electric-ity consumption which in 1998 cost the government L 259 million (orUS$175 million at an exchange rate at that time of L 1480 to the US
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
279
(c) The International Bank for Reconstruction and Development The World Bank
dollar) The second subsidy with a cost of L 114 million in 1990 is for bustransportation in the capital city of Tegucigalpa These subsidies are largein comparison with other social programs because only onemdashthe Hon-duras Social Investment Fundmdashhas a larger budget Here we focus on theelectricity subsidy which the state paid to the national electric utility onbehalf of beneficiaries
The subsidy is targeted through the lifeline principle however becausethe consumption threshold for eligibility is relatively high (300 kWh permonth) and because those with access to electricity tend to be less poorthan those without access the programrsquos overall performance is low interms of poverty reduction Targeting through means-testing rather than alifeline or at least a lower threshold for the lifeline could help improve theimpact of the subsidy and based on experience in other countries wouldnot necessarily imply high administrative costs
The rest of the paper is structured as follows Section 82 describes thesubsidy scheme and assesses its performance Section 83 illustrates howsimple techniques based on so-called receiver operating characteristics(ROC) curves can be used to show how much improvement might comefrom an alternative way of targeting the subsidy Section 84 concludes
82 Targeting Performance of Hondurasrsquos Electricity Subsidy
Electricity access rates in Honduras can be estimated using the PermanentMultiple Purpose Household Survey (Encuesta Permanente de Hogares dePropoacutesitos Muacuteltiples or EPHPM) This is a nationally representative laborforce survey implemented with support from the US Agency for Interna-tional Development and conducted by the General Directorate for Statisticsand Censuses The sample consists of 6423 households stratified into fourgeographic regions Tegucigalpa San Pedro Sula other urban and rural
As table 81 reveals Empresa Nacional de Energiacutea Eleacutectrica (ENEE)the public provider of electricity is by far the predominant provider ofelectricity for households A few households obtain their electricity fromcollectives or privately owned electricity generators Clearly access to elec-tricity varies a good deal by income group and location As expectedhigher-income households have higher access rates than lower-income
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
280
(c) The International Bank for Reconstruction and Development The World Bank
281
TAB
LE 8
1 A
cces
s to
Ele
ctri
city
by
Inco
me
Gro
up
Ho
nd
ura
s 1
999
(per
cen
t)
Inco
me
dec
ile
Loca
tio
n a
nd
so
urc
e1
23
45
67
89
10M
ean
Nat
ion
al
Non
e75
065
346
533
525
920
416
19
76
64
730
4
ENEE
245
347
532
657
739
787
821
899
919
933
688
Colle
ctiv
e0
00
10
30
50
00
70
70
30
40
50
4
Indi
vidu
ala
05
00
00
03
02
02
11
01
11
15
05
Urb
an
Non
e31
016
610
09
04
01
91
40
40
20
47
5
ENEE
691
834
901
907
960
981
986
996
997
994
925
Colle
ctiv
e0
00
00
00
30
00
00
00
00
10
20
1
Indi
vidu
ala
00
00
00
00
00
00
00
00
00
00
00
Ru
ral
Non
e80
276
759
450
850
046
840
733
823
325
848
8
ENEE
193
232
401
480
496
508
547
648
715
633
485
Colle
ctiv
e0
00
10
40
60
01
81
81
11
11
90
9
Indi
vidu
ala
06
00
00
06
04
06
29
03
41
90
19
a S
mal
l pri
vate
ly o
wn
ed g
ener
ato
rs
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
Mar
ch 1
999
EPH
PM s
urv
ey
(c) The International Bank for Reconstruction and Development The World Bank
households For instance 245 percent of all households in the 1st or poor-est decile have access to electricity provided by ENEE compared with 933percent of households in the 10th richest decile Also as expected house-holds in urban areas have higher access rates than households in ruralareas The fact that richer households have higher rates of access to elec-tricity than poorer households is one of the reasons why from the point ofview of poverty reduction and in comparison with other programs thatcould be implemented to fight poverty electricity subsidies may not be welltargeted
The design of the electricity subsidy in Honduras follows a basic self-targeting structure that is households self-select themselves for inclusionin the subsidy scheme by not consuming more than a certain amount Thisimplies that neither the utility nor the government needs to design a morecomplex targeting mechanism based on means-testing whereby they usehousehold characteristics as proxies to identify who is poor and who is notand thereby who is eligible to receive the subsidy and who is not As of July2002 the subsidy was provided to all households that purchased electricityfrom ENEE and consumed less than 300 kWh per month the electricityconsumption lifeline Specifically households consuming less than 300kWh paid the price of electricity at the 1994 rate plus an increase of 1631percent approved in 2000 Households that consumed more than 300 kWhper month paid the full price at the 1997 rate plus the same 1631 percentincrease The government paid the difference between the 1994 and 1997prices (L 238 million per month in mid-2002) directly to ENEE1
To assess the targeting performance of the subsidy that is what share ofthe subsidy goes to the poor we combine administrative information withhousehold survey data On the administrative data side we rely on infor-mation provided by ENEE This information includes (a) the percentage ofresidential clients by consumption level (b) the average amount of elec-tricity consumed by clients at each level (c) the total electricity bill with-out subsidy by level and (d) the total subsidy paid by level
On the household data side to estimate the proportion of poor householdsat the various consumption levels we use data from a survey implementedin 1999 by the Family Assistance Program (Programa de AsignacioacutenFamiliar or PRAF) with financing from the Inter-American DevelopmentBank and assistance from the International Food Policy Research InstituteThe survey was implemented in the bottom half of Hondurasrsquos municipali-
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
282
(c) The International Bank for Reconstruction and Development The World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
283
ties in terms of malnutrition as measured by the Ministry of Educationrsquosannual census of weight and height during the first year of primary schoolThe survey has modules on expenditures education health and the impactof Hurricane Mitch The advantage of this survey is that it includes dataon electricity consumption which are not available in the labor force sur-vey The disadvantage is that the survey is not nationally representativebut we assume that the results obtained here would also apply if the surveywere nationally representative2
Table 82 provides the results For example according to ENEErsquos admin-istrative records the share of households connected to its network withmonthly consumption below 20 kWh is 2031 percent (115723 house-holds) Of these we estimate from the PRAF survey that 4493 percent arepoor That is among all households that consume less than 20 kWh accord-ing to the PRAF survey about 45 percent have a level of per capita con-sumption that is below a reasonable poverty line for Honduras With aver-age consumption of 336 kWh per household per month the totalconsumption for this group is 388626 kWh Without the subsidy this groupwould have to pay a total bill of L 929256 but this bill is reduced toL 333282 when the subsidy of L 595973 is taken into account
Presenting statistics on errors of inclusion and exclusion is commonwhen assessing the targeting performance of a subsidy (see for exampleCornia and Stewart 1995) There are various ways to present these twotypes of errors The simplest way used in table 82 consists of computingthe share of households in poverty and receiving the subsidy as well as theshare of households not in poverty that are also receiving the subsidy Ingeneral as errors of inclusion increase errors of exclusion decrease andvice versa If all households receive the subsidy there are no errors ofexclusion but the errors of inclusion will likely be large because all non-poor households benefit from the subsidy
In table 82 among those households that are connected to ENEE6019 percent are nonpoor and receive the subsidy By contrast only 168percent are poor and do not receive the subsidy (nationally the share ofthe population in poverty not receiving the subsidy is larger becausemany poor households are not connected to the grid) In addition 2328percent of households that are poor receive the subsidy and 1485 percentof households that are nonpoor do not receive the subsidy From theseresults we can estimate that the ratio of poor versus nonpoor beneficiaries
(c) The International Bank for Reconstruction and Development The World Bank
284
TAB
LE 8
2 T
arg
etin
g P
erfo
rman
ce f
or
the
Elec
tric
ity
Sub
sid
y H
on
du
ras
Mid
-200
2
Tota
l bill
Sh
are
of
Co
nsu
mp
tio
nSh
are
of
clie
nts
Er
ror
of
A
vera
ge
wit
ho
ut
Tota
lsu
bsi
dy
leve
l Sh
are
of
in p
ove
rty
incl
usi
on
Er
ror
of
con
sum
pti
on
sub
sid
y b
y su
bsi
dy
by
spen
t o
n
(kW
h p
er)
clie
nts
b
ased
on
199
9 (1
)[1
-(2)
] ex
clu
sio
n
(kW
h p
er
cate
go
ryca
teg
ory
no
np
oo
r m
on
th)
()
(1)
dat
a (
) (2
)(
)(1
)(2
) (
)m
on
th)
(L t
ho
usa
nd
s)(L
th
ou
san
ds)
ho
use
ho
lds
()
0ndash20
203
144
93
111
8n
a3
3692
959
61
38
20ndash1
0022
69
356
614
60
na
586
75
096
271
77
34
100ndash
150
126
316
82
105
0n
a12
509
738
73
762
131
4
150ndash
200
111
610
98
994
na
175
3510
314
514
919
24
200ndash
250
925
156
47
81n
a22
454
116
185
746
203
5
250ndash
300
743
170
96
16n
a27
577
118
965
851
203
6
300+
165
310
15
na
168
mdash
mdashmdash
mdash
Tota
l10
024
96
601
91
6810
858
472
4123
820
818
1
mdash N
ot
avai
lab
le
na
No
t ap
plic
able
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
EN
EE (
2002
) an
d t
he
1999
PR
AF
surv
ey d
ata
(c) The International Bank for Reconstruction and Development The World Bank
is 039 (23286019) and so the number of nonpoor households receivingthe subsidy is more than twice as large as the number of poor householdsreceiving the subsidy3
The most important statistic however is the share of the subsidy givento the nonpoor This tells us how much poverty reduction is obtained (interms of the poverty gap as defined in appendix 81) for each lempira spenton the subsidy As computed in the last column of table 82 this share isabove 80 percent Part of the reason for the subsidyrsquos poor targeting per-formance is related to the level of the lifeline threshold which is set toohigh Some 835 percent of households with access to electricity consumeless than 300 kWh per month and hence qualify for the subsidy Further-more while the level of poverty is higher among households that consumeless than 100 kWh most of the subsidy is spent on households that con-sume between 100 and 300 kWh per month but these households are lesslikely to be poor
In part because the subsidy is not well targeted its impact on poverty issmall This is illustrated in table 83 which provides poverty measures withand without incorporating the value of the electricity subsidy in the overallconsumption aggregate The table provides three measures of poverty theheadcount index (the share of the population in poverty) the poverty gap(the distance separating the poor from the poverty line) and the squaredpoverty gap (appendix 81 provides a formal definition of these povertymeasures) The measures are provided for two alternative poverty lines cor-responding roughly to the extreme poor (L 400 per person per month) andthe poor (L 600 per person per month) Overall the changes in povertywhen subsidies are taken into account are small and these changes arelikely to be slightly overestimated because we do not take substitutioneffects due to the subsidy into account (if the subsidy were eliminatedelectricity prices would go up and households would substitute consump-tion toward other goods)
Although this is not reported here we carried out additional simulations toassess if the results are sensitive to the choice of the PRAF survey for theanalysis That is using a large set of variables common to both the PRAFand the nationally representative EHPHM survey and fitting a predictivemodel of electricity consumption in the PRAF survey we obtained a predic-tion for electricity consumption in the EHPHM survey and redid estimationsregarding targeting performance and the impact of the subsidy on poverty
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
285
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
277
8
Lifeline or Means-Testing ElectricUtility Subsidies in Honduras
Quentin Wodon Mohamed Ihsan Ajwadand Corinne Siaens
(c) The International Bank for Reconstruction and Development The World Bank
81 Introduction
Many countries around the world have implemented subsidies for utilityconsumption especially in the case of water and electricity Most subsi-dies take the form of a lifeline or increasing block tariff whereby house-holds that consume less pay less on a unit basis The idea is that house-holds with low consumption levels are likely to be poor and someintervention is warranted to enable them to meet their basic needs (the life-line) at an affordable cost Whether such subsidies are successful at help-ing the poor is not clear as illustrated by the experience of a number ofCentral American and Latin American countries
Goacutemez-Lobo and Contreras (2000) suggested that in Chile and Colom-bia errors of exclusion (poor households not benefiting from the subsidy)and inclusion (nonpoor households benefiting from the subsidy) for waterand electricity subsidies are large (for a review of this and other studiessee Estache Foster and Wodon 2002) In Colombia dwelling and neigh-borhood characteristics are used as proxies for income in a six-tier classi-fication of households Households classified as belonging to the lowerstrata get a percentage reduction in their water and energy bills that isfinanced by a surcharge for households from the upper strata Subsidiza-tion takes place primarily within each utility that is the subsidies arefinanced by higher tariffs on unsubsidized customers but the governmentalso provides public funds to utilities in areas with few high-income house-holds Because eligibility rules are not stringent the subsidy reaches alarge share of the poor that is the rate of errors of exclusion is low but thisleads to large errors of inclusion up to 80 percent of beneficiaries are non-poor households depending on the definition of who is poor
In Chile eligible households receive a subsidy of 20 to 85 percent of thewater bill for the first 15 cubic meters of monthly consumption The target-ing mechanism is based both on regional data on water consumption andtariffs and on household characteristics as measured by a national means-testing system While errors of inclusion for nonpoor households are lowerthan in Colombia the errors of exclusion are much higher at up to 80 per-cent In addition research by Clert and Wodon (2001) suggests that the
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
278
The research was funded through a grant from the Energy Sector Management Assistance Programunder the project Low Income Energy Assistance in Latin America
(c) The International Bank for Reconstruction and Development The World Bank
water subsidyrsquos overall performance in reducing poverty and inequality issubstantially weaker than that of other social programs targeted using thenational means-testing system These other programs include noncontribu-tory pensions for the elderly poor family allowances for poor families withchildren or pregnant women subsidized child care centers and housingsubsidies Given that their impact on poverty and inequality tends to belarger than that of utility subsidies this raises the question of whetherkeeping these subsidies is worthwhile or whether the funds used for thesubsidies should be invested in these other programs
