China’s Sloping Land Conversion Program: Does Expansion ... · tious land...

China’s Sloping Land Conversion Program: DoesExpansion Equal Success?

Jintao Xu, Ran Tao, Zhigang Xu, and Michael T. Bennett

ABSTRACT. This paper uses a 2003 householdsurvey to examine implementation and impacts ofChina’s Sloping Land Conversion Program. We findthat land targeting has been strongly influenced byprogram goals, but that mistargeting also occurred.Using a treatment effects approach to evaluateprogram impact, we find evidence of positive impacton cropping, husbandry, and total income, though theresults are not robust enough to support governmentclaims of huge gains. We also find evidence that lackof participant choice could be dampening programimpacts, and that allowing households autonomycould lead to improvement in program cost-effec-tiveness and outcomes. (JEL Q28, Q57)

I. INTRODUCTION

In 1999, China initiated the most ambi-tious land conversion/afforestation pro-gram in the developing world. With a totalbudget of CNY 337 billion (around $48billion), the Sloping Land ConversionProgram (SLCP) plans to convert 14.67million ha of fragile cropland to forests (4.4million of which will be land with slopes of25 degrees or greater) and has an additional‘‘soft’’ target of afforesting an equal area ofwasteland (State Forestry Administration2003). To date, the program has alreadyretired and afforested more than 9 millionha of cropland (Economic Daily 2007). Assuch, pending successful completion, theprogram could have significant implicationsfor China’s forests and remaining naturalecosystems, representing a 10% to 20%increase in current national forest area,and roughly a 10% decrease in China’s

cultivated area (Hyde, Belcher, and Xu2003; ZGTJNJ 2001).1

The SLCP has expanded very fast sinceits initiation in 1999. By the end of 2003,fully 7.2 million ha of cropland had beenenrolled, and the program encompassedmore than 2,000 counties in 25 provincesacross China. In 2004, the central govern-ment continued to push expansion andclaimed that it is a key policy for restoringChina’s degraded ecosystems (e.g., ChinaGreen Times 2003, 2004, 2005). However,fast expansion does not necessarily implysuccess. There is still lack of empiricalevidence to evaluate whether the earlierphase of the SLCP was successful enough tojustify fast program expansion in recentyears. This paper aims to fill this gap on thebasis of a household and village-level survey

The authors are, respectively, professor and head ofthe Environmental Economics Program in China(EEPC), College of Environmental Sciences and Engi-neering, Peking University; senior research fellow, Centerfor Chinese Agricultural Policy (CCAP), Chinese Acad-emy of Sciences; senior research fellow, CCAP; andresearch fellow, EEPC, Peking University, and seniorresearcher, Forest Trends, Washington, D.C. Thisresearch was funded by the World Bank and the FordFoundation for the Taskforce on Forests and Grasslands,China Council for International Cooperation on Envi-ronment and Development. The authors wish to thankUma Lele and Susan Shen of the World Bank, GuofangShen of the Chinese Academy of Engineering, HeinMallee of the Ford Foundation, Scott Rozelle and EmiUchida of University of California, Davis, and YazhenGong and Jikun Huang of CCAP for their help indesigning and implementing the survey. The authors alsowish to thank one anonymous referee for helping tosignificantly improve the paper via several excellentcomments and suggestions. All findings, opinions, andmistakes are the sole responsibility of the authors.

Land Economics N May 2010 N 86 (2): 219–244ISSN 0023-7639; E-ISSN 1543-8325

E 2010 by the Board of Regents of theUniversity of Wisconsin System

1 This is comparable to the other large land set-asideprogram in the world, the U.S. Conservation ReserveProgram, which in 2000 had enrolled some 13.56 millionha, or nearly 10% of cropland in the United States(Heimlich 2003).

collected in three western provinces targetedby the program.

II. BACKGROUND

Stricken by a historic 267-day YellowRiver dry-out in 1997, and the 1998 YangtzeRiver floods that caused significant economicdamage and loss of life, the Chinese govern-ment claimed that it is necessary to takeimmediate action to alleviate water and soilerosion. In the aftermath of the floods,experts generally agreed that high rates ofdeforestation and the consequent increase inrates of soil erosion in the upper reaches ofthe Yangtze River Basin exacerbated, if notprecipitated, the disaster (World Bank 2001a;World Wildlife Fund 2003). In general, soilerosion is one of China’s most pressingenvironmental problems (Huang 2000). Anestimated 2 billion tons of silt is released intothe Yangtze and the Yellow rivers annually,with 65% of this coming from slopingcropland. Data suggest that west China, with70% of the approximately 6.07 million ha ofagricultural land with slopes greater than 25degrees, contributes the majority of this (Xu,Katsigris, and White 2001). The centralgovernment thus initiated the SLCP in 1999with particular emphasis on west China.

The SLCP is an important departurefrom China’s other water and soil conser-vation and forestry programs. It is one ofthe first, and certainly the most ambitious,‘‘payment for environmental services’’ pro-grams in China. Most other large nationalforestry programs, such as the NaturalForest Protection Program (initiated in1998) and the Northeast, North, andNorthwest China Green Belt Program(initiated in the late 1970s) are directlyimplemented by either state-owned forestenterprises or by local forest authorities. Incontrast, the SLCP uses a public paymentscheme that directly engages millions ofrural households as core agents of projectimplementation. As such, central to theprogram’s long-term success is its incentivecompatibility for participating farmers.

The program stipulates that farmers whoconvert degraded and highly sloping crop-

land back to either ‘‘ecological forest’’(defined as timber-producing forests), ‘‘eco-nomic forest’’ (orchards, or plantations oftrees with medicinal value), or grasslandwill be compensated with (1) an annual in-kind subsidy of grain, (2) a cash subsidy,and (3) free seedlings, provided to thefarmer at the beginning of the plantingperiod. To account for differences inregional average yields, the annual grainsubsidy has been set at 2,250 kg/ha in theYangtze River Basin, and 1,500 kg/ha in theYellow River Basin. The cash subsidy isCNY 300/ha of eligible land ($43/ha) peryear. Both grain and cash subsidies are foreight years if ecological forests are plantedand for five years or two years if economicforests or grasses are planted, respectively(State Forestry Administration 2003).

Finally, SLCP is most notable for its sheersize. Under the program, the State ForestryAdministration plans to convert around14.67 million ha of fragile cropland to forestby 2010 (4.4 million of which is estimated tobe on land with slopes of 25 degrees ormore) and also has a ‘‘soft’’ target ofafforesting an equal area of wasteland (StateForestry Administration 2003; World Wild-life Fund 2003). The central governmentpoured CNY7.68 billion ($1.1 billion) ingrain and cash subsidies into the programduring the three-year pilot period alone, andby the end of 2003 total accumulatedgovernment expenditures approachedCNY50 billion (over $7 billion), around68% of which has been for grain subsidies(authors’ calculations from State ForestryAdministration 2003). Fifteen million farm-ers have entered the program in just the firstfive years, and leaders have estimated thatupon completion it will affect 40 to 60million rural households (Uchida, Xu, andRozelle 2005). If the program is to becompleted as the State Forestry Adminis-tration has planned, total program expen-ditures will reach CNY 337 billion (around$48 billion). In comparison, 13.56 million haof cropland was enrolled in the U.S.Conservation Reserve Program in 2000,with estimated outlays in 2001 of $1.7 billion(Heimlich 2003).

220 Land Economics May 2010

III. PROGRAM EXPANSION

The SLCP has expanded very fast sinceits initiation in 1999. As detailed in Fig-ure 1, during the pilot phase (1999–2001) anaverage of 402,000 ha of cropland wasenrolled into SLCP annually. According tointernal government reports, upon fullimplementation the rate of enrollmentincreased more than sixfold, averagingalmost 3 million ha of cropland convertedper year during 2002–2003. By the end ofthe pilot phase, 1.2 million ha of croplandand 0.47 million ha of barren land had beenconverted, and SLCP was being implement-ed in 400 counties across 20 provinces (Xu,Katsigris, and White 2001; Uchida, Xu, andRozelle 2005). Just two years after this, atthe end of 2003, fully 7.2 million ha ofcropland had been enrolled and 4.92 millionha of barren land afforested, and the

program encompassed more than 2,000counties in 25 provinces.