In Guatemala following reform of the electricity sector prices for resi-dential customers increased after 1998 in some case by up to 85 percentover a three-year period To ensure affordability for poor customers thegovernment introduced a social tariff for those consuming less than 500kilowatt hours (kWh) per month In January 2001 the government reducedthe lifeline to 300 kWh per household per month This threshold which issimilar to the threshold in Honduras examined in this chapter is still highbecause an average household consumes only 102 kWh per month Anevaluation by Foster and Araujo (forthcoming) suggests that two-thirds ofbeneficiaries are not poor and because the nonpoor consume more elec-tricity 90 percent of the funds go to the nonpoor
Finally in El Salvador until recently the Electricity-Telephone Invest-ment Fund subsidized residential electricity consumption for householdsconsuming less than 200 kWh per month The subsidy was a flat 75 per-cent reduction on the bill paid by the fund to the electricity company everysix months An additional subsidy came from the payment of the consump-tion tax (13 percent) on that share of the householdrsquos consumption As inother countries this lifeline subsidy had high leakage rates to the nonpoorand it benefited mostly urban households The subsidy was recently termi-nated
In this chapter we assess the targeting performance of a similar subsidyfor electricity implemented in Honduras Honduras is the second largestcountry in Central America and one of the poorest in the Western Hemi-sphere (World Bank 2001) The country has 65 million inhabitants Aspart of its efforts to reduce poverty the government is funding two largesubsidies for basic infrastructure services The first subsidy is for electric-ity consumption which in 1998 cost the government L 259 million (orUS$175 million at an exchange rate at that time of L 1480 to the US
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
279
(c) The International Bank for Reconstruction and Development The World Bank
dollar) The second subsidy with a cost of L 114 million in 1990 is for bustransportation in the capital city of Tegucigalpa These subsidies are largein comparison with other social programs because only onemdashthe Hon-duras Social Investment Fundmdashhas a larger budget Here we focus on theelectricity subsidy which the state paid to the national electric utility onbehalf of beneficiaries
The subsidy is targeted through the lifeline principle however becausethe consumption threshold for eligibility is relatively high (300 kWh permonth) and because those with access to electricity tend to be less poorthan those without access the programrsquos overall performance is low interms of poverty reduction Targeting through means-testing rather than alifeline or at least a lower threshold for the lifeline could help improve theimpact of the subsidy and based on experience in other countries wouldnot necessarily imply high administrative costs
The rest of the paper is structured as follows Section 82 describes thesubsidy scheme and assesses its performance Section 83 illustrates howsimple techniques based on so-called receiver operating characteristics(ROC) curves can be used to show how much improvement might comefrom an alternative way of targeting the subsidy Section 84 concludes
82 Targeting Performance of Hondurasrsquos Electricity Subsidy
Electricity access rates in Honduras can be estimated using the PermanentMultiple Purpose Household Survey (Encuesta Permanente de Hogares dePropoacutesitos Muacuteltiples or EPHPM) This is a nationally representative laborforce survey implemented with support from the US Agency for Interna-tional Development and conducted by the General Directorate for Statisticsand Censuses The sample consists of 6423 households stratified into fourgeographic regions Tegucigalpa San Pedro Sula other urban and rural
As table 81 reveals Empresa Nacional de Energiacutea Eleacutectrica (ENEE)the public provider of electricity is by far the predominant provider ofelectricity for households A few households obtain their electricity fromcollectives or privately owned electricity generators Clearly access to elec-tricity varies a good deal by income group and location As expectedhigher-income households have higher access rates than lower-income
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
280
(c) The International Bank for Reconstruction and Development The World Bank
281
TAB
LE 8
1 A
cces
s to
Ele
ctri
city
by
Inco
me
Gro
up
Ho
nd
ura
s 1
999
(per
cen
t)
Inco
me
dec
ile
Loca
tio
n a
nd
so
urc
e1
23
45
67
89
10M
ean
Nat
ion
al
Non
e75
065
346
533
525
920
416
19
76
64
730
4
ENEE
245
347
532
657
739
787
821
899
919
933
688
Colle
ctiv
e0
00
10
30
50
00
70
70
30
40
50
4
Indi
vidu
ala
05
00
00
03
02
02
11
01
11
15
05
Urb
an
Non
e31
016
610
09
04
01
91
40
40
20
47
5
ENEE
691
834
901
907
960
981
986
996
997
994
925
Colle
ctiv
e0
00
00
00
30
00
00
00
00
10
20
1
Indi
vidu
ala
00
00
00
00
00
00
00
00
00
00
00
Ru
ral
Non
e80
276
759
450
850
046
840
733
823
325
848
8
ENEE
193
232
401
480
496
508
547
648
715
633
485
Colle
ctiv
e0
00
10
40
60
01
81
81
11
11
90
9
Indi
vidu
ala
06
00
00
06
04
06
29
03
41
90
19
a S
mal
l pri
vate
ly o
wn
ed g
ener
ato
rs
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
Mar
ch 1
999
EPH
PM s
urv
ey
(c) The International Bank for Reconstruction and Development The World Bank
households For instance 245 percent of all households in the 1st or poor-est decile have access to electricity provided by ENEE compared with 933percent of households in the 10th richest decile Also as expected house-holds in urban areas have higher access rates than households in ruralareas The fact that richer households have higher rates of access to elec-tricity than poorer households is one of the reasons why from the point ofview of poverty reduction and in comparison with other programs thatcould be implemented to fight poverty electricity subsidies may not be welltargeted
The design of the electricity subsidy in Honduras follows a basic self-targeting structure that is households self-select themselves for inclusionin the subsidy scheme by not consuming more than a certain amount Thisimplies that neither the utility nor the government needs to design a morecomplex targeting mechanism based on means-testing whereby they usehousehold characteristics as proxies to identify who is poor and who is notand thereby who is eligible to receive the subsidy and who is not As of July2002 the subsidy was provided to all households that purchased electricityfrom ENEE and consumed less than 300 kWh per month the electricityconsumption lifeline Specifically households consuming less than 300kWh paid the price of electricity at the 1994 rate plus an increase of 1631percent approved in 2000 Households that consumed more than 300 kWhper month paid the full price at the 1997 rate plus the same 1631 percentincrease The government paid the difference between the 1994 and 1997prices (L 238 million per month in mid-2002) directly to ENEE1
To assess the targeting performance of the subsidy that is what share ofthe subsidy goes to the poor we combine administrative information withhousehold survey data On the administrative data side we rely on infor-mation provided by ENEE This information includes (a) the percentage ofresidential clients by consumption level (b) the average amount of elec-tricity consumed by clients at each level (c) the total electricity bill with-out subsidy by level and (d) the total subsidy paid by level
On the household data side to estimate the proportion of poor householdsat the various consumption levels we use data from a survey implementedin 1999 by the Family Assistance Program (Programa de AsignacioacutenFamiliar or PRAF) with financing from the Inter-American DevelopmentBank and assistance from the International Food Policy Research InstituteThe survey was implemented in the bottom half of Hondurasrsquos municipali-
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
282
(c) The International Bank for Reconstruction and Development The World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
283
ties in terms of malnutrition as measured by the Ministry of Educationrsquosannual census of weight and height during the first year of primary schoolThe survey has modules on expenditures education health and the impactof Hurricane Mitch The advantage of this survey is that it includes dataon electricity consumption which are not available in the labor force sur-vey The disadvantage is that the survey is not nationally representativebut we assume that the results obtained here would also apply if the surveywere nationally representative2
Table 82 provides the results For example according to ENEErsquos admin-istrative records the share of households connected to its network withmonthly consumption below 20 kWh is 2031 percent (115723 house-holds) Of these we estimate from the PRAF survey that 4493 percent arepoor That is among all households that consume less than 20 kWh accord-ing to the PRAF survey about 45 percent have a level of per capita con-sumption that is below a reasonable poverty line for Honduras With aver-age consumption of 336 kWh per household per month the totalconsumption for this group is 388626 kWh Without the subsidy this groupwould have to pay a total bill of L 929256 but this bill is reduced toL 333282 when the subsidy of L 595973 is taken into account
Presenting statistics on errors of inclusion and exclusion is commonwhen assessing the targeting performance of a subsidy (see for exampleCornia and Stewart 1995) There are various ways to present these twotypes of errors The simplest way used in table 82 consists of computingthe share of households in poverty and receiving the subsidy as well as theshare of households not in poverty that are also receiving the subsidy Ingeneral as errors of inclusion increase errors of exclusion decrease andvice versa If all households receive the subsidy there are no errors ofexclusion but the errors of inclusion will likely be large because all non-poor households benefit from the subsidy
In table 82 among those households that are connected to ENEE6019 percent are nonpoor and receive the subsidy By contrast only 168percent are poor and do not receive the subsidy (nationally the share ofthe population in poverty not receiving the subsidy is larger becausemany poor households are not connected to the grid) In addition 2328percent of households that are poor receive the subsidy and 1485 percentof households that are nonpoor do not receive the subsidy From theseresults we can estimate that the ratio of poor versus nonpoor beneficiaries
(c) The International Bank for Reconstruction and Development The World Bank
284
TAB
LE 8
2 T
arg
etin
g P
erfo
rman
ce f
or
the
Elec
tric
ity
Sub
sid
y H
on
du
ras
Mid
-200
2
Tota
l bill
Sh
are
of
Co
nsu
mp
tio
nSh
are
of
clie
nts
Er
ror
of
A
vera
ge
wit
ho
ut
Tota
lsu
bsi
dy
leve
l Sh
are
of
in p
ove
rty
incl
usi
on
Er
ror
of
con
sum
pti
on
sub
sid
y b
y su
bsi
dy
by
spen
t o
n
(kW
h p
er)
clie
nts
b
ased
on
199
9 (1
)[1
-(2)
] ex
clu
sio
n
(kW
h p
er
cate
go
ryca
teg
ory
no
np
oo
r m
on
th)
()
(1)
dat
a (
) (2
)(
)(1
)(2
) (
)m
on
th)
(L t
ho
usa
nd
s)(L
th
ou
san
ds)
ho
use
ho
lds
()
0ndash20
203
144
93
111
8n
a3
3692
959
61
38
20ndash1
0022
69
356
614
60
na
586
75
096
271
77
34
100ndash
150
126
316
82
105
0n
a12
509
738
73
762
131
4
150ndash
200
111
610
98
994
na
175
3510
314
514
919
24
200ndash
250
925
156
47
81n
a22
454
116
185
746
203
5
250ndash
300
743
170
96
16n
a27
577
118
965
851
203
6
300+
165
310
15
na
168
mdash
mdashmdash
mdash
Tota
l10
024
96
601
91
6810
858
472
4123
820
818
1
mdash N
ot
avai
lab
le
na
No
t ap
plic
able
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
EN
EE (
2002
) an
d t
he
1999
PR
AF
surv
ey d
ata
(c) The International Bank for Reconstruction and Development The World Bank
is 039 (23286019) and so the number of nonpoor households receivingthe subsidy is more than twice as large as the number of poor householdsreceiving the subsidy3
The most important statistic however is the share of the subsidy givento the nonpoor This tells us how much poverty reduction is obtained (interms of the poverty gap as defined in appendix 81) for each lempira spenton the subsidy As computed in the last column of table 82 this share isabove 80 percent Part of the reason for the subsidyrsquos poor targeting per-formance is related to the level of the lifeline threshold which is set toohigh Some 835 percent of households with access to electricity consumeless than 300 kWh per month and hence qualify for the subsidy Further-more while the level of poverty is higher among households that consumeless than 100 kWh most of the subsidy is spent on households that con-sume between 100 and 300 kWh per month but these households are lesslikely to be poor
In part because the subsidy is not well targeted its impact on poverty issmall This is illustrated in table 83 which provides poverty measures withand without incorporating the value of the electricity subsidy in the overallconsumption aggregate The table provides three measures of poverty theheadcount index (the share of the population in poverty) the poverty gap(the distance separating the poor from the poverty line) and the squaredpoverty gap (appendix 81 provides a formal definition of these povertymeasures) The measures are provided for two alternative poverty lines cor-responding roughly to the extreme poor (L 400 per person per month) andthe poor (L 600 per person per month) Overall the changes in povertywhen subsidies are taken into account are small and these changes arelikely to be slightly overestimated because we do not take substitutioneffects due to the subsidy into account (if the subsidy were eliminatedelectricity prices would go up and households would substitute consump-tion toward other goods)
Although this is not reported here we carried out additional simulations toassess if the results are sensitive to the choice of the PRAF survey for theanalysis That is using a large set of variables common to both the PRAFand the nationally representative EHPHM survey and fitting a predictivemodel of electricity consumption in the PRAF survey we obtained a predic-tion for electricity consumption in the EHPHM survey and redid estimationsregarding targeting performance and the impact of the subsidy on poverty
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
285
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
81 Introduction
Many countries around the world have implemented subsidies for utilityconsumption especially in the case of water and electricity Most subsi-dies take the form of a lifeline or increasing block tariff whereby house-holds that consume less pay less on a unit basis The idea is that house-holds with low consumption levels are likely to be poor and someintervention is warranted to enable them to meet their basic needs (the life-line) at an affordable cost Whether such subsidies are successful at help-ing the poor is not clear as illustrated by the experience of a number ofCentral American and Latin American countries
Goacutemez-Lobo and Contreras (2000) suggested that in Chile and Colom-bia errors of exclusion (poor households not benefiting from the subsidy)and inclusion (nonpoor households benefiting from the subsidy) for waterand electricity subsidies are large (for a review of this and other studiessee Estache Foster and Wodon 2002) In Colombia dwelling and neigh-borhood characteristics are used as proxies for income in a six-tier classi-fication of households Households classified as belonging to the lowerstrata get a percentage reduction in their water and energy bills that isfinanced by a surcharge for households from the upper strata Subsidiza-tion takes place primarily within each utility that is the subsidies arefinanced by higher tariffs on unsubsidized customers but the governmentalso provides public funds to utilities in areas with few high-income house-holds Because eligibility rules are not stringent the subsidy reaches alarge share of the poor that is the rate of errors of exclusion is low but thisleads to large errors of inclusion up to 80 percent of beneficiaries are non-poor households depending on the definition of who is poor
In Chile eligible households receive a subsidy of 20 to 85 percent of thewater bill for the first 15 cubic meters of monthly consumption The target-ing mechanism is based both on regional data on water consumption andtariffs and on household characteristics as measured by a national means-testing system While errors of inclusion for nonpoor households are lowerthan in Colombia the errors of exclusion are much higher at up to 80 per-cent In addition research by Clert and Wodon (2001) suggests that the
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
278
The research was funded through a grant from the Energy Sector Management Assistance Programunder the project Low Income Energy Assistance in Latin America
(c) The International Bank for Reconstruction and Development The World Bank
water subsidyrsquos overall performance in reducing poverty and inequality issubstantially weaker than that of other social programs targeted using thenational means-testing system These other programs include noncontribu-tory pensions for the elderly poor family allowances for poor families withchildren or pregnant women subsidized child care centers and housingsubsidies Given that their impact on poverty and inequality tends to belarger than that of utility subsidies this raises the question of whetherkeeping these subsidies is worthwhile or whether the funds used for thesubsidies should be invested in these other programs
In Guatemala following reform of the electricity sector prices for resi-dential customers increased after 1998 in some case by up to 85 percentover a three-year period To ensure affordability for poor customers thegovernment introduced a social tariff for those consuming less than 500kilowatt hours (kWh) per month In January 2001 the government reducedthe lifeline to 300 kWh per household per month This threshold which issimilar to the threshold in Honduras examined in this chapter is still highbecause an average household consumes only 102 kWh per month Anevaluation by Foster and Araujo (forthcoming) suggests that two-thirds ofbeneficiaries are not poor and because the nonpoor consume more elec-tricity 90 percent of the funds go to the nonpoor
Finally in El Salvador until recently the Electricity-Telephone Invest-ment Fund subsidized residential electricity consumption for householdsconsuming less than 200 kWh per month The subsidy was a flat 75 per-cent reduction on the bill paid by the fund to the electricity company everysix months An additional subsidy came from the payment of the consump-tion tax (13 percent) on that share of the householdrsquos consumption As inother countries this lifeline subsidy had high leakage rates to the nonpoorand it benefited mostly urban households The subsidy was recently termi-nated
In this chapter we assess the targeting performance of a similar subsidyfor electricity implemented in Honduras Honduras is the second largestcountry in Central America and one of the poorest in the Western Hemi-sphere (World Bank 2001) The country has 65 million inhabitants Aspart of its efforts to reduce poverty the government is funding two largesubsidies for basic infrastructure services The first subsidy is for electric-ity consumption which in 1998 cost the government L 259 million (orUS$175 million at an exchange rate at that time of L 1480 to the US
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
279
(c) The International Bank for Reconstruction and Development The World Bank
dollar) The second subsidy with a cost of L 114 million in 1990 is for bustransportation in the capital city of Tegucigalpa These subsidies are largein comparison with other social programs because only onemdashthe Hon-duras Social Investment Fundmdashhas a larger budget Here we focus on theelectricity subsidy which the state paid to the national electric utility onbehalf of beneficiaries
The subsidy is targeted through the lifeline principle however becausethe consumption threshold for eligibility is relatively high (300 kWh permonth) and because those with access to electricity tend to be less poorthan those without access the programrsquos overall performance is low interms of poverty reduction Targeting through means-testing rather than alifeline or at least a lower threshold for the lifeline could help improve theimpact of the subsidy and based on experience in other countries wouldnot necessarily imply high administrative costs
The rest of the paper is structured as follows Section 82 describes thesubsidy scheme and assesses its performance Section 83 illustrates howsimple techniques based on so-called receiver operating characteristics(ROC) curves can be used to show how much improvement might comefrom an alternative way of targeting the subsidy Section 84 concludes
82 Targeting Performance of Hondurasrsquos Electricity Subsidy
Electricity access rates in Honduras can be estimated using the PermanentMultiple Purpose Household Survey (Encuesta Permanente de Hogares dePropoacutesitos Muacuteltiples or EPHPM) This is a nationally representative laborforce survey implemented with support from the US Agency for Interna-tional Development and conducted by the General Directorate for Statisticsand Censuses The sample consists of 6423 households stratified into fourgeographic regions Tegucigalpa San Pedro Sula other urban and rural
As table 81 reveals Empresa Nacional de Energiacutea Eleacutectrica (ENEE)the public provider of electricity is by far the predominant provider ofelectricity for households A few households obtain their electricity fromcollectives or privately owned electricity generators Clearly access to elec-tricity varies a good deal by income group and location As expectedhigher-income households have higher access rates than lower-income
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
280
(c) The International Bank for Reconstruction and Development The World Bank
281
TAB
LE 8
1 A
cces
s to
Ele
ctri
city
by
Inco
me
Gro
up
Ho
nd
ura
s 1
999
(per
cen
t)
Inco
me
dec
ile
Loca
tio
n a
nd
so
urc
e1
23
45
67
89
10M
ean
Nat
ion
al
Non
e75
065
346
533
525
920
416
19
76
64
730
4
ENEE
245
347
532
657
739
787
821
899
919
933
688
Colle
ctiv
e0
00
10
30
50
00
70
70
30
40
50
4
Indi
vidu
ala
05
00
00
03
02
02
11
01
11
15
05
Urb
an
Non
e31
016
610
09
04
01
91
40
40
20
47
5
ENEE
691
834
901
907
960
981
986
996
997
994
925
Colle
ctiv
e0
00
00
00
30
00
00
00
00
10
20
1
Indi
vidu
ala
00
00
00
00
00
00
00
00
00
00
00
Ru
ral
Non
e80
276
759
450
850
046
840
733
823
325
848
8
ENEE
193
232
401
480
496
508
547
648
715
633
485
Colle
ctiv
e0
00
10
40
60
01
81
81
11
11
90
9
Indi
vidu
ala
06
00
00
06
04
06
29
03
41
90
19
a S
mal
l pri
vate
ly o
wn
ed g
ener
ato
rs
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
Mar
ch 1
999
EPH
PM s
urv
ey
(c) The International Bank for Reconstruction and Development The World Bank
households For instance 245 percent of all households in the 1st or poor-est decile have access to electricity provided by ENEE compared with 933percent of households in the 10th richest decile Also as expected house-holds in urban areas have higher access rates than households in ruralareas The fact that richer households have higher rates of access to elec-tricity than poorer households is one of the reasons why from the point ofview of poverty reduction and in comparison with other programs thatcould be implemented to fight poverty electricity subsidies may not be welltargeted
The design of the electricity subsidy in Honduras follows a basic self-targeting structure that is households self-select themselves for inclusionin the subsidy scheme by not consuming more than a certain amount Thisimplies that neither the utility nor the government needs to design a morecomplex targeting mechanism based on means-testing whereby they usehousehold characteristics as proxies to identify who is poor and who is notand thereby who is eligible to receive the subsidy and who is not As of July2002 the subsidy was provided to all households that purchased electricityfrom ENEE and consumed less than 300 kWh per month the electricityconsumption lifeline Specifically households consuming less than 300kWh paid the price of electricity at the 1994 rate plus an increase of 1631percent approved in 2000 Households that consumed more than 300 kWhper month paid the full price at the 1997 rate plus the same 1631 percentincrease The government paid the difference between the 1994 and 1997prices (L 238 million per month in mid-2002) directly to ENEE1
To assess the targeting performance of the subsidy that is what share ofthe subsidy goes to the poor we combine administrative information withhousehold survey data On the administrative data side we rely on infor-mation provided by ENEE This information includes (a) the percentage ofresidential clients by consumption level (b) the average amount of elec-tricity consumed by clients at each level (c) the total electricity bill with-out subsidy by level and (d) the total subsidy paid by level
On the household data side to estimate the proportion of poor householdsat the various consumption levels we use data from a survey implementedin 1999 by the Family Assistance Program (Programa de AsignacioacutenFamiliar or PRAF) with financing from the Inter-American DevelopmentBank and assistance from the International Food Policy Research InstituteThe survey was implemented in the bottom half of Hondurasrsquos municipali-
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
282
(c) The International Bank for Reconstruction and Development The World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
283
ties in terms of malnutrition as measured by the Ministry of Educationrsquosannual census of weight and height during the first year of primary schoolThe survey has modules on expenditures education health and the impactof Hurricane Mitch The advantage of this survey is that it includes dataon electricity consumption which are not available in the labor force sur-vey The disadvantage is that the survey is not nationally representativebut we assume that the results obtained here would also apply if the surveywere nationally representative2
Table 82 provides the results For example according to ENEErsquos admin-istrative records the share of households connected to its network withmonthly consumption below 20 kWh is 2031 percent (115723 house-holds) Of these we estimate from the PRAF survey that 4493 percent arepoor That is among all households that consume less than 20 kWh accord-ing to the PRAF survey about 45 percent have a level of per capita con-sumption that is below a reasonable poverty line for Honduras With aver-age consumption of 336 kWh per household per month the totalconsumption for this group is 388626 kWh Without the subsidy this groupwould have to pay a total bill of L 929256 but this bill is reduced toL 333282 when the subsidy of L 595973 is taken into account
Presenting statistics on errors of inclusion and exclusion is commonwhen assessing the targeting performance of a subsidy (see for exampleCornia and Stewart 1995) There are various ways to present these twotypes of errors The simplest way used in table 82 consists of computingthe share of households in poverty and receiving the subsidy as well as theshare of households not in poverty that are also receiving the subsidy Ingeneral as errors of inclusion increase errors of exclusion decrease andvice versa If all households receive the subsidy there are no errors ofexclusion but the errors of inclusion will likely be large because all non-poor households benefit from the subsidy
In table 82 among those households that are connected to ENEE6019 percent are nonpoor and receive the subsidy By contrast only 168percent are poor and do not receive the subsidy (nationally the share ofthe population in poverty not receiving the subsidy is larger becausemany poor households are not connected to the grid) In addition 2328percent of households that are poor receive the subsidy and 1485 percentof households that are nonpoor do not receive the subsidy From theseresults we can estimate that the ratio of poor versus nonpoor beneficiaries
(c) The International Bank for Reconstruction and Development The World Bank
284
TAB
LE 8
2 T
arg
etin
g P
erfo
rman
ce f
or
the
Elec
tric
ity
Sub
sid
y H
on
du
ras
Mid
-200
2
Tota
l bill
Sh
are
of
Co
nsu
mp
tio
nSh
are
of
clie
nts
Er
ror
of
A
vera
ge
wit
ho
ut
Tota
lsu
bsi
dy
leve
l Sh
are
of
in p
ove
rty
incl
usi
on
Er
ror
of
con
sum
pti
on
sub
sid
y b
y su
bsi
dy
by
spen
t o
n
(kW
h p
er)
clie
nts
b
ased
on
199
9 (1
)[1
-(2)
] ex
clu
sio
n
(kW
h p
er
cate
go
ryca
teg
ory
no
np
oo
r m
on
th)
()
(1)
dat
a (
) (2
)(
)(1
)(2
) (
)m
on
th)
(L t
ho
usa
nd
s)(L
th
ou
san
ds)
ho
use
ho
lds
()
0ndash20
203
144
93
111
8n
a3
3692
959
61
38
20ndash1
0022
69
356
614
60
na
586
75
096
271
77
34
100ndash
150
126
316
82
105
0n
a12
509
738
73
762
131
4
150ndash
200
111
610
98
994
na
175
3510
314
514
919
24
200ndash
250
925
156
47
81n
a22
454
116
185
746
203
5
250ndash
300
743
170
96
16n
a27
577
118
965
851
203
6
300+
165
310
15
na
168
mdash
mdashmdash
mdash
Tota
l10
024
96
601
91
6810
858
472
4123
820
818
1
mdash N
ot
avai
lab
le
na
No
t ap
plic
able
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
EN
EE (
2002
) an
d t
he
1999
PR
AF
surv
ey d
ata
(c) The International Bank for Reconstruction and Development The World Bank
is 039 (23286019) and so the number of nonpoor households receivingthe subsidy is more than twice as large as the number of poor householdsreceiving the subsidy3
The most important statistic however is the share of the subsidy givento the nonpoor This tells us how much poverty reduction is obtained (interms of the poverty gap as defined in appendix 81) for each lempira spenton the subsidy As computed in the last column of table 82 this share isabove 80 percent Part of the reason for the subsidyrsquos poor targeting per-formance is related to the level of the lifeline threshold which is set toohigh Some 835 percent of households with access to electricity consumeless than 300 kWh per month and hence qualify for the subsidy Further-more while the level of poverty is higher among households that consumeless than 100 kWh most of the subsidy is spent on households that con-sume between 100 and 300 kWh per month but these households are lesslikely to be poor
In part because the subsidy is not well targeted its impact on poverty issmall This is illustrated in table 83 which provides poverty measures withand without incorporating the value of the electricity subsidy in the overallconsumption aggregate The table provides three measures of poverty theheadcount index (the share of the population in poverty) the poverty gap(the distance separating the poor from the poverty line) and the squaredpoverty gap (appendix 81 provides a formal definition of these povertymeasures) The measures are provided for two alternative poverty lines cor-responding roughly to the extreme poor (L 400 per person per month) andthe poor (L 600 per person per month) Overall the changes in povertywhen subsidies are taken into account are small and these changes arelikely to be slightly overestimated because we do not take substitutioneffects due to the subsidy into account (if the subsidy were eliminatedelectricity prices would go up and households would substitute consump-tion toward other goods)
Although this is not reported here we carried out additional simulations toassess if the results are sensitive to the choice of the PRAF survey for theanalysis That is using a large set of variables common to both the PRAFand the nationally representative EHPHM survey and fitting a predictivemodel of electricity consumption in the PRAF survey we obtained a predic-tion for electricity consumption in the EHPHM survey and redid estimationsregarding targeting performance and the impact of the subsidy on poverty
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
285
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