Though the sheer scale and budget of theSLCP are encouraging signals of theChinese government’s growing commit-ment toward the environment, they do notguarantee success. In fact, a growing bodyof evidence suggests that there are problemsin program design and implementation thatmight risk the SLCP’s long-term goals (Xuand Cao 2001; Zuo 2001). A key worry isthat the program places undue burden andcost on local governments, which in turncould be causing problems in programadministration observed during the pilotphase, such as low survival rates of plantedtrees, insufficient delivery of compensationto farmers, lack of respect for the principalsof volunteerism, and difficulties in targetingand monitoring (Zuo 2001; Xu and Cao2001). It has only been since 2002, in fact,

FIGURE 1SLOPING LAND CONVERSION PROGRAM TOTAL CONVERTED AREA, 1999–2003

(MILLION HECTARES)SOURCE: DATA FROM XU ET AL. 2004; UCHIDA, XU, AND ROZELLE 2005; INTERNAL

CHINESE GOVERNMENT REPORTS.

86(2) Xu et al.: Sloping Land Conversion Program 221

that the central government has allocatedadministrative fees to provincial govern-ments for SLCP implementation, and thesestill appear to be insufficient. Fast expan-sion has thus created heavy administrativeneeds that have potentially exacerbatedshortfalls in required funds, thus leadingto problems in implementation and subsidydelivery.2

An important backdrop to the SLCP’sfast expansion is China’s unsuccessful grainpolicy, which during the second half of1990s involved large-scaled grain procure-ments at above-market prices and a subse-quent failed attempt at recentralizing grainmarkets. By 1999, the State Grain Bureauwas burdened by severe financial stress andstocks of aging and unsold grain estimatedto be larger than China’s annual production(Lu 1999; Zhong 2001).3 From the perspec-tive of the central government, the ever-increasing deficit called for action to reducegrain stocks. The SLCP was then consid-ered by the center as a way out, since itcould not only reduce grain output but alsohelp to consume the State Grain Bureau’sexisting grain stocks. Therefore, behind theSLCP’s high grain subsidy standard andfast expansion has been the additional goalof aiding the state grain storage system. Infact, program grain has been purchasedfrom state stocks at CNY 0.4/kg abovemarket prices, which by the end of 2003 hasresulted in a 24.55 million ton drawdown ofstocks and an implicit CNY 9.8 billionsubsidy to the State Grain Bureau (accord-ing to authors’ calculations based on StateForestry Administration 2003).

Another backdrop to the SLCP’s fastexpansion is the financial incentives facedby local governments, which in manyregions of China viewed the program asan opportunity to bring in much-neededgovernment funding, and thus made everyeffort to push for larger land conversionquotas. Given that state subsidies had to gothrough the hands of local implementingagencies, local governments usually firstoverreached their land conversion quotasand then bargained for more subsidies. Asshown later in this paper, diversion of fundsin the SLCP has been no less frequent thanin many other centrally funded programs asindicated in the literature on povertyreduction (World Bank 2001b; Park, Wang,and Wu 2002).

Since an important justification behindfast expansion has been claims made aboutthe success of the pilot phase, the remainderof this paper analyzes, on the basis of a 2003survey, the degree to which substantiveevidence exists to justify the fast programexpansion after the pilot phase.

IV. LACK OF VOLUNTEERISM,SHORTFALLS IN SUBSIDIES

Our data come from a household andvillage-level survey completed in 2003 bythe Center for Chinese Agricultural Policy(CCAP), Chinese Academy of Sciences. Thesurvey was conducted in the three provinceswhere the SLCP was first implemented,located at the upper reaches of the YellowRiver Basin and the Yangtze River Basin:Shaanxi, Gansu, and Sichuan. Two coun-ties per province, three townships percounty, two participating villages per town-ship, and 10 households per village wererandomly selected, for a total of 36 villagesurveys and 360 household surveys. Thesurveys provide a comprehensive and com-paratively long window into program im-plementation, since both the household andvillage surveys collected detailed informa-tion for 1999 and 2002 regarding bothgeneral characteristics and SLCP imple-mentation.

3 Between 1996 and 1998, the financial loss of stategrain enterprises grew by an astonishing CNY 100 billionand became the nonperforming loans of the StateAgricultural Development Bank, which dispatched loansto the State Grain Bureau for grain procurement (Lu1998).

2 For example, in a township in a key project countyin Shaanxi Province, half of the participating plots werenot inspected and compensated on time. In anothertownship of the same county, many participating plotshad yet to be inspected even three years after they hadentered the SLCP; though the county governmentrecruited 30 additional staff to deal with these problems,manpower has still been far short of that required toinspect some 67,000 ha of converted land.


One discovery from the survey is thepredominantly topdown approach towardimplementation that has been taken in thesample villages. Although the State Forest-ry Administration’s plan emphasizes thatfarmers should have autonomy of choice inthe participation decision, a significantportion of households reported that theyhad little or no choice about whether toparticipate. Based on the State ForestryAdministration’s plan, most if not all of thefarmers in our sample should be eligible toparticipate, since while emphasizing thathighly sloped plots should be targeted first,the plan also allows some leeway intargeting lower-sloped marginal land thathas an impact on the local watershed (StateForestry Administration 2003). However,as detailed in Table 1, only around 53% ofsurveyed households felt that they couldchoose to participate (61.7% of the partic-ipants and only 25.9% of nonparticipants).4

This ranges from 65.8% of households inShaanxi, to 45.5% in Sichuan, to only 31%

in Gansu Province. From our fieldwork, webelieve that these responses can be taken atface value since households were aware ofthe details of the program and cognizant oftheir choices. For example, respondentsreported lower levels of choice for aspectsof program implementation requiring tech-nical expertise, and therefore more likelythe purview of program officials; only 36%of participant households said they couldchoose what kinds of trees to plant on theirenrolled land, and only 34.5% and 29.9% ofparticipant households felt that they couldchoose which areas and which plots,respectively, to retire.

Survey results also give evidence that lackof autonomy is, in part, the result ofsystematic differences in local implementa-tion. Village share of households reportingthat they have autonomy of choice, com-pared between villages in the same town-ship, has a correlation coefficient of 0.7,significant at 1%. Results from binomiallogit analysis of household autonomy statusalso support this. Two models using,alternately, township and county fixedeffects, and conditioning on 1999 householdand village characteristics, both find thathousehold characteristics are statisticallyinsignificant. At the same time, the margin-al effects for 8 out of the 17 townshipindicators and 3 out of 5 county indicators

4 These numbers do not change significantly whencontrolling for eligibility in terms of having sloping land;88% of the sample has land with slope greater than 15degrees, and 76.5% has land with slope greater than 25degrees. Furthermore, of those households in the samplethat had no sloped land (comprising 12% of the sample),63% were found to be participating in the program.

TABLE 1

FARMER AUTONOMY IN SLOPING LAND CONVERSION PROGRAM PARTICIPATION (N 5 345)

Measure of Autonomy

% That Said ‘‘Yes’’

Participants,All (n 5 264)

Shaanxi(n 5 103)

Gansu(n 5 85)

Sichuan(n 5 76)

Were the villagers asked their opinion about the project andhow it could be best designed prior to the time that theproject was implemented? 42.8 41.7 41.2 46.1

When your village began the SLCP, did your household haveautonomy to choose whether to participate? 61.7 72.8 43.5 67.1

Did you have autonomy in choosing the types of trees toplant? 36.0 47.6 34.1 22.4

Did you have autonomy in choosing which areas to retire? 34.5 53.4 15.3 30.3Did you have autonomy in choosing which plots to retire? 29.9 40.8 12.9 34.2

Nonparticipants,All (n 5 81)

Shaanxi(n 5 11)

Gansu(n 5 34)

Sichuan(n 5 36)

Could you participate in the SLCP if you wanted to? 25.9 45.5 29.4 16.7

Source: Adapted from Table 5 in Xu et al. 2004, with a different subsample.


are significant at 5% or better.5 Theseindicators, furthermore, have large margin-al effects: the average absolute value ofimpact on probability is 48%. These resultssuggest that households in the sample havehad unequal access to the program and, insome cases, have been forced to participatewhen they would otherwise have not.