water subsidyrsquos overall performance in reducing poverty and inequality issubstantially weaker than that of other social programs targeted using thenational means-testing system These other programs include noncontribu-tory pensions for the elderly poor family allowances for poor families withchildren or pregnant women subsidized child care centers and housingsubsidies Given that their impact on poverty and inequality tends to belarger than that of utility subsidies this raises the question of whetherkeeping these subsidies is worthwhile or whether the funds used for thesubsidies should be invested in these other programs
In Guatemala following reform of the electricity sector prices for resi-dential customers increased after 1998 in some case by up to 85 percentover a three-year period To ensure affordability for poor customers thegovernment introduced a social tariff for those consuming less than 500kilowatt hours (kWh) per month In January 2001 the government reducedthe lifeline to 300 kWh per household per month This threshold which issimilar to the threshold in Honduras examined in this chapter is still highbecause an average household consumes only 102 kWh per month Anevaluation by Foster and Araujo (forthcoming) suggests that two-thirds ofbeneficiaries are not poor and because the nonpoor consume more elec-tricity 90 percent of the funds go to the nonpoor
Finally in El Salvador until recently the Electricity-Telephone Invest-ment Fund subsidized residential electricity consumption for householdsconsuming less than 200 kWh per month The subsidy was a flat 75 per-cent reduction on the bill paid by the fund to the electricity company everysix months An additional subsidy came from the payment of the consump-tion tax (13 percent) on that share of the householdrsquos consumption As inother countries this lifeline subsidy had high leakage rates to the nonpoorand it benefited mostly urban households The subsidy was recently termi-nated
In this chapter we assess the targeting performance of a similar subsidyfor electricity implemented in Honduras Honduras is the second largestcountry in Central America and one of the poorest in the Western Hemi-sphere (World Bank 2001) The country has 65 million inhabitants Aspart of its efforts to reduce poverty the government is funding two largesubsidies for basic infrastructure services The first subsidy is for electric-ity consumption which in 1998 cost the government L 259 million (orUS$175 million at an exchange rate at that time of L 1480 to the US
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
279
(c) The International Bank for Reconstruction and Development The World Bank
dollar) The second subsidy with a cost of L 114 million in 1990 is for bustransportation in the capital city of Tegucigalpa These subsidies are largein comparison with other social programs because only onemdashthe Hon-duras Social Investment Fundmdashhas a larger budget Here we focus on theelectricity subsidy which the state paid to the national electric utility onbehalf of beneficiaries
The subsidy is targeted through the lifeline principle however becausethe consumption threshold for eligibility is relatively high (300 kWh permonth) and because those with access to electricity tend to be less poorthan those without access the programrsquos overall performance is low interms of poverty reduction Targeting through means-testing rather than alifeline or at least a lower threshold for the lifeline could help improve theimpact of the subsidy and based on experience in other countries wouldnot necessarily imply high administrative costs
The rest of the paper is structured as follows Section 82 describes thesubsidy scheme and assesses its performance Section 83 illustrates howsimple techniques based on so-called receiver operating characteristics(ROC) curves can be used to show how much improvement might comefrom an alternative way of targeting the subsidy Section 84 concludes
82 Targeting Performance of Hondurasrsquos Electricity Subsidy
Electricity access rates in Honduras can be estimated using the PermanentMultiple Purpose Household Survey (Encuesta Permanente de Hogares dePropoacutesitos Muacuteltiples or EPHPM) This is a nationally representative laborforce survey implemented with support from the US Agency for Interna-tional Development and conducted by the General Directorate for Statisticsand Censuses The sample consists of 6423 households stratified into fourgeographic regions Tegucigalpa San Pedro Sula other urban and rural
As table 81 reveals Empresa Nacional de Energiacutea Eleacutectrica (ENEE)the public provider of electricity is by far the predominant provider ofelectricity for households A few households obtain their electricity fromcollectives or privately owned electricity generators Clearly access to elec-tricity varies a good deal by income group and location As expectedhigher-income households have higher access rates than lower-income
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
280
(c) The International Bank for Reconstruction and Development The World Bank
281
TAB
LE 8
1 A
cces
s to
Ele
ctri
city
by
Inco
me
Gro
up
Ho
nd
ura
s 1
999
(per
cen
t)
Inco
me
dec
ile
Loca
tio
n a
nd
so
urc
e1
23
45
67
89
10M
ean
Nat
ion
al
Non
e75
065
346
533
525
920
416
19
76
64
730
4
ENEE
245
347
532
657
739
787
821
899
919
933
688
Colle
ctiv
e0
00
10
30
50
00
70
70
30
40
50
4
Indi
vidu
ala
05
00
00
03
02
02
11
01
11
15
05
Urb
an
Non
e31
016
610
09
04
01
91
40
40
20
47
5
ENEE
691
834
901
907
960
981
986
996
997
994
925
Colle
ctiv
e0
00
00
00
30
00
00
00
00
10
20
1
Indi
vidu
ala
00
00
00
00
00
00
00
00
00
00
00
Ru
ral
Non
e80
276
759
450
850
046
840
733
823
325
848
8
ENEE
193
232
401
480
496
508
547
648
715
633
485
Colle
ctiv
e0
00
10
40
60
01
81
81
11
11
90
9
Indi
vidu
ala
06
00
00
06
04
06
29
03
41
90
19
a S
mal
l pri
vate
ly o
wn
ed g
ener
ato
rs
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
Mar
ch 1
999
EPH
PM s
urv
ey
(c) The International Bank for Reconstruction and Development The World Bank
households For instance 245 percent of all households in the 1st or poor-est decile have access to electricity provided by ENEE compared with 933percent of households in the 10th richest decile Also as expected house-holds in urban areas have higher access rates than households in ruralareas The fact that richer households have higher rates of access to elec-tricity than poorer households is one of the reasons why from the point ofview of poverty reduction and in comparison with other programs thatcould be implemented to fight poverty electricity subsidies may not be welltargeted
The design of the electricity subsidy in Honduras follows a basic self-targeting structure that is households self-select themselves for inclusionin the subsidy scheme by not consuming more than a certain amount Thisimplies that neither the utility nor the government needs to design a morecomplex targeting mechanism based on means-testing whereby they usehousehold characteristics as proxies to identify who is poor and who is notand thereby who is eligible to receive the subsidy and who is not As of July2002 the subsidy was provided to all households that purchased electricityfrom ENEE and consumed less than 300 kWh per month the electricityconsumption lifeline Specifically households consuming less than 300kWh paid the price of electricity at the 1994 rate plus an increase of 1631percent approved in 2000 Households that consumed more than 300 kWhper month paid the full price at the 1997 rate plus the same 1631 percentincrease The government paid the difference between the 1994 and 1997prices (L 238 million per month in mid-2002) directly to ENEE1
To assess the targeting performance of the subsidy that is what share ofthe subsidy goes to the poor we combine administrative information withhousehold survey data On the administrative data side we rely on infor-mation provided by ENEE This information includes (a) the percentage ofresidential clients by consumption level (b) the average amount of elec-tricity consumed by clients at each level (c) the total electricity bill with-out subsidy by level and (d) the total subsidy paid by level
On the household data side to estimate the proportion of poor householdsat the various consumption levels we use data from a survey implementedin 1999 by the Family Assistance Program (Programa de AsignacioacutenFamiliar or PRAF) with financing from the Inter-American DevelopmentBank and assistance from the International Food Policy Research InstituteThe survey was implemented in the bottom half of Hondurasrsquos municipali-
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
282
(c) The International Bank for Reconstruction and Development The World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
283
ties in terms of malnutrition as measured by the Ministry of Educationrsquosannual census of weight and height during the first year of primary schoolThe survey has modules on expenditures education health and the impactof Hurricane Mitch The advantage of this survey is that it includes dataon electricity consumption which are not available in the labor force sur-vey The disadvantage is that the survey is not nationally representativebut we assume that the results obtained here would also apply if the surveywere nationally representative2
Table 82 provides the results For example according to ENEErsquos admin-istrative records the share of households connected to its network withmonthly consumption below 20 kWh is 2031 percent (115723 house-holds) Of these we estimate from the PRAF survey that 4493 percent arepoor That is among all households that consume less than 20 kWh accord-ing to the PRAF survey about 45 percent have a level of per capita con-sumption that is below a reasonable poverty line for Honduras With aver-age consumption of 336 kWh per household per month the totalconsumption for this group is 388626 kWh Without the subsidy this groupwould have to pay a total bill of L 929256 but this bill is reduced toL 333282 when the subsidy of L 595973 is taken into account
Presenting statistics on errors of inclusion and exclusion is commonwhen assessing the targeting performance of a subsidy (see for exampleCornia and Stewart 1995) There are various ways to present these twotypes of errors The simplest way used in table 82 consists of computingthe share of households in poverty and receiving the subsidy as well as theshare of households not in poverty that are also receiving the subsidy Ingeneral as errors of inclusion increase errors of exclusion decrease andvice versa If all households receive the subsidy there are no errors ofexclusion but the errors of inclusion will likely be large because all non-poor households benefit from the subsidy
In table 82 among those households that are connected to ENEE6019 percent are nonpoor and receive the subsidy By contrast only 168percent are poor and do not receive the subsidy (nationally the share ofthe population in poverty not receiving the subsidy is larger becausemany poor households are not connected to the grid) In addition 2328percent of households that are poor receive the subsidy and 1485 percentof households that are nonpoor do not receive the subsidy From theseresults we can estimate that the ratio of poor versus nonpoor beneficiaries
(c) The International Bank for Reconstruction and Development The World Bank
284
TAB
LE 8
2 T
arg
etin
g P
erfo
rman
ce f
or
the
Elec
tric
ity
Sub
sid
y H
on
du
ras
Mid
-200
2
Tota
l bill
Sh
are
of
Co
nsu
mp
tio
nSh
are
of
clie
nts
Er
ror
of
A
vera
ge
wit
ho
ut
Tota
lsu
bsi
dy
leve
l Sh
are
of
in p
ove
rty
incl
usi
on
Er
ror
of
con
sum
pti
on
sub
sid
y b
y su
bsi
dy
by
spen
t o
n
(kW
h p
er)
clie
nts
b
ased
on
199
9 (1
)[1
-(2)
] ex
clu
sio
n
(kW
h p
er
cate
go
ryca
teg
ory
no
np
oo
r m
on
th)
()
(1)
dat
a (
) (2
)(
)(1
)(2
) (
)m
on
th)
(L t
ho
usa
nd
s)(L
th
ou
san
ds)
ho
use
ho
lds
()
0ndash20
203
144
93
111
8n
a3
3692
959
61
38
20ndash1
0022
69
356
614
60
na
586
75
096
271
77
34
100ndash
150
126
316
82
105
0n
a12
509
738
73
762
131
4
150ndash
200
111
610
98
994
na
175
3510
314
514
919
24
200ndash
250
925
156
47
81n
a22
454
116
185
746
203
5
250ndash
300
743
170
96
16n
a27
577
118
965
851
203
6
300+
165
310
15
na
168
mdash
mdashmdash
mdash
Tota
l10
024
96
601
91
6810
858
472
4123
820
818
1
mdash N
ot
avai
lab
le
na
No
t ap
plic
able
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
EN
EE (
2002
) an
d t
he
1999
PR
AF
surv
ey d
ata
(c) The International Bank for Reconstruction and Development The World Bank
is 039 (23286019) and so the number of nonpoor households receivingthe subsidy is more than twice as large as the number of poor householdsreceiving the subsidy3
The most important statistic however is the share of the subsidy givento the nonpoor This tells us how much poverty reduction is obtained (interms of the poverty gap as defined in appendix 81) for each lempira spenton the subsidy As computed in the last column of table 82 this share isabove 80 percent Part of the reason for the subsidyrsquos poor targeting per-formance is related to the level of the lifeline threshold which is set toohigh Some 835 percent of households with access to electricity consumeless than 300 kWh per month and hence qualify for the subsidy Further-more while the level of poverty is higher among households that consumeless than 100 kWh most of the subsidy is spent on households that con-sume between 100 and 300 kWh per month but these households are lesslikely to be poor
In part because the subsidy is not well targeted its impact on poverty issmall This is illustrated in table 83 which provides poverty measures withand without incorporating the value of the electricity subsidy in the overallconsumption aggregate The table provides three measures of poverty theheadcount index (the share of the population in poverty) the poverty gap(the distance separating the poor from the poverty line) and the squaredpoverty gap (appendix 81 provides a formal definition of these povertymeasures) The measures are provided for two alternative poverty lines cor-responding roughly to the extreme poor (L 400 per person per month) andthe poor (L 600 per person per month) Overall the changes in povertywhen subsidies are taken into account are small and these changes arelikely to be slightly overestimated because we do not take substitutioneffects due to the subsidy into account (if the subsidy were eliminatedelectricity prices would go up and households would substitute consump-tion toward other goods)
Although this is not reported here we carried out additional simulations toassess if the results are sensitive to the choice of the PRAF survey for theanalysis That is using a large set of variables common to both the PRAFand the nationally representative EHPHM survey and fitting a predictivemodel of electricity consumption in the PRAF survey we obtained a predic-tion for electricity consumption in the EHPHM survey and redid estimationsregarding targeting performance and the impact of the subsidy on poverty
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
285
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
dollar) The second subsidy with a cost of L 114 million in 1990 is for bustransportation in the capital city of Tegucigalpa These subsidies are largein comparison with other social programs because only onemdashthe Hon-duras Social Investment Fundmdashhas a larger budget Here we focus on theelectricity subsidy which the state paid to the national electric utility onbehalf of beneficiaries
The subsidy is targeted through the lifeline principle however becausethe consumption threshold for eligibility is relatively high (300 kWh permonth) and because those with access to electricity tend to be less poorthan those without access the programrsquos overall performance is low interms of poverty reduction Targeting through means-testing rather than alifeline or at least a lower threshold for the lifeline could help improve theimpact of the subsidy and based on experience in other countries wouldnot necessarily imply high administrative costs
The rest of the paper is structured as follows Section 82 describes thesubsidy scheme and assesses its performance Section 83 illustrates howsimple techniques based on so-called receiver operating characteristics(ROC) curves can be used to show how much improvement might comefrom an alternative way of targeting the subsidy Section 84 concludes
82 Targeting Performance of Hondurasrsquos Electricity Subsidy
Electricity access rates in Honduras can be estimated using the PermanentMultiple Purpose Household Survey (Encuesta Permanente de Hogares dePropoacutesitos Muacuteltiples or EPHPM) This is a nationally representative laborforce survey implemented with support from the US Agency for Interna-tional Development and conducted by the General Directorate for Statisticsand Censuses The sample consists of 6423 households stratified into fourgeographic regions Tegucigalpa San Pedro Sula other urban and rural
As table 81 reveals Empresa Nacional de Energiacutea Eleacutectrica (ENEE)the public provider of electricity is by far the predominant provider ofelectricity for households A few households obtain their electricity fromcollectives or privately owned electricity generators Clearly access to elec-tricity varies a good deal by income group and location As expectedhigher-income households have higher access rates than lower-income
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
280
(c) The International Bank for Reconstruction and Development The World Bank
281
TAB
LE 8
1 A
cces
s to
Ele
ctri
city
by
Inco
me
Gro
up
Ho
nd
ura
s 1
999
(per
cen
t)
Inco
me
dec
ile
Loca
tio
n a
nd
so
urc
e1
23
45
67
89
10M
ean
Nat
ion
al
Non
e75
065
346
533
525
920
416
19
76
64
730
4
ENEE
245
347
532
657
739
787
821
899
919
933
688
Colle
ctiv
e0
00
10
30
50
00
70
70
30
40
50
4
Indi
vidu
ala
05
00
00
03
02
02
11
01
11
15
05
Urb
an
Non
e31
016
610
09
04
01
91
40
40
20
47
5
ENEE
691
834
901
907
960
981
986
996
997
994
925
Colle
ctiv
e0
00
00
00
30
00
00
00
00
10
20
1
Indi
vidu
ala
00
00
00
00
00
00
00
00
00
00
00
Ru
ral
Non
e80
276
759
450
850
046
840
733
823
325
848
8
ENEE
193
232
401
480
496
508
547
648
715
633
485
Colle
ctiv
e0
00
10
40
60
01
81
81
11
11
90
9
Indi
vidu
ala
06
00
00
06
04
06
29
03
41
90
19
a S
mal
l pri
vate
ly o
wn
ed g
ener
ato
rs
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
Mar
ch 1
999
EPH
PM s
urv
ey
(c) The International Bank for Reconstruction and Development The World Bank
households For instance 245 percent of all households in the 1st or poor-est decile have access to electricity provided by ENEE compared with 933percent of households in the 10th richest decile Also as expected house-holds in urban areas have higher access rates than households in ruralareas The fact that richer households have higher rates of access to elec-tricity than poorer households is one of the reasons why from the point ofview of poverty reduction and in comparison with other programs thatcould be implemented to fight poverty electricity subsidies may not be welltargeted
The design of the electricity subsidy in Honduras follows a basic self-targeting structure that is households self-select themselves for inclusionin the subsidy scheme by not consuming more than a certain amount Thisimplies that neither the utility nor the government needs to design a morecomplex targeting mechanism based on means-testing whereby they usehousehold characteristics as proxies to identify who is poor and who is notand thereby who is eligible to receive the subsidy and who is not As of July2002 the subsidy was provided to all households that purchased electricityfrom ENEE and consumed less than 300 kWh per month the electricityconsumption lifeline Specifically households consuming less than 300kWh paid the price of electricity at the 1994 rate plus an increase of 1631percent approved in 2000 Households that consumed more than 300 kWhper month paid the full price at the 1997 rate plus the same 1631 percentincrease The government paid the difference between the 1994 and 1997prices (L 238 million per month in mid-2002) directly to ENEE1
To assess the targeting performance of the subsidy that is what share ofthe subsidy goes to the poor we combine administrative information withhousehold survey data On the administrative data side we rely on infor-mation provided by ENEE This information includes (a) the percentage ofresidential clients by consumption level (b) the average amount of elec-tricity consumed by clients at each level (c) the total electricity bill with-out subsidy by level and (d) the total subsidy paid by level
On the household data side to estimate the proportion of poor householdsat the various consumption levels we use data from a survey implementedin 1999 by the Family Assistance Program (Programa de AsignacioacutenFamiliar or PRAF) with financing from the Inter-American DevelopmentBank and assistance from the International Food Policy Research InstituteThe survey was implemented in the bottom half of Hondurasrsquos municipali-
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
282
(c) The International Bank for Reconstruction and Development The World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
283
ties in terms of malnutrition as measured by the Ministry of Educationrsquosannual census of weight and height during the first year of primary schoolThe survey has modules on expenditures education health and the impactof Hurricane Mitch The advantage of this survey is that it includes dataon electricity consumption which are not available in the labor force sur-vey The disadvantage is that the survey is not nationally representativebut we assume that the results obtained here would also apply if the surveywere nationally representative2
Table 82 provides the results For example according to ENEErsquos admin-istrative records the share of households connected to its network withmonthly consumption below 20 kWh is 2031 percent (115723 house-holds) Of these we estimate from the PRAF survey that 4493 percent arepoor That is among all households that consume less than 20 kWh accord-ing to the PRAF survey about 45 percent have a level of per capita con-sumption that is below a reasonable poverty line for Honduras With aver-age consumption of 336 kWh per household per month the totalconsumption for this group is 388626 kWh Without the subsidy this groupwould have to pay a total bill of L 929256 but this bill is reduced toL 333282 when the subsidy of L 595973 is taken into account
Presenting statistics on errors of inclusion and exclusion is commonwhen assessing the targeting performance of a subsidy (see for exampleCornia and Stewart 1995) There are various ways to present these twotypes of errors The simplest way used in table 82 consists of computingthe share of households in poverty and receiving the subsidy as well as theshare of households not in poverty that are also receiving the subsidy Ingeneral as errors of inclusion increase errors of exclusion decrease andvice versa If all households receive the subsidy there are no errors ofexclusion but the errors of inclusion will likely be large because all non-poor households benefit from the subsidy
In table 82 among those households that are connected to ENEE6019 percent are nonpoor and receive the subsidy By contrast only 168percent are poor and do not receive the subsidy (nationally the share ofthe population in poverty not receiving the subsidy is larger becausemany poor households are not connected to the grid) In addition 2328percent of households that are poor receive the subsidy and 1485 percentof households that are nonpoor do not receive the subsidy From theseresults we can estimate that the ratio of poor versus nonpoor beneficiaries
(c) The International Bank for Reconstruction and Development The World Bank
284
TAB
LE 8
2 T
arg
etin
g P
erfo
rman
ce f
or
the
Elec
tric
ity
Sub
sid
y H
on
du
ras
Mid
-200
2
Tota
l bill
Sh
are
of
Co
nsu
mp
tio
nSh
are
of
clie
nts
Er
ror
of
A
vera
ge
wit
ho
ut
Tota
lsu
bsi
dy
leve
l Sh
are
of
in p
ove
rty
incl
usi
on
Er
ror
of
con
sum
pti
on
sub
sid
y b
y su
bsi
dy
by
spen
t o
n
(kW
h p
er)
clie
nts
b
ased
on
199
9 (1
)[1
-(2)
] ex
clu
sio
n
(kW
h p
er
cate
go
ryca
teg
ory
no
np
oo
r m
on
th)
()
(1)
dat
a (
) (2
)(
)(1
)(2
) (
)m
on
th)
(L t
ho
usa
nd
s)(L
th
ou
san
ds)
ho
use
ho
lds
()
0ndash20
203
144
93
111
8n
a3
3692
959
61
38
20ndash1
0022
69
356
614
60
na
586
75
096
271
77
34
100ndash
150
126
316
82
105
0n
a12
509
738
73
762
131
4
150ndash
200
111
610
98
994
na
175
3510
314
514
919
24
200ndash
250
925
156
47
81n
a22
454
116
185
746
203
5
250ndash
300
743
170
96
16n
a27
577
118
965
851
203
6
300+
165
310
15
na
168
mdash
mdashmdash
mdash
Tota
l10
024
96
601
91
6810
858
472
4123
820
818
1
mdash N
ot
avai
lab
le
na
No
t ap
plic
able
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
EN
EE (
2002
) an
d t
he
1999
PR
AF
surv
ey d
ata
(c) The International Bank for Reconstruction and Development The World Bank
is 039 (23286019) and so the number of nonpoor households receivingthe subsidy is more than twice as large as the number of poor householdsreceiving the subsidy3
The most important statistic however is the share of the subsidy givento the nonpoor This tells us how much poverty reduction is obtained (interms of the poverty gap as defined in appendix 81) for each lempira spenton the subsidy As computed in the last column of table 82 this share isabove 80 percent Part of the reason for the subsidyrsquos poor targeting per-formance is related to the level of the lifeline threshold which is set toohigh Some 835 percent of households with access to electricity consumeless than 300 kWh per month and hence qualify for the subsidy Further-more while the level of poverty is higher among households that consumeless than 100 kWh most of the subsidy is spent on households that con-sume between 100 and 300 kWh per month but these households are lesslikely to be poor
In part because the subsidy is not well targeted its impact on poverty issmall This is illustrated in table 83 which provides poverty measures withand without incorporating the value of the electricity subsidy in the overallconsumption aggregate The table provides three measures of poverty theheadcount index (the share of the population in poverty) the poverty gap(the distance separating the poor from the poverty line) and the squaredpoverty gap (appendix 81 provides a formal definition of these povertymeasures) The measures are provided for two alternative poverty lines cor-responding roughly to the extreme poor (L 400 per person per month) andthe poor (L 600 per person per month) Overall the changes in povertywhen subsidies are taken into account are small and these changes arelikely to be slightly overestimated because we do not take substitutioneffects due to the subsidy into account (if the subsidy were eliminatedelectricity prices would go up and households would substitute consump-tion toward other goods)
Although this is not reported here we carried out additional simulations toassess if the results are sensitive to the choice of the PRAF survey for theanalysis That is using a large set of variables common to both the PRAFand the nationally representative EHPHM survey and fitting a predictivemodel of electricity consumption in the PRAF survey we obtained a predic-tion for electricity consumption in the EHPHM survey and redid estimationsregarding targeting performance and the impact of the subsidy on poverty
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
285
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
281
TAB
LE 8
1 A
cces
s to
Ele
ctri
city
by
Inco
me
Gro
up
Ho
nd
ura
s 1
999
(per
cen
t)
Inco
me
dec
ile
Loca
tio
n a
nd
so
urc
e1
23
45
67
89
10M
ean
Nat
ion
al
Non
e75
065
346
533
525
920
416
19
76
64
730
4
ENEE
245
347
532
657
739
787
821
899
919
933
688
Colle
ctiv
e0
00
10
30
50
00
70
70
30
40
50
4
Indi
vidu
ala
05
00
00
03
02
02
11
01
11
15
05
Urb
an
Non
e31
016
610
09
04
01
91
40
40
20
47
5
ENEE
691
834
901
907
960
981
986
996
997
994
925
Colle
ctiv
e0
00
00
00
30
00
00
00
00
10
20
1
Indi
vidu
ala
00
00
00
00
00
00
00
00
00
00
00
Ru
ral
Non
e80
276
759
450
850
046
840
733
823
325
848
8
ENEE
193
232
401
480
496
508
547
648
715
633
485
Colle
ctiv
e0
00
10
40
60
01
81
81
11
11
90
9
Indi
vidu
ala
06
00
00
06
04
06
29
03
41
90
19
a S
mal
l pri
vate
ly o
wn
ed g
ener
ato
rs
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
Mar
ch 1
999
EPH
PM s
urv
ey
(c) The International Bank for Reconstruction and Development The World Bank
households For instance 245 percent of all households in the 1st or poor-est decile have access to electricity provided by ENEE compared with 933percent of households in the 10th richest decile Also as expected house-holds in urban areas have higher access rates than households in ruralareas The fact that richer households have higher rates of access to elec-tricity than poorer households is one of the reasons why from the point ofview of poverty reduction and in comparison with other programs thatcould be implemented to fight poverty electricity subsidies may not be welltargeted
The design of the electricity subsidy in Honduras follows a basic self-targeting structure that is households self-select themselves for inclusionin the subsidy scheme by not consuming more than a certain amount Thisimplies that neither the utility nor the government needs to design a morecomplex targeting mechanism based on means-testing whereby they usehousehold characteristics as proxies to identify who is poor and who is notand thereby who is eligible to receive the subsidy and who is not As of July2002 the subsidy was provided to all households that purchased electricityfrom ENEE and consumed less than 300 kWh per month the electricityconsumption lifeline Specifically households consuming less than 300kWh paid the price of electricity at the 1994 rate plus an increase of 1631percent approved in 2000 Households that consumed more than 300 kWhper month paid the full price at the 1997 rate plus the same 1631 percentincrease The government paid the difference between the 1994 and 1997prices (L 238 million per month in mid-2002) directly to ENEE1
To assess the targeting performance of the subsidy that is what share ofthe subsidy goes to the poor we combine administrative information withhousehold survey data On the administrative data side we rely on infor-mation provided by ENEE This information includes (a) the percentage ofresidential clients by consumption level (b) the average amount of elec-tricity consumed by clients at each level (c) the total electricity bill with-out subsidy by level and (d) the total subsidy paid by level
On the household data side to estimate the proportion of poor householdsat the various consumption levels we use data from a survey implementedin 1999 by the Family Assistance Program (Programa de AsignacioacutenFamiliar or PRAF) with financing from the Inter-American DevelopmentBank and assistance from the International Food Policy Research InstituteThe survey was implemented in the bottom half of Hondurasrsquos municipali-
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
282
(c) The International Bank for Reconstruction and Development The World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
283
ties in terms of malnutrition as measured by the Ministry of Educationrsquosannual census of weight and height during the first year of primary schoolThe survey has modules on expenditures education health and the impactof Hurricane Mitch The advantage of this survey is that it includes dataon electricity consumption which are not available in the labor force sur-vey The disadvantage is that the survey is not nationally representativebut we assume that the results obtained here would also apply if the surveywere nationally representative2
Table 82 provides the results For example according to ENEErsquos admin-istrative records the share of households connected to its network withmonthly consumption below 20 kWh is 2031 percent (115723 house-holds) Of these we estimate from the PRAF survey that 4493 percent arepoor That is among all households that consume less than 20 kWh accord-ing to the PRAF survey about 45 percent have a level of per capita con-sumption that is below a reasonable poverty line for Honduras With aver-age consumption of 336 kWh per household per month the totalconsumption for this group is 388626 kWh Without the subsidy this groupwould have to pay a total bill of L 929256 but this bill is reduced toL 333282 when the subsidy of L 595973 is taken into account