Lack of household autonomy in partici-pation choice runs counter to the program’sexplicitly stated principal of volunteerism(State Forestry Administration 2003). As iswell known in the literature (Pagiola, Land-ell-Mills, and Joshua Bishop 2002), use ofmarket-based voluntary mechanisms ofparticipation is key to the efficiency gainspromised by payment for environmentalservices programs over traditional com-mand-and-control approaches. In the caseof the SLCP, since no bidding mechanismexists to optimally match payer benefitswith participant costs, participation should,at minimum, be voluntary. This couldimprove cost-effectiveness by ensuring thathouseholds with the lowest opportunitycosts participate, and would minimize thepossibility that program participation ishaving negative welfare effects on someparticipants. The survey data, in fact,

provide evidence that both of these out-comes have occurred.

Table 2 compares the 1999 (pre-SLCP)net income per hectare of retired plots, usedas a rough measure of plot opportunitycost, with the real value of SLCP compen-sation standards (calculated as the cashstandard plus the monetized grain subsidystandard). Though on average the partici-pant rural households received a subsidyhigher than their opportunity costs, at leastin the case of Shaanxi and Sichuan, theSLCP standard was below pre-SLCP netincome of enrolled plots in a number ofcases. Among the 103 participant house-holds in Shannxi province, 7 householdsincurred net loss. In Gansu and Sichuan, 42out of 85 and 23 out of 76 participanthouseholds were net losers. respectively.6

Some may argue that this does notnecessarily indicate problematic implemen-tation, since one year’s observation of netincome is an imperfect measure of plotopportunity cost, and poorer householdsmight prefer a low guaranteed subsidy overa high but highly variable expected plotincome. However, if we take into accountthe fact that 1999 was a year of bad harvestdue to serious drought in the surveyedregions (and thus the 1999 cropping incomeis below the cropping income of an averageyear), this implies that using the 1999cropping income as the opportunity costfor program participation is more likely tounderestimate participant farmers’ realopportunity costs. Therefore, governmentsubsidies lower than the 1999 croppingincome for a significant share of participantfarmers is indeed a serious issue.

5 Full regression results are omitted here due to spacelimitations and are available from the authors uponrequest. Basically, the number of years a village has beenimplementing the program was also statistically insignif-icant in all models, providing evidence that householdresponses are not reflecting lack of information about theprogram. Household characteristics included are house-hold head age and years of education, householdpopulation, labor, per capita income, per capita land,nonagricultural share of income and labor, and whetherthe household head is a party member. Village charac-teristics are village population, average per capita income,average per capita agricultural land, share of villagepopulation engaged in nonfarm work, village leader andsecretary age and years of education, number of years thevillage has implemented the SLCP, share of village landwith slope greater than 15 degrees, whether the villageleader or village secretary worked previously in a forestrydepartment, and number of villagers working in county-level forestry departments. Village characteristics weresignificant only in the model with county indicators; thesewere household population (2), share of village popula-tion in off-farm work (+), and whether the villagesecretary worked before in a forestry department (2).

6 Further calculations show that the subsidy standardis below the 1999 net income of retired plots for a numberof nonautonomous households. This group makes up26% of the Gansu Province sample, in fact. For thesehouseholds, the 1999 net income from retired plots was, intotal, CNY 8,503 larger than the SLCP subsidy standard,equaling on a household basis about 5% of the averagetotal 1999 household net income. In Sichuan, 9% ofsample participants are nonautonomous households thatare net losers, totaling a loss of CNY 7,386, or roughly15% of the average total 1999 household net income. Thissuggests that these households would not have willinglyparticipated had they been given a choice.


TA

BL

E2

PA

RT

ICIP

AN

T1

999

NE

TIN

CO

ME

FR

OM

EN

RO

LL

ED

LA

ND

VE

RS

US

SL

OP

ING

LA

ND

CO

NV

ER

SIO

NP

RO

GR

AM

(SL

CP

)C

OM

PE

NS

AT

ION

ST

AN

DA

RD

S

Net

Lo

sin

gH

ou

seh

old

sN

etG

ain

ing

Ho

use

ho

lds

All

Part

icip

an

ts

All

No

Au

ton

om

yA

uto

no

my

All

No

Au

ton

om

yA

uto

no

my

All

No

Au

ton

om

yA

uto

no

my

Sha

an

xi

(n

510

3)

Nu

mb

ero

fh

ou

seh

old

s7

07

96

28

68

103

28

75

19

99

av

erage

net

inco

me

fro

men

roll

edla

nd

(RM

B/h

a)

4,8

33

—4,8

33

181

173

186

507

173

67

3T

ota

lco

nv

erte

dla

nd

are

a(h

a)

5.1

3—

5.1

36

8.1

12

4.2

843

.83

73.2

42

4.2

848

.95

Av

erag

ed

iffe

ren

ceb

etw

een

SL

CP

stan

da

rdan

d19

99

net

inco

me

(RM

B/h

a)a

23

,03

3—

23,0

33

1,6

19

1,6

27

1,6

14

1,2

93

1,6

27

1,1

27

Sh

are

of

sub

sid

yb

eyo

nd

199

9n

etin

com

eb2

1.6

8—

21.6

80.9

00.9

00

.90

0.7

20.9

00

.63

Ga

nsu

(n

585

)

Nu

mb

ero

fh

ou

seh

old

s4

222

20

43

26

17

85

48

37

19

99

av

erage

net

inco

me

fro

men

roll

edla

nd

(RM

B/h

a)

3,4

85

3,9

46

3,0

85

940

974

89

52,0

26

2,1

02

1,9

44

To

tal

con

ver

ted

lan

da

rea

(ha

)8.5

23.9

64.5

61

1.4

46.4

84

.96

19.9

71

0.4

49

.53

Av

erag

ed

iffe

ren

ceb

etw

een

SL

CP

stan

da

rdan

d19

99

net

inco

me

(RM

B/h

a)a

21

,68

52

2,1

46

21,2

85

860

826

90

52

226

23

02

214

4S

ha

reo

fsu

bsi

dy

bey

on

d1

99

9n

etin

com

eb2

0.9

42

1.1

92

0.7

10.4

80.4

60

.50

20.1

32

0.1

72

0.0

8

Sic

hu

an

(n

576

)

Nu

mb

ero

fh

ou

seh

old

s2

37

16

53

18

35

76

25

51

19

99

av

erage

net

inco

me

fro

men

roll

edla

nd

(RM

B/h

a)

5,3

71

5,9

18

5,1

43

1,0

31

785

1,1

57

2,4

57

2,3

19

2,5

23

To

tal

con

ver

ted

lan

da

rea

(ha

)7.4

72.1

95.2

81

5.2

65.1

410

.12

22.7

37.3

415.4

0A

ver

ag

ed

iffe

ren

ceb

etw

een

SL

CP

stan

da

rdan

d19

99

net

inco

me

(RM

B/h

a)a

22

,82

12

3,3

68

22,5

93

1,5

19

1,7

65

1,3

93

93

231

27

Sh

are

of

sub

sid

yb

eyo

nd

199

9n

etin

com

eb2

1.1

12

1.3

22

1.0

20.6

00.6

90

.55

0.0

40.0

90

.01

Sourc

e:2003

SL

CP

surv

eyd

ata

.a

Su

bsi

dy

gra

inw

as

con

ver

ted

toca

shb

ase

do

nth

en

ati

on

al

mark

etp

rice

of

RM

B1/k

g.

bT

his

isth

ed

iffe

ren

ceb

etw

een

the

SL

CP

stan

dard

an

d1999

net

inco

me,

as

ash

are

of

the

SL

CP

stan

dard

.T

hu

s,fo

rex

am

ple

,0.9

0m

ean

sth

at

90%

of

the

sub

sid

yst

an

dard

com

pen

sate

sa

ho

use

ho

ldb

eyo

nd

1999

net

inco

me.