Presenting statistics on errors of inclusion and exclusion is commonwhen assessing the targeting performance of a subsidy (see for exampleCornia and Stewart 1995) There are various ways to present these twotypes of errors The simplest way used in table 82 consists of computingthe share of households in poverty and receiving the subsidy as well as theshare of households not in poverty that are also receiving the subsidy Ingeneral as errors of inclusion increase errors of exclusion decrease andvice versa If all households receive the subsidy there are no errors ofexclusion but the errors of inclusion will likely be large because all non-poor households benefit from the subsidy
In table 82 among those households that are connected to ENEE6019 percent are nonpoor and receive the subsidy By contrast only 168percent are poor and do not receive the subsidy (nationally the share ofthe population in poverty not receiving the subsidy is larger becausemany poor households are not connected to the grid) In addition 2328percent of households that are poor receive the subsidy and 1485 percentof households that are nonpoor do not receive the subsidy From theseresults we can estimate that the ratio of poor versus nonpoor beneficiaries
(c) The International Bank for Reconstruction and Development The World Bank
284
TAB
LE 8
2 T
arg
etin
g P
erfo
rman
ce f
or
the
Elec
tric
ity
Sub
sid
y H
on
du
ras
Mid
-200
2
Tota
l bill
Sh
are
of
Co
nsu
mp
tio
nSh
are
of
clie
nts
Er
ror
of
A
vera
ge
wit
ho
ut
Tota
lsu
bsi
dy
leve
l Sh
are
of
in p
ove
rty
incl
usi
on
Er
ror
of
con
sum
pti
on
sub
sid
y b
y su
bsi
dy
by
spen
t o
n
(kW
h p
er)
clie
nts
b
ased
on
199
9 (1
)[1
-(2)
] ex
clu
sio
n
(kW
h p
er
cate
go
ryca
teg
ory
no
np
oo
r m
on
th)
()
(1)
dat
a (
) (2
)(
)(1
)(2
) (
)m
on
th)
(L t
ho
usa
nd
s)(L
th
ou
san
ds)
ho
use
ho
lds
()
0ndash20
203
144
93
111
8n
a3
3692
959
61
38
20ndash1
0022
69
356
614
60
na
586
75
096
271
77
34
100ndash
150
126
316
82
105
0n
a12
509
738
73
762
131
4
150ndash
200
111
610
98
994
na
175
3510
314
514
919
24
200ndash
250
925
156
47
81n
a22
454
116
185
746
203
5
250ndash
300
743
170
96
16n
a27
577
118
965
851
203
6
300+
165
310
15
na
168
mdash
mdashmdash
mdash
Tota
l10
024
96
601
91
6810
858
472
4123
820
818
1
mdash N
ot
avai
lab
le
na
No
t ap
plic
able
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
EN
EE (
2002
) an
d t
he
1999
PR
AF
surv
ey d
ata
(c) The International Bank for Reconstruction and Development The World Bank
is 039 (23286019) and so the number of nonpoor households receivingthe subsidy is more than twice as large as the number of poor householdsreceiving the subsidy3
The most important statistic however is the share of the subsidy givento the nonpoor This tells us how much poverty reduction is obtained (interms of the poverty gap as defined in appendix 81) for each lempira spenton the subsidy As computed in the last column of table 82 this share isabove 80 percent Part of the reason for the subsidyrsquos poor targeting per-formance is related to the level of the lifeline threshold which is set toohigh Some 835 percent of households with access to electricity consumeless than 300 kWh per month and hence qualify for the subsidy Further-more while the level of poverty is higher among households that consumeless than 100 kWh most of the subsidy is spent on households that con-sume between 100 and 300 kWh per month but these households are lesslikely to be poor
In part because the subsidy is not well targeted its impact on poverty issmall This is illustrated in table 83 which provides poverty measures withand without incorporating the value of the electricity subsidy in the overallconsumption aggregate The table provides three measures of poverty theheadcount index (the share of the population in poverty) the poverty gap(the distance separating the poor from the poverty line) and the squaredpoverty gap (appendix 81 provides a formal definition of these povertymeasures) The measures are provided for two alternative poverty lines cor-responding roughly to the extreme poor (L 400 per person per month) andthe poor (L 600 per person per month) Overall the changes in povertywhen subsidies are taken into account are small and these changes arelikely to be slightly overestimated because we do not take substitutioneffects due to the subsidy into account (if the subsidy were eliminatedelectricity prices would go up and households would substitute consump-tion toward other goods)
Although this is not reported here we carried out additional simulations toassess if the results are sensitive to the choice of the PRAF survey for theanalysis That is using a large set of variables common to both the PRAFand the nationally representative EHPHM survey and fitting a predictivemodel of electricity consumption in the PRAF survey we obtained a predic-tion for electricity consumption in the EHPHM survey and redid estimationsregarding targeting performance and the impact of the subsidy on poverty
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
285
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
households For instance 245 percent of all households in the 1st or poor-est decile have access to electricity provided by ENEE compared with 933percent of households in the 10th richest decile Also as expected house-holds in urban areas have higher access rates than households in ruralareas The fact that richer households have higher rates of access to elec-tricity than poorer households is one of the reasons why from the point ofview of poverty reduction and in comparison with other programs thatcould be implemented to fight poverty electricity subsidies may not be welltargeted
The design of the electricity subsidy in Honduras follows a basic self-targeting structure that is households self-select themselves for inclusionin the subsidy scheme by not consuming more than a certain amount Thisimplies that neither the utility nor the government needs to design a morecomplex targeting mechanism based on means-testing whereby they usehousehold characteristics as proxies to identify who is poor and who is notand thereby who is eligible to receive the subsidy and who is not As of July2002 the subsidy was provided to all households that purchased electricityfrom ENEE and consumed less than 300 kWh per month the electricityconsumption lifeline Specifically households consuming less than 300kWh paid the price of electricity at the 1994 rate plus an increase of 1631percent approved in 2000 Households that consumed more than 300 kWhper month paid the full price at the 1997 rate plus the same 1631 percentincrease The government paid the difference between the 1994 and 1997prices (L 238 million per month in mid-2002) directly to ENEE1
To assess the targeting performance of the subsidy that is what share ofthe subsidy goes to the poor we combine administrative information withhousehold survey data On the administrative data side we rely on infor-mation provided by ENEE This information includes (a) the percentage ofresidential clients by consumption level (b) the average amount of elec-tricity consumed by clients at each level (c) the total electricity bill with-out subsidy by level and (d) the total subsidy paid by level
On the household data side to estimate the proportion of poor householdsat the various consumption levels we use data from a survey implementedin 1999 by the Family Assistance Program (Programa de AsignacioacutenFamiliar or PRAF) with financing from the Inter-American DevelopmentBank and assistance from the International Food Policy Research InstituteThe survey was implemented in the bottom half of Hondurasrsquos municipali-
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
282
(c) The International Bank for Reconstruction and Development The World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
283
ties in terms of malnutrition as measured by the Ministry of Educationrsquosannual census of weight and height during the first year of primary schoolThe survey has modules on expenditures education health and the impactof Hurricane Mitch The advantage of this survey is that it includes dataon electricity consumption which are not available in the labor force sur-vey The disadvantage is that the survey is not nationally representativebut we assume that the results obtained here would also apply if the surveywere nationally representative2
Table 82 provides the results For example according to ENEErsquos admin-istrative records the share of households connected to its network withmonthly consumption below 20 kWh is 2031 percent (115723 house-holds) Of these we estimate from the PRAF survey that 4493 percent arepoor That is among all households that consume less than 20 kWh accord-ing to the PRAF survey about 45 percent have a level of per capita con-sumption that is below a reasonable poverty line for Honduras With aver-age consumption of 336 kWh per household per month the totalconsumption for this group is 388626 kWh Without the subsidy this groupwould have to pay a total bill of L 929256 but this bill is reduced toL 333282 when the subsidy of L 595973 is taken into account
Presenting statistics on errors of inclusion and exclusion is commonwhen assessing the targeting performance of a subsidy (see for exampleCornia and Stewart 1995) There are various ways to present these twotypes of errors The simplest way used in table 82 consists of computingthe share of households in poverty and receiving the subsidy as well as theshare of households not in poverty that are also receiving the subsidy Ingeneral as errors of inclusion increase errors of exclusion decrease andvice versa If all households receive the subsidy there are no errors ofexclusion but the errors of inclusion will likely be large because all non-poor households benefit from the subsidy
In table 82 among those households that are connected to ENEE6019 percent are nonpoor and receive the subsidy By contrast only 168percent are poor and do not receive the subsidy (nationally the share ofthe population in poverty not receiving the subsidy is larger becausemany poor households are not connected to the grid) In addition 2328percent of households that are poor receive the subsidy and 1485 percentof households that are nonpoor do not receive the subsidy From theseresults we can estimate that the ratio of poor versus nonpoor beneficiaries
(c) The International Bank for Reconstruction and Development The World Bank
284
TAB
LE 8
2 T
arg
etin
g P
erfo
rman
ce f
or
the
Elec
tric
ity
Sub
sid
y H
on
du
ras
Mid
-200
2
Tota
l bill
Sh
are
of
Co
nsu
mp
tio
nSh
are
of
clie
nts
Er
ror
of
A
vera
ge
wit
ho
ut
Tota
lsu
bsi
dy
leve
l Sh
are
of
in p
ove
rty
incl
usi
on
Er
ror
of
con
sum
pti
on
sub
sid
y b
y su
bsi
dy
by
spen
t o
n
(kW
h p
er)
clie
nts
b
ased
on
199
9 (1
)[1
-(2)
] ex
clu
sio
n
(kW
h p
er
cate
go
ryca
teg
ory
no
np
oo
r m
on
th)
()
(1)
dat
a (
) (2
)(
)(1
)(2
) (
)m
on
th)
(L t
ho
usa
nd
s)(L
th
ou
san
ds)
ho
use
ho
lds
()
0ndash20
203
144
93
111
8n
a3
3692
959
61
38
20ndash1
0022
69
356
614
60
na
586
75
096
271
77
34
100ndash
150
126
316
82
105
0n
a12
509
738
73
762
131
4
150ndash
200
111
610
98
994
na
175
3510
314
514
919
24
200ndash
250
925
156
47
81n
a22
454
116
185
746
203
5
250ndash
300
743
170
96
16n
a27
577
118
965
851
203
6
300+
165
310
15
na
168
mdash
mdashmdash
mdash
Tota
l10
024
96
601
91
6810
858
472
4123
820
818
1
mdash N
ot
avai
lab
le
na
No
t ap
plic
able
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
EN
EE (
2002
) an
d t
he
1999
PR
AF
surv
ey d
ata
(c) The International Bank for Reconstruction and Development The World Bank
is 039 (23286019) and so the number of nonpoor households receivingthe subsidy is more than twice as large as the number of poor householdsreceiving the subsidy3
The most important statistic however is the share of the subsidy givento the nonpoor This tells us how much poverty reduction is obtained (interms of the poverty gap as defined in appendix 81) for each lempira spenton the subsidy As computed in the last column of table 82 this share isabove 80 percent Part of the reason for the subsidyrsquos poor targeting per-formance is related to the level of the lifeline threshold which is set toohigh Some 835 percent of households with access to electricity consumeless than 300 kWh per month and hence qualify for the subsidy Further-more while the level of poverty is higher among households that consumeless than 100 kWh most of the subsidy is spent on households that con-sume between 100 and 300 kWh per month but these households are lesslikely to be poor
In part because the subsidy is not well targeted its impact on poverty issmall This is illustrated in table 83 which provides poverty measures withand without incorporating the value of the electricity subsidy in the overallconsumption aggregate The table provides three measures of poverty theheadcount index (the share of the population in poverty) the poverty gap(the distance separating the poor from the poverty line) and the squaredpoverty gap (appendix 81 provides a formal definition of these povertymeasures) The measures are provided for two alternative poverty lines cor-responding roughly to the extreme poor (L 400 per person per month) andthe poor (L 600 per person per month) Overall the changes in povertywhen subsidies are taken into account are small and these changes arelikely to be slightly overestimated because we do not take substitutioneffects due to the subsidy into account (if the subsidy were eliminatedelectricity prices would go up and households would substitute consump-tion toward other goods)
Although this is not reported here we carried out additional simulations toassess if the results are sensitive to the choice of the PRAF survey for theanalysis That is using a large set of variables common to both the PRAFand the nationally representative EHPHM survey and fitting a predictivemodel of electricity consumption in the PRAF survey we obtained a predic-tion for electricity consumption in the EHPHM survey and redid estimationsregarding targeting performance and the impact of the subsidy on poverty
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
285
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
283
ties in terms of malnutrition as measured by the Ministry of Educationrsquosannual census of weight and height during the first year of primary schoolThe survey has modules on expenditures education health and the impactof Hurricane Mitch The advantage of this survey is that it includes dataon electricity consumption which are not available in the labor force sur-vey The disadvantage is that the survey is not nationally representativebut we assume that the results obtained here would also apply if the surveywere nationally representative2
Table 82 provides the results For example according to ENEErsquos admin-istrative records the share of households connected to its network withmonthly consumption below 20 kWh is 2031 percent (115723 house-holds) Of these we estimate from the PRAF survey that 4493 percent arepoor That is among all households that consume less than 20 kWh accord-ing to the PRAF survey about 45 percent have a level of per capita con-sumption that is below a reasonable poverty line for Honduras With aver-age consumption of 336 kWh per household per month the totalconsumption for this group is 388626 kWh Without the subsidy this groupwould have to pay a total bill of L 929256 but this bill is reduced toL 333282 when the subsidy of L 595973 is taken into account
Presenting statistics on errors of inclusion and exclusion is commonwhen assessing the targeting performance of a subsidy (see for exampleCornia and Stewart 1995) There are various ways to present these twotypes of errors The simplest way used in table 82 consists of computingthe share of households in poverty and receiving the subsidy as well as theshare of households not in poverty that are also receiving the subsidy Ingeneral as errors of inclusion increase errors of exclusion decrease andvice versa If all households receive the subsidy there are no errors ofexclusion but the errors of inclusion will likely be large because all non-poor households benefit from the subsidy