Even more troubling are the findings,detailed in Table 3, that subsidies actuallyreceived by participants in the samplegenerally fell short of SLCP compensationstandards. Nonautonomous households re-ceived on average only 46% of their owedsubsidies in 2002, compared with theaverage 62% received by autonomoushouseholds, with this difference significantat 1%. In terms of the cash subsidy alone,nonautonomous households received onaverage only 21%, compared with 34% forautonomous households, with this differ-ence significant at 5%. This suggests, atminimum, that significant problems inimplementation exist. There were two mainreasons for such shortfalls. The first is thatlocal governments, in program implemen-tation, have retained some subsidies tomake up for expenditure shortfalls andtax arrears.7 The second is that programexpansion had been so fast that localgovernment agencies responsible for pro-gram supervision have not had sufficientmanpower to check whether the convertedland satisfies government requirements(such as tree types and survival rates).Therefore, delivery of compensation wasdelayed in many regions.8 Though it isworth examining whether these factors arerelated to the fact that nonautonomoushouseholds in the sample appear to be

carrying the brunt of these shortfalls,analysis of this is beyond the scope of thispaper. It is a fact, however, that manyfarmers have not received the full amount ofsubsidies owed them.

Overall, the State Forestry Administra-tion and provincial and subprovincialforestry departments have been primarilyresponsible for targeting general areas ofland for enrollment in the SLCP, as well asfor setting and distributing enrollmentquotas to local governments (Zuo 2001).9

This topdown approach raises the questionof whether participant welfare is beingadequately considered when choosing landto enroll, and whether political/institutionalfactors unrelated to participant welfare orenvironmental or economic conditionscould be influencing this choice. Shortfallsin subsidies actually received suggest thatthis could be a concern. Fundamentally, toevaluate implementation it is important tolook both at the targeting of land forenrollment, and the program impact onparticipant income, since it is only byhelping participants shift to other suffi-ciently lucrative and sustainable income-generating activities that program goals willbe achievable in the long run. We examinethis in the following sections.

V. LAND TARGETING

The survey finds evidence of mistargetingof plots for retirement in terms of theSLCP’s stated target of highly sloping land.Low-sloping plots (with slope less than 15degrees) were enrolled in the program in 26of the sample villages, comprising onaverage 21% of total sample retired land.On average 71% of this land in each village(100% in 17 villages) could have beenreplaced with unenrolled highly sloping

7 Our fieldwork found that this is related to recentrural tax reforms that deprived local governments of thepower to levy various fees on farmers, and also accordingto the SLCP plan, local governments were no longer ableto levy agricultural tax and fees on the retired land.Consequent high local government budget deficitscombined with serious rural tax incompliance in manyregions have created incentives for governments toexpand their SLCP enrollment quotas so as to increaseinflows of subsidies, a portion of which can be thenretained for program costs and tax arrears.

8 We observed in our survey in the three pilotprovinces, and in visits to other provinces (e.g., Hunanand Hebei) that retention of SLCP subsidy funds by localgovernments is prevalent. In many regions, the cashsubsidies never reached participating farmers. Again, thiswas related to the huge fiscal pressures local governmentsfaced after the rural tax reform and the fact that noagricultural tax could be levied on retired land so thatlocal governments in SLCP areas lost a significant shareof their revenue and had to resort to retention of upper-level transfers such as the SLCP subsidy.

9 In practice, bargaining between the central and thelocal governments on the land conversion quota hasalways been a part of the program. Given that subsidiesare in most cases higher than forgone income ofcultivation and need to go through the hands of localimplementing agencies and local governments, suchagencies usually overreached the land retirement quotaset by the center to bargain for more subsidies.


TA

BL

E3

AV

ER

AG

ES

HO

RT

FA

LL

SIN

GR

AIN

AN

DC

AS

HS

UB

SID

YIN

SU

RV

EY

ED

AR

EA

S,

200

2

Co

un

tya

nd

To

wn

ship

Gra

in(k

g/h

a)a

Cash

(RM

B/h

a)

To

tal

Sh

ort

fall

(RM

B/h

a)b

SL

CP

Sta

nd

ard

Act

ual

Del

iver

y

SL

CP

Sta

nd

ard

Act

ua

lD

eliv

ery

All

No

Au

ton

om

yA

uto

no

my

All

No

Au

ton

om

yA

uto

no

my

All

No

Au

ton

om

yA

uto

no

my

Shaa

nx

i(

n5

103

)

Yan

chu

an

1,5

00

300

Ya

nsh

uig

ua

n50

61

,29

93

91

25

02

91,2

69

501

1,3

80

Maji

ah

e46

623

85

45

59

38

66

1,2

76

1,5

24

1,1

89

Yu

ju9

42

31

93

81

30

1,6

98

1,7

63

1,6

07

Liq

ua

n1

,50

0300

Ya

nx

ia1

,07

486

31,1

22

11

20

13

76

14

937

542

Jian

lin

g1

,50

01

,50

01,5

00

48

07

12

52

300

229

Ch

igen

1,4

71

1,5

00

1,4

68

78

96

76

251

204

257

Gan

su(

n5

85)

Jin

gn

ing

1,5

00

300

Zh

ipin

g57

451

76

38

10

411

49

41,1

22

1,1

69

1,0

68

Gan

go

u95

763

91,1

79

13

79

416

77

07

1,0

67

454

Lin

gzh

i1

,17

01

,10

01,2

17

20

119

820

34

29

502

380

Lin

xia

1,5

00

300

Zh

an

gzi

go

u49

960

20

86

83

10

01,2

15

1,1

14

1,7

00

Tie

zhai

00

05

05

61,7

95

1,8

00

1,7

44

Hex

i58

827

01,1

31

36

14

74

1.1

76

1,5

16

595

Sic

hua

n(

n5

76)

Ch

ao

tian

2,2

50

300

Data

n1

,84

92

,25

01,4

08

87

11

75

46

14

183

1,0

88

Zh

on

gzi

2,0

50

2,0

20

2,1

22

00

05

00

530

428

Sh

ah

e2

,17

72

,12

82,2

50

39

50

24

334

372

276

Li

2,5

00

300

Sh

an

gm

eng

2,1

60

2,2

50

2,1

50

10

70

11

82

84

300

282

Pu

xi

2,2

50

2,2

50

2,2

50

23

130

022

569

075

Gu

erg

ou

61

8—

618

50

—5

01,8

82

—1,8

82

Aver

age

85

670

51,3

43

70

49

11

11,0

21

1,1

77

1,6

02

Sourc

e:2003

SL

CP

surv

eyd

ata

.N

ote

:S

LC

P,

Slo

pin

gL

an

dC

on

ver

sio

nP

rogra

m.

aS

um

of

corn

,w

hea

t,w

hit

ean

dp

ad

dy

rice

,an

dw

hea

tfl

ou

rsu

bsi

die

s.B

oth

wh

ite

rice

an

dw

hea

tfl

ou

rw

ere

con

ver

ted

tou

nh

usk

edw

eigh

teq

uiv

ale

nts

at

afa

cto

ro

f1:1

.4.

bT

his

valu

esgra

inat

the

nati

on

al

pri

ceo

fR

MB

1/k

g.


land (slope greater than 25 degrees) in thesample. This indicates that considerationsother than plot slope have been importantin the enrollment choice in these villages.Plot quality and opportunity cost for thehousehold is likely an important factor andis not necessarily directly associated withslope.10 In villages that enrolled low-slopingland, an average of 37.4% of it wasdescribed as ‘‘low quality’’ by the house-hold, and 36.4% was affected by a disaster(mainly drought) in 1999. The program’sother goal of poverty alleviation raises thepossibility that enrollment targeting mightalso be influenced by household character-istics independent of plot traits. The trans-action costs of program implementationand the political economy of the village,where considerations of equity likely comeinto play when deciding who gets programsubsidies, could also play a part, as couldupper-level pressure to enroll land, as wellas rent-seeking behavior by local govern-ments.