In table 82 among those households that are connected to ENEE6019 percent are nonpoor and receive the subsidy By contrast only 168percent are poor and do not receive the subsidy (nationally the share ofthe population in poverty not receiving the subsidy is larger becausemany poor households are not connected to the grid) In addition 2328percent of households that are poor receive the subsidy and 1485 percentof households that are nonpoor do not receive the subsidy From theseresults we can estimate that the ratio of poor versus nonpoor beneficiaries
(c) The International Bank for Reconstruction and Development The World Bank
284
TAB
LE 8
2 T
arg
etin
g P
erfo
rman
ce f
or
the
Elec
tric
ity
Sub
sid
y H
on
du
ras
Mid
-200
2
Tota
l bill
Sh
are
of
Co
nsu
mp
tio
nSh
are
of
clie
nts
Er
ror
of
A
vera
ge
wit
ho
ut
Tota
lsu
bsi
dy
leve
l Sh
are
of
in p
ove
rty
incl
usi
on
Er
ror
of
con
sum
pti
on
sub
sid
y b
y su
bsi
dy
by
spen
t o
n
(kW
h p
er)
clie
nts
b
ased
on
199
9 (1
)[1
-(2)
] ex
clu
sio
n
(kW
h p
er
cate
go
ryca
teg
ory
no
np
oo
r m
on
th)
()
(1)
dat
a (
) (2
)(
)(1
)(2
) (
)m
on
th)
(L t
ho
usa
nd
s)(L
th
ou
san
ds)
ho
use
ho
lds
()
0ndash20
203
144
93
111
8n
a3
3692
959
61
38
20ndash1
0022
69
356
614
60
na
586
75
096
271
77
34
100ndash
150
126
316
82
105
0n
a12
509
738
73
762
131
4
150ndash
200
111
610
98
994
na
175
3510
314
514
919
24
200ndash
250
925
156
47
81n
a22
454
116
185
746
203
5
250ndash
300
743
170
96
16n
a27
577
118
965
851
203
6
300+
165
310
15
na
168
mdash
mdashmdash
mdash
Tota
l10
024
96
601
91
6810
858
472
4123
820
818
1
mdash N
ot
avai
lab
le
na
No
t ap
plic
able
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
EN
EE (
2002
) an
d t
he
1999
PR
AF
surv
ey d
ata
(c) The International Bank for Reconstruction and Development The World Bank
is 039 (23286019) and so the number of nonpoor households receivingthe subsidy is more than twice as large as the number of poor householdsreceiving the subsidy3
The most important statistic however is the share of the subsidy givento the nonpoor This tells us how much poverty reduction is obtained (interms of the poverty gap as defined in appendix 81) for each lempira spenton the subsidy As computed in the last column of table 82 this share isabove 80 percent Part of the reason for the subsidyrsquos poor targeting per-formance is related to the level of the lifeline threshold which is set toohigh Some 835 percent of households with access to electricity consumeless than 300 kWh per month and hence qualify for the subsidy Further-more while the level of poverty is higher among households that consumeless than 100 kWh most of the subsidy is spent on households that con-sume between 100 and 300 kWh per month but these households are lesslikely to be poor
In part because the subsidy is not well targeted its impact on poverty issmall This is illustrated in table 83 which provides poverty measures withand without incorporating the value of the electricity subsidy in the overallconsumption aggregate The table provides three measures of poverty theheadcount index (the share of the population in poverty) the poverty gap(the distance separating the poor from the poverty line) and the squaredpoverty gap (appendix 81 provides a formal definition of these povertymeasures) The measures are provided for two alternative poverty lines cor-responding roughly to the extreme poor (L 400 per person per month) andthe poor (L 600 per person per month) Overall the changes in povertywhen subsidies are taken into account are small and these changes arelikely to be slightly overestimated because we do not take substitutioneffects due to the subsidy into account (if the subsidy were eliminatedelectricity prices would go up and households would substitute consump-tion toward other goods)
Although this is not reported here we carried out additional simulations toassess if the results are sensitive to the choice of the PRAF survey for theanalysis That is using a large set of variables common to both the PRAFand the nationally representative EHPHM survey and fitting a predictivemodel of electricity consumption in the PRAF survey we obtained a predic-tion for electricity consumption in the EHPHM survey and redid estimationsregarding targeting performance and the impact of the subsidy on poverty
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
285
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
284
TAB
LE 8
2 T
arg
etin
g P
erfo
rman
ce f
or
the
Elec
tric
ity
Sub
sid
y H
on
du
ras
Mid
-200
2
Tota
l bill
Sh
are
of
Co
nsu
mp
tio
nSh
are
of
clie
nts
Er
ror
of
A
vera
ge
wit
ho
ut
Tota
lsu
bsi
dy
leve
l Sh
are
of
in p
ove
rty
incl
usi
on
Er
ror
of
con
sum
pti
on
sub
sid
y b
y su
bsi
dy
by
spen
t o
n
(kW
h p
er)
clie
nts
b
ased
on
199
9 (1
)[1
-(2)
] ex
clu
sio
n
(kW
h p
er
cate
go
ryca
teg
ory
no
np
oo
r m
on
th)
()
(1)
dat
a (
) (2
)(
)(1
)(2
) (
)m
on
th)
(L t
ho
usa
nd
s)(L
th
ou
san
ds)
ho
use
ho
lds
()
0ndash20
203
144
93
111
8n
a3
3692
959
61
38
20ndash1
0022
69
356
614
60
na
586
75
096
271
77
34
100ndash
150
126
316
82
105
0n
a12
509
738
73
762
131
4
150ndash
200
111
610
98
994
na
175
3510
314
514
919
24
200ndash
250
925
156
47
81n
a22
454
116
185
746
203
5
250ndash
300
743
170
96
16n
a27
577
118
965
851
203
6
300+
165
310
15
na
168
mdash
mdashmdash
mdash
Tota
l10
024
96
601
91
6810
858
472
4123
820
818
1
mdash N
ot
avai
lab
le
na
No
t ap
plic
able
Sou
rce
Au
tho
rsrsquo e
stim
ates
usi
ng
EN
EE (
2002
) an
d t
he
1999
PR
AF
surv
ey d
ata
(c) The International Bank for Reconstruction and Development The World Bank
is 039 (23286019) and so the number of nonpoor households receivingthe subsidy is more than twice as large as the number of poor householdsreceiving the subsidy3
The most important statistic however is the share of the subsidy givento the nonpoor This tells us how much poverty reduction is obtained (interms of the poverty gap as defined in appendix 81) for each lempira spenton the subsidy As computed in the last column of table 82 this share isabove 80 percent Part of the reason for the subsidyrsquos poor targeting per-formance is related to the level of the lifeline threshold which is set toohigh Some 835 percent of households with access to electricity consumeless than 300 kWh per month and hence qualify for the subsidy Further-more while the level of poverty is higher among households that consumeless than 100 kWh most of the subsidy is spent on households that con-sume between 100 and 300 kWh per month but these households are lesslikely to be poor
In part because the subsidy is not well targeted its impact on poverty issmall This is illustrated in table 83 which provides poverty measures withand without incorporating the value of the electricity subsidy in the overallconsumption aggregate The table provides three measures of poverty theheadcount index (the share of the population in poverty) the poverty gap(the distance separating the poor from the poverty line) and the squaredpoverty gap (appendix 81 provides a formal definition of these povertymeasures) The measures are provided for two alternative poverty lines cor-responding roughly to the extreme poor (L 400 per person per month) andthe poor (L 600 per person per month) Overall the changes in povertywhen subsidies are taken into account are small and these changes arelikely to be slightly overestimated because we do not take substitutioneffects due to the subsidy into account (if the subsidy were eliminatedelectricity prices would go up and households would substitute consump-tion toward other goods)
Although this is not reported here we carried out additional simulations toassess if the results are sensitive to the choice of the PRAF survey for theanalysis That is using a large set of variables common to both the PRAFand the nationally representative EHPHM survey and fitting a predictivemodel of electricity consumption in the PRAF survey we obtained a predic-tion for electricity consumption in the EHPHM survey and redid estimationsregarding targeting performance and the impact of the subsidy on poverty
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
285
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
is 039 (23286019) and so the number of nonpoor households receivingthe subsidy is more than twice as large as the number of poor householdsreceiving the subsidy3
The most important statistic however is the share of the subsidy givento the nonpoor This tells us how much poverty reduction is obtained (interms of the poverty gap as defined in appendix 81) for each lempira spenton the subsidy As computed in the last column of table 82 this share isabove 80 percent Part of the reason for the subsidyrsquos poor targeting per-formance is related to the level of the lifeline threshold which is set toohigh Some 835 percent of households with access to electricity consumeless than 300 kWh per month and hence qualify for the subsidy Further-more while the level of poverty is higher among households that consumeless than 100 kWh most of the subsidy is spent on households that con-sume between 100 and 300 kWh per month but these households are lesslikely to be poor
In part because the subsidy is not well targeted its impact on poverty issmall This is illustrated in table 83 which provides poverty measures withand without incorporating the value of the electricity subsidy in the overallconsumption aggregate The table provides three measures of poverty theheadcount index (the share of the population in poverty) the poverty gap(the distance separating the poor from the poverty line) and the squaredpoverty gap (appendix 81 provides a formal definition of these povertymeasures) The measures are provided for two alternative poverty lines cor-responding roughly to the extreme poor (L 400 per person per month) andthe poor (L 600 per person per month) Overall the changes in povertywhen subsidies are taken into account are small and these changes arelikely to be slightly overestimated because we do not take substitutioneffects due to the subsidy into account (if the subsidy were eliminatedelectricity prices would go up and households would substitute consump-tion toward other goods)
Although this is not reported here we carried out additional simulations toassess if the results are sensitive to the choice of the PRAF survey for theanalysis That is using a large set of variables common to both the PRAFand the nationally representative EHPHM survey and fitting a predictivemodel of electricity consumption in the PRAF survey we obtained a predic-tion for electricity consumption in the EHPHM survey and redid estimationsregarding targeting performance and the impact of the subsidy on poverty
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
285
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
with these predictive values for electricity consumption in the EHPHM sur-vey The results obtained using this procedure were fairly similar
83 Alternative Targeting Indicators
The evidence suggests that Hondurasrsquos lifeline subsidy is badly targetedand therefore fails to benefit the poor very much In this section we com-pare the lifeline targeting technique to other means of targeting thatcould use more and better information to determine eligibility for utilitysubsidies
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
286
TABLE 83 Impact of the Electricity Subsidy on Poverty Honduras 1999
Without subsidy With subsidy
kWh consumed Headcount Poverty Squared Headcount Poverty Squaredmonth () gap () pov gap () gap () pov gap
Poverty line of L 400personmonth
0ndash20 4493 1242 599 4493 1229 592
20ndash100 3600 1045 426 3566 1019 411
100ndash150 2057 606 261 1682 560 235
150ndash200 1098 267 093 1098 224 072
200ndash250 1564 532 200 1564 456 151
250ndash300 1709 306 102 1709 238 079
More than 300 1015 270 112 1015 219 087
Poverty line of L 600personmonth
0ndash20 7101 2900 1456 7101 2885 1444
20ndash100 6347 2370 1160 6347 2336 1136
100ndash150 4474 1397 675 4339 1329 634
150ndash200 3126 819 335 2745 747 295
200ndash250 3580 1315 605 3416 1188 521
250ndash300 2915 1041 437 2915 936 373
More than 300 1797 633 303 1797 565 260
Source Authorsrsquo estimates using 1999 PRAF survey data
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
As mentioned in the previous section investigators often analyze the tar-geting performance of any given indicator such as the lifeline level of elec-tricity consumption used in Honduras using simple summary statisticssuch as the errors of inclusion and exclusion for a given targeting mecha-nism A generalization of this approach consists of using ROC curves toassess which indicatormdashin our case lifeline versus various potentialmeans-testing mechanismsmdashhas the best performance in identifying thepoor More precisely the idea is to use simple categorical regressions toassess how various targeting indicators predict the probability of beingpoor and to see how the two types of errors (exclusion of some poor house-holds and inclusion of some nonpoor households) vary with the choice of aparticular level of the indicator to determine eligibility In some cases onecan find a best overall eligibility criterion independently of the weightingof the two types of errors in policymakersrsquo objective function In other casessome weighting scheme is needed and for any given weighting scheme theROC curve can help select the best indicator
Our objective here is not to discuss the method in detail (see appendix82 for an outline of the basic idea behind the ROC curve) Rather wefocus on the empirical results for Hondurasrsquos electricity subsidy For eachindicator that can be used for targeting (lifeline or other) one associates acurve that plots the probability that a poor household will be classified aspoor against the probability that a nonpoor household will be classified aspoor for every possible value given to the indicator Note that the indicatorcan be complex that is it can consist of a combination of indicators as theregression can be multivariate If the ROC curve lies on the 45 degree linethe model has no predictive power because the probability that a poorhousehold would be classified as poor is no higher than the probability thata nonpoor household would be classified as poor The more the ROC curvebows upward the greater the modelrsquos predictive power A summary meas-ure of predictive power is the area underneath the ROC curve If the areais above 50 percent then the model has some predictive power An area of100 percent implies that the model predicts poverty perfectly
We used the methodology to assess how well various indicators per-formed for identifying the poor among the sample of households with aconnection to the electricity grid in Honduras We employed povertylines of L 400 (extreme poverty) and L 600 (poverty) to define the poorThe first model in table 84 (household characteristics) combines infor-
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
287
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
mation on a number of household characteristics including demograph-ics education employment status and geographic location The modelis better at identifying the extreme poor (area under the ROC curve of087) than the poor (area of 083) Within these household characteris-tics (that is with separate models with subsets of variables) demograph-ics and education variables are better than employment and locationvariables at identifying the poor Housing characteristics can also beused to identify the poor with a similar level of performance (area underthe ROC curve of 082 for the extreme poor and 081 for the poor) Withinhousing characteristics the size and quality of the house are better atidentifying the poor than other characteristics Finally the lifelinethreshold (related to the level of energy consumption in the household)has some predictive power (the area under the ROC curve is above 05)but less so than some other easily identifiable variables The bottom lineis that if the objective is to target the poor variables are available thatare better at doing so than the level of energy consumption (see appendix82 for examples of actual ROC curves)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
288
TABLE 84 Areas under ROC Curves for Alternative Targeting Mechanisms Honduras
Performance in identifying Performance in the extreme poor identifying the poor
(area under ROC curve (area under ROC curveModel percent) percent)
Household characteristics 87 83
Demographics 72 71
Educational attainment 71 72
Employment status 69 66
Geographic location (department) 66 63
Housing characteristics 82 81
Size of house 77 77
Quality of house 72 72
Access to water and sanitation 61 58
Electricity consumption 70 73
Note A larger area indicates better targeting performance
Source Authorsrsquo estimation
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
While it is not surprising that household or housing-based targeting indi-cators would be better at identifying the poor than householdsrsquo level ofenergy consumption one might believe that for a service provider or utilityto gather such information could be difficult or expensive Clearly means-testing (using correlates of poverty for targeting) requires information andgathering this information requires effort However experiences in othercountries suggest that the cost of doing so need not be very high if the sametype of information is used for targeting a range of social programs ratherthan utility subsidies only
More specifically one clear possibility for reducing the administrativecost of means-testing is to use a single system of means-testing at thenational level for many different programs In Latin America this has beendone with some success in Colombia (the System for Selection of Benefi-ciaries) and in Chile (the Committee for Social Municipal Assistance orCAS) among others In Chile for example as documented by Clert andWodon (2001) the CAS system is used as a targeting instrument not onlyfor water subsidies but also for the family income subsidy the social hous-ing subsidy and the pension subsidy scheme Because the fixed adminis-trative costs are spread across several programs the CAS is cost-effectiveIn 1996 for example administrative costs represented a mere 12 percentof the benefits distributed using the CAS score If the administrative costsof the CAS system had had to be borne by the water subsidy scheme alonethey would have represented 178 percent of the value of the subsidies Thecost of interviews for determining eligibility for the subsidies was US$865per household and the Ministry of Planning estimates that 30 percent ofChilean households underwent interviews which seems reasonable giventhat the target group for the subsidy programs is the poorest 20 percent ofthe population
84 Conclusion
Governments have a range of options for helping to reduce poverty Simi-larly private utilities have a number of options for helping their low-incomecustomers While decisions about the choice of a specific instrument arebased on various criteria interventions whose benefits are immediate visi-ble and administratively easy to implement and are supported not only by
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
289
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
the poor but also by the nonpoor or not as poor are attractive from a politi-cal economy point of view4
Lifeline subsidies for basic infrastructure or utility services have allthese characteristics The subsidies may take various forms but their keycharacteristic is that they are provided to all customers with a consump-tion level below a minimum threshold considered necessary for meetingbasic needs hence the use of the ldquolifelinerdquo expression Lifeline subsidieshave an immediate impact by reducing beneficiariesrsquo expenditures for agiven level of provision The benefits of the subsidies are easily understood(even though their costs may not be) Lifeline subsidies often enjoy wide-spread political support especially when the lifeline threshold is set suffi-ciently high so as to benefit the less poor as well as the poor or the medianconsumer as well as the low-income customer The subsidies are easy toimplement at relatively low administrative cost because no means-testingis involved
For poverty reduction however while the characteristics of lifeline sub-sidies help to muster support they may not ensure effectiveness or a goodcost-benefit ratio Indeed one of the major drawbacks of lifeline subsidiesis that they may not be well targeted When they are well designed life-lines can reach the poor through self-targeting That is if the lifelinethreshold is low enough only those who consume little will be eligible andthese customers may be comparatively poor In many instances howeverthe leakage of lifeline subsidies to the nonpoor is such that it dilutes theeffectiveness of the policy for poverty reduction
In this chapter we provided a partial evaluation of the lifeline or increas-ing block tariff electricity subsidy in Honduras With funding from the gov-ernment the public utility is offering electricity at greatly subsidized ratesfor those households with monthly consumption below 300 kWh Becausethe lifeline threshold is set so high 835 percent of the utilityrsquos residentialclients benefit from the subsidy At the same time 818 percent of the sub-sidy may well be spent on nonpoor households While this last statisticcould be lower if we were using a different method for measuring povertyit remains true that the impact on poverty of the subsidy is rather small incomparison to its cost The fact that the current subsidy is badly targeteddoes not mean that it could not be improved by reducing the lifeline thresh-old A lower lifeline subsidy as currently being considered by the govern-ment would have the potential of being more effective Alternative proxy
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
290
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
means-testing targeting mechanisms based on household or housing char-acteristics could also be used to improve targeting Nevertheless experi-ence in other countries such as Chile suggests that even with better means-testing other types of interventions would probably have a better impacton poverty per dollar spent than utility subsidies
Appendix 81 Definition of Poverty Measures
This appendix which is reproduced with minor changes from CoudouelHentschel and Wodon (2002) provides mathematical expressions for thepoverty measures used in table 83
Poverty Headcount
This is the share of the population that is poor that is the proportion of thepopulation for whom consumption or income y is less than the poverty linez Suppose we have a population of size n in which q people are poor Thenthe headcount index is defined as
H = nq
Poverty Gap
The poverty gap which is often considered as representing the depth ofpoverty is the mean distance separating the population from the povertyline with the nonpoor being given a distance of zero The poverty gap is ameasure of the poverty deficit of the entire population where the notion ofpoverty deficit captures the resources that would be needed to lift all thepoor out of poverty through perfectly targeted cash transfers It is definedas follows
PG = 1n
q
i=1z ndash
zyi
where yi is the income of individual i and the sum is taken only on thoseindividuals who are poor The poverty gap can be written as being equal tothe product of the income gap ratio and the headcount index of povertywhere the income gap ratio is itself defined as
PG = I H with
I = z ndash
zyq
where yq = 1q
q
i=1yi is the average income of the poor
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
291
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
Squared Poverty Gap
This is often described as a measure of the severity of poverty While thepoverty gap takes into account the distance separating the poor from thepoverty line the squared poverty gap takes the square of that distance intoaccount When using the squared poverty gap the poverty gap is weightedby itself so as to give more weight to the very poor Said differently thesquared poverty gap takes into account the inequality among the poor It isobtained as follows
P2 = 1n
q
i=1z ndash
zyi
2
It is important to use the poverty gap or the squared poverty gap in addi-tion to the headcount for evaluation purposes since these measure differ-ent aspects of income poverty Indeed basing evaluation on the headcountratio would consider as more effective those policies that lift the richest ofthe poor (those close to the line) out of poverty Using the poverty gap PGand the squared poverty gap P2 on the other hand puts the emphasis onhelping those who are further away from the line the poorest of the poor
Appendix 82 Identifying the Targeting Performanceof Various Indicators Using ROC Curves
Following Wodon (1997) denote by P Pndash and P+ the number of the poorthe number of the poor classified as nonpoor and the number of the poorclassified as poor by a model Also denote by NP NPndash and NP+ the num-ber of the nonpoor the number of the nonpoor classified as nonpoor andthe number of the nonpoor classified as poor Sensitivity SE = P+(Pndash + P+)= P+P is the fraction of poor households classified as poor Specificity SP= NPndash(NPndash + NP+) = NPndashNP is the fraction of nonpoor households clas-sified as nonpoor The errors of inclusion and exclusion can be defined as1 minus SP and 1 minus SE (other definitions could be used as well butROC curves are based on these definitions)
Nonpoor Poor
Predicted nonpoor SP = NPndash (NPndash + NP+) 1 ndash SE = Pndash (Pndash + P+)
Predicted poor 1 ndash SP = NP+ (NPndash + NP+) SE = P+ (Pndash + P+)
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
292
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
293
1 ndash
SP
SE
Hous
ing
Elec
tric
ity
000
010
020
030
040
050
060
070
080
090
100
100
090
080
070
060
050
040
030
020
010
000
FIG
UR
E A
21
RO
C C
urv
es U
sin
g H
ou
sin
g a
nd
Ele
ctri
city
Co
nsu
mp
tio
n M
od
els
(po
vert
y lin
e at
L 6
00 p
er p
erso
n p
er m
on
th)
Sou
rce
Au
tho
rs
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
When using a statistical package and running a probit or logit regressionfor poverty each observation is given an index value equal to the predictedright-hand-side of the regression This predicted value is used to classifythe households as poor or nonpoor with the computer typically using one-half as the cutoff point which we will denote by c (those above the cutoffpoint are classified as poor) However this cutoff point can be changed AROC curve is a graph that plots SE as a function of 1 ndash SP for alternativevalues of the cutoff point Figure A21 shows the ROC curves estimated forthose with access to electricity in Honduras with two different models thehousing model and the level of electricity consumption of the households(which is a generalization of the lifeline used for targeting the subsidy byENEE) At the origin c = 1 SE = 0 and SP = 1 At the upper right cornerc = 0 SE = 1 and SP = 0 The higher the ROC curve the better its pre-dictive power (a 45 degree line has no predictive power while a verticalline from the origin to the top of the box followed by a horizontal line untilthe upper right corner has perfect predictive power) Clearly the housingmodel performs better than the level of electricity consumption of house-holds in identifying the poor
The area below a ROC curve provides a summary statistic of the predic-tive value of the underlying model An area of 05 corresponds to the 45degree line which has no explanatory power An area of 1 corresponds toperfect prediction If the ROC curve of one targeting indicator or set ofindicators lies above the ROC curves of all the alternatives at all pointsthat indicator will typically be the best to target the poor for the class ofsocial welfare functions based on the two types of errors that can be com-mitted through targeting If two ROC curves intersect the choice of thebest indicator will depend on the normative weights the policymakerattaches to the two types of errors
Notes
1 The tariff structure for 1997 distinguishes between households consumingmore or less than 500 kWh per month For those households that consume lessthan 500 kWh per month the price is a flat rate of L 69 for the first 0 to 20 kWhThereafter the unit price per kWh is L 06979 for 20 to 99 kWh L 10173 from100 to 299 kWh and L 11829 from 300 to 499 kWh For households consum-ing more than 500 kWh per month the flat rate for the first 20 kWh is 70800Lempiras Then the unit rate per kWh is L 07161 for the next 80 kWh L 10438
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
294
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
for the next 200 kWh L 12137 for the next 200 kWh and L 13352 above500 kWh
2 Using variables common to the PRAF survey and the nationally representa-tive EHPHM labor force survey we predicted energy consumption in the EHPHMusing a model fitted in the PRAF survey The results obtained with the EHPHMsurvey for the assessment of the targeting performance of the subsidy and itsimpact on poverty were similar
3 Another way of defining the errors of inclusion and exclusion consists of con-sidering the fraction of subsidy recipients that are nonpoor as errors of inclusionand the fraction of households that are not recipients but are poor as errors ofexclusion According to this alternative definition the errors of inclusion are equalto 072 [6019(6019 + 2328)] while the errors of exclusion are equal to 010[168(168 + 1485)]
4 For issues relating to targeting its costs and the interplay with the politicaleconomy see for instance Besley and Kanbur (1993) Sen (1995)
References
The word ldquoprocessedrdquo describes informally reproduced works that may notbe commonly available through libraries
Besley T and R Kanbur 1993 ldquoThe Principles of Targetingrdquo In MLipton and J Van der Gaag eds Including the Poor Washington DCWorld Bank
Clert C and Q Wodon 2001 ldquoThe Targeting of Government Programsin Chile A Quantitative and Qualitative Assessmentrdquo In E Gacitua-Mario and Q Wodon eds Measurement and Meaning CombiningQuantitative and Qualitative Methods for the Analysis of Poverty andSocial Exclusion in Latin America Technical Paper no 518 Washing-ton DC World Bank
Cornia G A and F Stewart 1995 ldquoTwo Errors of Targetingrdquo In D Vande Walle and K Nead eds Public Spending and the Poor Theory andEvidence Baltimore Maryland The John Hopkins University Press
Coudouel A J Hentschel and Q Wodon 2002 ldquoPoverty Measurementand Analysisrdquo In J Klugman ed Poverty Reduction Strategies Source-book Washington DC World Bank
LIFELINE OR MEANS-TESTING ELECTRIC UTILITY SUBSIDIES IN HONDURAS
295
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank
Goacutemez-Lobo A and D Contreras 2000 ldquoSubsidy Policies for the UtilityIndustries A Comparison of the Chilean and Colombian Water SubsidySchemesrdquo University of Chile Department of Economics SantiagoProcessed
ENEE (Empresa Nacional de Energiacutea Eleacutectrica) 2002 ldquoAnalisis del sub-sidio que ortoga el Gobierno Central de Honduras a los Abonados Resi-denciales con consumos iguales o menores a 300 hWhmesrdquo Teguci-galpa Processed
Estache A V Foster and Q Wodon 2002 Accounting for Poverty inInfrastructure Reform Learning from Latin Americarsquos Experience Devel-opment Studies Washington DC World Bank Institute
Foster V and C Araujo Forthcoming ldquoA Case Study from GuatemalardquoIn V Foster and Q Wodon eds Energy and Utility Reform Lessonsfrom Latin America for Poverty Reduction Technical Paper WashingtonDC World Bank
Sen A 1995 ldquoThe Political Economy of Targetingrdquo In D Van de Walleand K Nead eds Public Spending and the Poor Theory and EvidenceBaltimore Maryland The John Hopkins University Press
Wodon Q 1997 ldquoTargeting the Poor Using ROC Curvesrdquo World Develop-ment 25(12) 2083ndash92
World Bank 2001 ldquoHonduras Poverty Diagnostic 2000rdquo Report no20531-HO Washington DC
QUENTIN WODON MOHAMED IHSAN AJWAD AND CORINNE SIAENS
296
(c) The International Bank for Reconstruction and Development The World Bank