To examine these issues, we modelenrollment of plot i in the SLCP by theend of 2002 as the outcome of a latentchoice process,

Qi~xiazzHi bHzzV

i bVzdidzei, [1]

which is a function of 1999 plot character-istics (xi), the 1999 characteristics of thehousehold (zi

H) and village (ziV) associated

with the plot, as well as provincial andtownship indicators (di) and other unob-served aspects of the choice (ei). Whetherplot i is enrolled in the SLCP is thus theobserved outcome of this process whereby

plot i isenrolled in SLCP if Qi (:)v0

not enrolled in SLCP if Qi (:)§0

�:

Assuming that the values of ei are anindependent and identically distributed

logistic, the probability that plot i isenrolled in the SLCP is

P(plot i is enrolled in SLCP)~

L(xiazzHi bHzzV

i bVzdid), [2]

where L(.) denotes the logistic cumulativedistribution function.

After data cleaning, 345 households wereselected for the analysis, with a total of 2,004plots. Table 4 details the characteristics ofthese sample plots. Overall, 27% of thesample plots were enrolled in the SLCP by2002, ranging from 48% in Shaanxi Province,to 20.9% in Sichuan, and 18% in Gansu. Plotcharacteristics included in the model are pre-SLCP (1999) plot slope, size, land quality,irrigation conditions, land rights held by thehousehold over the plot, and plot accessibil-ity. Generally speaking, the Shaanxi andSichuan samples have a large share of highlysloping plots and land that is low quality,whereas the Gansu sample has a much highershare of low-sloping, high-quality plots. TheShaanxi plots are also, on average, almosttwice as large as those in Gansu and Sichuanand were much more often affected by adisaster in 1999.

Table 5 details the characteristics of thehouseholds and villages in the sample.Program implementation generally beganearliest in Shaanxi Province (where 67% ofthe sample villages started the SLCP in1999), followed by Sichuan (where 83% ofthe sample villages started in 2000) and thenGansu (where 50% of the sample villagesdid not start until 2001 or 2002). Toexamine the influence of household charac-teristics on plot choice, the model includesas explanatory variables household popu-lation, household head age and years ofeducation, per capita income and land,household labor, the share of income fromoff-farm sources, and the share of laborengaged in part-time or full-time off-farmwork. Households in Shaanxi Province aregenerally the poorest in the sample (1999per capita income of CNY 991) and havethe highest share of steeply sloping land(72%). They also have the highest per capitaland area and lowest degree of land

10 In the sample, the correlation coefficients betweenwhether a plot is highly sloping (slope greater than 25degrees), and two measures of plot quality (1999 per hanet income from the plot, and whether the householdconsiders the plot to be ‘‘high quality’’) are 20.19 and20.33, respectively, both significant at 0.1%.


fragmentation. In comparison, Sichuan andGansu households are richer (CNY 1,435and CNY 1,566 per capita, respectively)and have lower per capita land and agreater degree of land fragmentation.Gansu households also have a relativelylow average share of land that is steeplysloped (31%).

To control for and examine the impactsof heterogeneity in local conditions, modelexplanatory variables also include village1999 per capita income, per capita land, andshare of village population in off-farm wagework (defined as part- or full-time off-farmday work, both in and out of the village, aswell as work that involves leaving the villagewithout returning for at least a week).

Villages in Shaanxi have lower populationdensity but depend more heavily on agri-culture for their livelihood, as reflected intheir low average number of rural enter-prises in the village (0.25) and low averageshare of village population engaged innonfarm wage work (11%, defined as workthat involves leaving the village for a weekor more). Gansu villages have on average0.33 rural enterprises and fully 23% ofvillage population engaged in nonfarmwage work in 1999, and Sichuan villageshave 1.75 rural enterprises and 15% ofvillage population engaged in nonfarmwage work. Villages in Shaanxi are alsogenerally smaller, with the average 1999population being 510, as compared with

TABLE 4

SAMPLE PLOT CHARACTERISTICS, 1999

All (n 5 2,004)Shaanxi

(n 5 560)Gansu

(n 5 755)Sichuan

(n 5 689)

MeanSt.

Dev. MeanSt.

Dev. MeanSt.

Dev. MeanSt.

Dev.

Plot Characteristics

Plot size (ha) 0.15 0.17 0.22 0.21 0.12 0.11 0.12 0.181999 net income/ha (RMB) 4,473 10,260 2,312 6,445 3,830 6,647 6,935 14,628Distance

To nearest road (km) 0.76 1.34 0.82 1.18 0.82 1.65 0.65 1.03To nearest gully/ditch (km) 1.03 2.19 1.44 2.85 1.07 2.24 0.66 1.25To home (km) 0.88 1.08 0.97 0.87 0.91 1.11 0.76 1.17

Share of Plots (%)

Enrolled in the SLCP 27.3 47.9 17.9 20.9Affected by a disaster in 1999a 24.6 62.0 9.0 11.2With slope

.25u 37.6 48.6 19.2 48.915u–25u 14.4 18.9 10.2 15.2,15ub 48.0 32.5 70.6 35.8

With land qualityHigh 32.8 25.4 44.5 26.1Medium 29.1 34.3 26.8 27.4Low 38.1 40.4 28.7 46.4

Irrigated withSurfacewater 12.5 2.7 14.0 18.9Groundwater 1.4 0.7 2.5 0.9Other 1.4 0.0 1.2 2.9No irrigation 84.6 96.6 82.3 77.4

That arePrivate land 6.1 3.9 2.9 11.5Responsibility or ration land 81.2 74.5 90.3 76.6Contract land 8.2 17.9 5.2 3.8Otherc 4.4 3.8 1.6 8.1

a Defined as an exogenous negative production shock, including drought, flood, severe insect infestation, windstorm, and hail.b Includes paddy and terraced fields, which comprised 5.8% of the plots in the sample.c Includes developed wasteland, and land transferred into or out of the household.


TA

BL

E5

HO

US

EH

OL

DA

ND

VIL

LA

GE

CH

AR

AC

TE

RIS

TIC

S,

1999

Vari

ab

le

All

Sh

aan

xi

Ga

nsu

Sic

hu

an

Mea

nS

t.D

ev.

Mea

nS

t.D

ev.

Mea

nS

t.D

ev.

Mea

nS

t.D

ev.

Ho

use

ho

ldch

ara

cter

isti

csn

5345

n5

11

4n

511

9n

511

2%

of

sam

ple

part

icip

ati

ng

inS

LC

P7

6.5

90

.471.4

67

.9%

of

part

icip

an

tsw

ith

slo

pin

gla

nd

(slo

pe

.15u)

90.2

97

.176.5

96

.1%

of

no

np

art

icip

an

tsw

ith

slo

pin

gla

nd

(slo

pe

.15u)

81.5

81

.873.5

88

.9H

ou

seh

old

hea

da

ge

44.3

11

.444

.41

0.6

44.9

11.7

43

.71

1.9

Ho

use

ho

ldh

ead

yea

rso

fed

uca

tio

n4.6

73

.45

.53

3.3

4.5

73.7

3.8

93.1

Ho

use

ho

ldp

op

ula

tio

n4.8

11

.64

.75

1.7

5.1

11.6

4.5

61.4

Ho

use

ho

ldp

erca

pit

ain

com

e(R

MB

)1

,33

01

,21

2991

1,0

53

1,4

35

1,2

36

1,5

66

1,2

71

No

na

gri

cult

ura

lsh

are

of

ho

use

ho

ldp

erca

pit

ain

com

e0.3

90

.39

0.4

00.4

60.4

40.3

20

.32

0.3

6H

ou

seh

old

per

cap

ita

ara

ble

lan

d(h

a)

0.1

90

.141

0.2

30.1

28

0.1

60.1

15

0.1

80.1

70

Ho

use

ho

ldla

bo

r3.5

11

.32

3.6

71.5

93.5

61.2

33

.29

1.0

6N

on

agri

cult

ura

lsh

are

of

ho

use

ho

ldla

bo

ra

0.3

40

.26

0.3

00.2

60.3

90.2

60

.34

0.2

5N

um

ber

of

plo

ts5.8

12

.23

4.9

11.8

36.3

42.2

46

.15

2.3

2S

hare

of

ag

ricu

ltu

ral

lan

dw

ith

slo

pe

.1

5u

0.5

70

.357

0.7

20.2

97

0.3

10.2

99

0.6

80.3

18

Vil

lag

ech

ara

cter

isti

csn

53

6n

51

2n

512

n5

12

Vil

lag

ep

op

ula

tio

n79

0.5

586

.25

10

.222

1.8

1,1

77.2

79

7.1

684

.238

4.0

Vil

lag

eaver

age

per

cap

ita

inco

me

(RM

B)

66

136

6535

221

672

251

776

532

Vil

lag

eaver

age

per

cap

ita

cro

pla

nd

(ha)

0.1

50

.107

0.1

90.1

56

0.1

50.0

90

0.1

20.0

25

Sh

are

of

vil

lage

po

pu

lati

on

wo

rkin

gin

no

nfa

rmw

ag

ew

ork

b0.1

60

.112

0.1

10.1

01

0.2

30.0

73

0.1

50.1

26

Mig

ran

tla

bo

ras

share

of

vil

lage

po

pu

lati

on

0.1

30

.099

0.0

60.0

54

0.2

00.0

80

0.1

10.1

03

Nu

mb

ero

fru

ral

ente

rpri

ses

0.7

81

.569

0.2

50.6

22

0.3

31.1

55

1.7

52.1

37

Vil

lag

ele

ad

eryea

rso

fed

uca

tio

n7.2

82

.77

.75

2.8

6.6

73.2

7.4

22.3

Vil

lag

ese

cret

ary

yea

rso

fed

uca

tio

n7.2

93

.68

.46

3.7

7.0

83.9

6.3

33.2

Vil

lag

ele

ad

erage

40.0

07

.441

.92

6.7

40.7

58.2

37

.33

7.0

Vil

lag

ese

cret

ary

age

45.0

88

.343

.50

6.7

47.5

87.8

44

.17

10.1

Inst

itu

tio

nal

fact

ors

Yea

rvil

lage

beg

an

the

SL

CP

200

00

.96

1999

0.6

720

00

1.2

42

000

0.4

3S

hare

of

vil

lage

agri

cult

ura

lla

nd

wit

hsl

op

e.

15u

0.5

80

.318

0.8

10.1

70

0.2

40.2

03

0.6

90.2

29

%o

fvil

lage

lead

ers

that

wo

rked

pre

vio

usl

yin

afo

rest

ryd

epart

men

t3

0.6

8.3

8.3

75

.0%

of

vil

lag

ese

cret

ari

esth

at

wo

rked

pre

vio

usl

yin

afo

rest

ryd

epart

men

t1

6.7

0.0

16.7

33

.3N

um

ber

of

vil

lag

ers

wo

rkin

gin

the

cou

nty

fore

stry

dep

art

men

t0.1

90

.62

0.3

30.8

90.2

50.6

20

Note

:S

LC

P,

Slo

pin

gL

an

dC

on

ver

sio

nP

rogra

m.

aC

alc

ula

ted

usi

ng

the

nu

mb

ero

fh

ou

seh

old

lab

ore

rsw

ork

ing

part

-tim

eo

rfu

ll-t

ime

off

-farm

,an

dso

no

tm

utu

all

yex

clu

sive

wit

hagri

cult

ura

lla

bo

r.bIn

clu

des

mig

ran

tla

bo

ran

dd

ay

wo

rker

sw

ork

ing

bo

tho

uts

ide

an

dw

ith

inth

evil

lage.


1,177 and 684 for Gansu and Sichuan,respectively.

The effects of lag time and programtransactions costs are captured with thenumber of years the village has beenimplementing the SLCP, and the villagepopulation, since implementation and mon-itoring likely take more time in largervillages. The model also includes share ofvillage agricultural land with slopes greaterthan 15 degrees to instrument for SLCPquotas distributed to villages, since thoughquota determination involves a degree ofnegotiation between local governments andforestry officials, local geographic condi-tions are an important baseline determi-nant. Finally, we include variables tocontrol for institutional heterogeneity thatcould influence program implementation.These include whether the village leader andwhether the village secretary previouslyworked at a forestry department, thenumber of villagers that work in the countyforestry department, and village leader andvillage secretary age and years of education.

To gain insight into the influence thathouseholds versus local governments havein plot enrollment choice, the model isestimated on the full sample as well as thesubsamples of autonomous and nonauton-omous households. In addition, two differ-ent forms of the model are estimated toexamine the degree to which the systematicregional variation in plot and householdcharacteristics could be picking up otherregional effects unrelated to plot traits. Thefirst model characterizes targeting as adirect function of plot, household, andvillage characteristics, while the secondmodel includes interaction terms betweenprovincial dummies and plot size, incomeper hectare, land quality, distance to nearestgully or ditch, household head per capitaincome, and household head per capitaland. Model marginal effects and signifi-cance levels are presented in Tables 6 and 7.

Model results provide evidence that whileplot targeting has been strongly influencedby program goals, other factors have alsobeen important. Overall, highly sloping,low-quality plots that are the least accessi-

ble to households are much more likely tobe enrolled in the SLCP. That a plot has aslope greater than 25 degrees increasesprobability of enrollment by 11% to 27%,and if it is irrigated with surface water, itsprobability of enrollment is reduced at themargin by 5% to 13%. Whether a plot wasaffected by a disaster in 1999 is even moreimportant, since if it was, the probability ofenrollment increased by fully 34% to 48%.Larger plots and plots with shorter-term,more flexible household property righttypes (i.e., either ‘‘responsibility,’’ ‘‘ration,’’or ‘‘contract’’ land) are also much morelikely to be retired, suggesting that transac-tion costs are being minimized in imple-mentation.11 The significant and negativeeffect of distance to the nearest road in fourof the six models also suggests this, sinceplots close to roads are easier to monitor.12

Though we are troubled by the lack ofhousehold autonomy in participation choiceseen in the sample, it is encouraging to seethat household characteristics are statisticallyinsignificant in the nonautonomous subsam-ple, since this suggests that selection ofhouseholds into the SLCP for this grouphas been based primarily on land character-istics.13 That said, comparison between the

11 Land in most Chinese villages can be divided intotwo types: private plots (ziliu di, around 6%) andcollectively controlled land (jiti di, more than 90%). Inmost villages, leaders do not intervene into decisions onprivate plots, and farmers enjoy a fairly high degree ofsecurity. Collectively controlled land includes threedifferent tenure forms: ration land (kouliang tian), whichgoes to farmers mainly to meet household subsistencerequirements with no tax obligations; responsibility land(zeren tian), which goes to farmers on the condition thatfarmers deliver a low-priced grain or cotton quota to thestate; and contract land (chengbao tian), which isauctioned off or allocated by village leaders for a fee(Rozelle et al. 2002; Liu, Carter, and Yao 1998).

12 However, this could also be viewed as supportinganecdotal evidence found in other case studies of oneaspect of potentially nonoptimal implementation, where-by local leaders target plots close to roads in order to‘‘showcase’’ implementation for higher-level officials(Zuo 2001; Xu and Cao 2001).

13 This interpretation is strengthened by evidence thatland distribution within villages in China in terms ofeffective per capita land is generally based on consider-ations of equity, suggesting that correlation between plotcharacteristics and household characteristics withinvillages is low (Rozelle et al. 2002; Kung 1995).


TABLE 6

PLOT ENROLLMENT IN THE SLOPING LAND CONVERSION PROGRAM (SLCP) BY THE END OF 2002, BINOMIAL LOGIT

MARGINAL EFFECTS

Explanatory Variablesa

Levels

All(n 5 2,004)

Autonomy(n 5 1,066)

No Autonomy(n 5 938)

1999 Plot Characteristics

Plot size (ha) 0.313**** 0.444**** 0.137***1999 income/ha (RMB) 20.000*** 20.000*** 20.000Plot affected by a disaster in 1999 (1 5 yes) 0.459**** 0.449**** 0.359****Slope . 25u 0.247**** 0.274**** 0.123****Slope 15u–25u 0.091** 0.082 0.060*High-quality land 20.034 20.062 20.004Low-quality land 0.053** 0.127*** 0.000Land rights

Responsibility or ration land 0.109*** 0.163** 0.023Contract land 0.201* 0.274* 0.021Other 20.062 20.194**** 20.012

IrrigationSurfacewater 20.129**** 20.125* 20.064****Groundwater 20.067 20.198**** 0.073Other 0.037 20.170** 0.172

DistanceTo nearest road (km) 20.014* 20.043** 20.001To nearest gully or ditch (km) 0.014*** 0.019*** 0.003To home (km) 0.041**** 0.095**** 0.009*

1999 Household Characteristics

Household head age 20.001 20.001 0.000Household head years of education 20.006* 20.014** 0.000Household population 20.013 20.008 20.001Household per capita income (RMB) 0.000** 0.000*** 0.000Nonagricultural share of household head per capita income 20.067** 20.051 20.028Household per capita land (ha) 20.166 20.198 20.012Household labor 0.010 20.004 0.004Nonagricultural share of laborb 0.075* 0.054 0.018

1999 Village Characteristics

Village population 0.000 0.000 0.000Village per capita income (RMB) 0.001**** 0.001**** 0.000**Village per capita agricultural land (ha) 0.865** 0.553 0.960***Share of village population in nonfarm wage work 1.178**** 0.778 1.349***Village leader years of education 0.015** 0.010 20.001Village secretary years of education 0.006 20.011 0.055***Village leader age 0.010* 0.017* 20.016***Village secretary age 0.007* 0.004 0.020***

Institutional Factors

Number of Years village has been implementing the SLCP 0.149**** 0.175** 20.008Share of village agricultural land with slope . 15u 0.294** 0.434 0.069Village leader worked previously at a forestry department (1 5 yes) 0.052 0.098 20.038Village secretary worked previously at a forestry department

(1 5 yes) 0.384**** 0.296 0.592***Number of villagers working in the county-level forestry department 0.010 0.102 20.189***Pseudo-R2 0.386 0.399 0.419% correctly predicted 84.6 82.0 89.3

*Significant at 10%; **significant at 5%; ***significant at 1%; ****significant at 0.1%. Statistical results are based on robust standarderrors clustered at the household level.

a Province and township dummies were used in the model but are not reported. Marginal effects for 0, 1 variables are for a discretechange from 0 to 1.

b Calculated using the number of head of household laborers working part-time or full-time off-farm.


TABLE 7

PLOT ENROLLMENT IN THE SLCP BY THE END OF 2002, BINOMIAL LOGIT MARGINAL EFFECTS

Explanatory Variablesa

Levels with Regional Interactionsb

All(n 5 2,004)

Autonomy(n 5 1,066)

No Autonomy(n 5 938)

1999 Plot Characteristics

Plot size (ha2) 0.337*** 0.704*** 0.079*1999 Income (RMB/ha2) 20.000 20.000* 20.000*Plot affected by a disaster in 1999 (1 5 yes) 0.479**** 0.460**** 0.341****Slope . 25u 0.246**** 0.267**** 0.105***Slope 15u–25u 0.095** 0.089 0.045High-quality landb 20.079* 20.059 20.023Low-quality landb 20.040 0.022 20.022Land rights

Responsibility or ration land 0.102** 0.149** 0.026Contract land 0.184* 0.241 0.077Other 20.057 20.196**** 0.003

IrrigationSurfacewater 20.134**** 20.135** 20.051****Groundwater 20.079 20.212**** 0.028Other 0.023 20.187*** 0.067

DistanceTo nearest road (km) 20.018** 20.049** 0.001To nearest gully or ditch (km)b 0.009 0.020 20.003To home (km) 0.041**** 0.094**** 0.006

1999 Household Characteristics

Household head age 20.001 20.001 0.000Household head years of education 20.006 20.012* 0.000Household population 20.015 20.016 0.003Household per capita income (RMB)b 0.000 0.000 0.000Nonagricultural share of household per capita income 20.075** 20.045 20.022Household per capita land (ha)b 20.328* 20.488 0.051Household labor 0.009 0.002 0.002Nonagricultural share of Laborc 0.078* 0.070 0.028

1999 Village Characteristics

Village population 0.000 0.000 0.000Village per capita income (RMB) 0.001**** 0.001*** 0.001**Village per capita agricultural land (ha) 0.945** 0.387 1.153***Share of village population in nonfarm wage work 1.033*** 0.577 1.570***Village leader years of education 0.015** 0.006 20.003Village secretary years of education 0.002 20.014 0.061***Village leader age 0.012** 0.020** 20.017***Village secretary age 0.006 0.002 0.023***

Institutional Factors

Number of years village has been implementing the SLCP 0.139**** 0.148* 20.003Share of village agricultural land with slope . 15u 0.307* 0.520 0.179Village leader worked previously at a forestry department

(1 5 yes) 0.046 0.021 20.017Village secretary worked previously at a forestry department

(1 5 yes) 0.314*** 0.164 0.902****Number of villagers working in the county-level forestry

department 0.013 0.124 20.209***Pseudo-R2 0.400 0.413 0.452% correctly predicted 85.6 82.5 90.4

*Significant at 10%; **significant at 5%; ***significant at 1%; ****significant at 0.1%. Statistical results are based on robust standarderrors clustered at the household level.

a Province and township dummies were used in the model but are not reported. Marginal effects for 0, 1 variables are for a discretechange from 0 to 1.

b To examine the affects of regional differences in natural and economic conditions, plot size, income/ha, land quality, distance tonearest gully or ditch, household per capita income, and household per capita land were interacted with province dummies. The marginaleffects of these interaction terms are not reported above.

c Calculated using the number of household laborers working part-time or full time off-farm.


autonomous and nonautonomous subsam-ples is still revealing and indicates that whenhouseholds have greater decision-makingpower in implementation they favor retire-ment of plots with lower opportunity costs.This can be seen in the indicator for low landquality (the household’s subjective evalua-tion), distance to home, the irrigation condi-tion variables, and the land rights variables,which are generally more often significantand larger in magnitude for the autonomoussubsample overall, and especially in compar-ison to the nonautonomous subsample.

In the nonautonomous subsample, forexample, land rights do not appear to playany significant role in enrollment targeting.From the perspective of the household,however, the degree to which rights over aplot are more stable and longer term hasdirect bearing on the plot’s position inhousehold input and investment choices.14

Similarly, the effect of a plot’s distance tohome, a clear indicator of its opportunitycost to the household, is highly significantand 10 times larger for the autonomous thanfor the nonautonomous households. In sum,these results are important; they suggest thatincreasing household autonomy in partici-pation choice could improve program cost-effectiveness by improving the likelihoodthat—pending eligibility—those plots ofleast cost for households will be chosen.

VI. PROGRAM IMPACT ON INCOME

Central to the realization of the SLCP’slong-term goals is whether it is adequatelyincentive-compatible for participants. Mostimmediately, program subsidies need to atminimum offset each participant’s opportu-nity cost of the enrolled land during thesubsidization period. Beyond that, theeconomic gain to farmers from the timberforests, orchards or pastures planted—andfrom other activities engaged in—as a resultof SLCP participation needs to be largeenough by the end of the subsidy period toensure that participants do not return

enrolled land back to cultivation. Postpro-gram land use decisions of participatingfarmers, in fact, have been one of the biggestconcerns in conservation set-aside programselsewhere (Cooper and Osborn 1998).

Though official reports and news ingovernment publications on the SLCPimplementation, progress, and socioeco-nomic impact are abundant (e.g., StateForestry Administration 2003; China GreenTimes 2003, 2004, 2005), rigorous analysesare rare. Not surprisingly, governmentreports all claim that the SLCP has had asignificant positive impact on programareas. However, the validity of these officialstatements needs to be examined furtherdue in part to questions about the quality ofthe survey data used, since it has beengathered via the government reportingsystem and thus may contain substantialbias in favor of program implementationagencies. The only rigorous analysis ofprogram impact to date is by Uchida et al.(2007), who used propensity scoring match-ing to evaluate the social and economicimpacts of the program. Overall, they foundevidence of a significant negative impact oncropping income, a significant positiveimpact on husbandry income and invento-ries, and a significant positive impact onproductive and housing assets. At the sametime, however, impact on total householdper capita income is estimated to be smalland statistically insignificant.

Table 8 presents the 1999 and 2002components of total income for participantand nonparticipant households, by prov-ince.15 These numbers suggest that the

14 For more evidence, please refer to Liu et al. (1998),Carter and Yao (1998), and Rozelle et al. (2002).

15 Cropping income consists of total crop productionvalued at average village market price, net of materialsand hired labor costs. Husbandry income includes bothsales income and own consumption, valued at marketprices. Off-farm income includes all nonagriculturalproduction activities, comprised mainly of sidelineactivities and wage labor income. Income from sidelineactivities is net of production costs and other business-related expenditures. Wage income includes both cashand in-kind income, valued at market prices. Otherincome consists of aquaculture, rental and interestincome, gifts, pension income, and government subsidiesand transfer payments. The SLCP subsidy is calculated asthe subsidy received by the household for 2002.


SLCP has indeed induced a restructuring ofagricultural production, whereby partici-pants have shifted relatively more of theirinputs out of cropping and into husbandry.In Shaanxi Province, growth rates forcropping income were 35% for nonpartici-pants compared with only 12% for partic-ipants (including subsidies received). InGansu, these were 226% and 232%,respectively, and in Sichuan cropping in-come declined by 30% for both groups.Conversely, growth rates for husbandrywere higher for participants than fornonparticipants. In Shaanxi, averagehousehold per capita husbandry incomefor participants increased more than 10-fold, compared to only 175% for nonpar-ticipants. In Gansu, participants’ husband-

ry income grew by 1,744%, compared withonly 586% for nonparticipants, and inSichuan these numbers are 845% and514%, respectively. Differences betweenparticipants and nonparticipants in changeof total income are less systematic acrossregions. In Shaanxi, total income (includingsubsidies received) increased by 41% and42% for participants and nonparticipants,respectively. For Gansu these numbers are2.3% and 12%, respectively, and for Sich-uan they are 26% and 17%, respectively.

Since such numbers could be the result offactors unrelated to SLCP implementation,we use a first-differences model explainingchange in household per capita net incomebetween 2002 and 1999 to more rigorouslyestimate program impact on income. A

TABLE 8

PER CAPITA NET INCOME OF PARTICIPANT AND NONPARTICIPANT HOUSEHOLDS, 1999 AND 2002

Income Componenta

Nonparticipant Households Participant Households

1999 2002 1999 2002

Mean St. Dev. Mean St. Dev. Mean St. Dev. Mean St. Dev.

Shaanxi

Total without subsidy 940 777 1,335 930 986 1,077 1,325 1,874Total with subsidy received — — — 1,394 1,877Cropping without subsidy 465 521 626 429 420 672 401 622Cropping with subsidy received — — — 470 628Husbandry 6 23 17 63 18 78 208 916Off-farm 388 623 590 947 401 554 525 680Other 82 233 101 234 147 686 191 826

Gansu

Total without subsidy 1,803 1,681 2,021 1,741 1,287 980 1,287 942Total with subsidy received — — — 1,317 942Cropping without subsidy 484 350 360 246 589 523 370 320Cropping with subsidy received — — — 399 345Husbandry 17 53 119 220 6 30 113 222Off-farm 1,192 1,570 1,346 1,624 633 679 681 647Other 110 515 196 541 59 204 124 393

Sichuan

Total without subsidy 1,419 1,425 1,654 1,271 1,635 1,195 1,961 1,524Total with subsidy received — — — 2,067 1,514Cropping without subsidy 721 938 506 633 829 931 472 590Cropping with subsidy received — — — 577 583Husbandry 33 42 202 200 49 75 459 1,187Off-Farm 543 953 714 987 674 897 869 971Other 122 295 232 476 83 251 161 375

Source: Data from Table 6, Xu et al. 2004.a All units are in 1999 RMB, adjusted using the Rural Consumer Price Index.


simple regression specification for explain-ing change in income is

Dyki ~azd(progi)zdL(pyrsi)zDxH

i bH

zDxVi bVzdiczDmi,

where Dyik denotes the change in household

i’s per capita net income component kbetween 1999 and 2002, progi 5 1 if house-

hold i is a participant, DxiH and Dxi

V arevectors of household and village-level char-acteristics, respectively, di is a vector ofprovincial- and county-specific time-trenddummies, and Dmi is the difference inidiosyncratic disturbances across periods.Since households entered the program atdifferent times between 1999 and 2002, wealso include in one specification the numberof years household i has been in the SLCPbefore the end of 2002 (pyrsi) to control forlagged program impacts not picked up inthe first difference.

Of interest for program evaluation is d,generally referred to as the difference-in-differences estimator. This captures thedifference, controlling for household andvillage-level factors, in average incomechange between participants and nonpar-ticipants that can be attributed to programparticipation (Wooldridge 2002). A centralconcern in the literature on programevaluation is the impact of selection biason estimates of program impact. If selectionbias is an issue, then the outcomes ofnonparticipants cannot be used to estimatethe counterfactual outcomes for partici-pants were they not to have participated.In the case of our survey, we believe theeffects of selection bias can be largelycontrolled for due, ironically, to the pre-dominantly topdown approach towardimplementation observed in the survey.This gives us a subsample of householdsthat have not self-selected to be partici-pants or nonparticipants, allowing forempirical examination of selection bias inthe sample.

To do this, we estimate [3] above in atreatment effects model framework whereinprogram participation is assumed to be theobserved outcome of a latent process that is

a function of household characteristics andcounty and provincial intercepts, so that

progi~1 if p1w0

0 otherwise

�

and

p1~xicxzzic

zzdilzei, [4]

where the vector xi contains householdcharacteristics shared with the incomedifference model [3], zi is a vector ofhousehold characteristics serving as instru-ments for identification of p* and thereforenot shared with [3], and di is a vector ofprovincial- and county-specific intercepts.Since selection bias occurs when participa-tion and outcomes are the result of com-mon, unobserved variables, the error termsof the income difference and selectionmodels, Dm and e, are assumed to bedistributed bivariate normal with zero meanand covariance matrix

sDm r

r 1

� �:

For our purposes, this characterization ispreferred to a two-stage IV regression, sinceit allows for direct estimation via maximumlikelihood of r—the correlation coefficientbetween e and Dm. The statistical signifi-cance of r as well as its estimated sign andmagnitude provide tests of the presence,direction, and importance of selection biasin the sample, all of which are of interest forevaluating program implementation, espe-cially given the significant share of samplehouseholds that reported lack of autonomyin participation choice.

To control for selection on observables,1999 household characteristics sharedacross the selection and outcome equationsinclude per capita income, per capitanumber of plots affected by disaster, yearsof education of the household head, percapita arable land, and number of house-hold laborers working in (1) agriculture, (2)‘‘outside area’’ wage labor, and (3) self-


employed in nonfarm work.16 The 1999–2002 changes in household population andlabor are also included in the outcomeequation [3]. To control for exogenousvillage-level factors impacting outcomesacross households, 1999–2002 change invillage population, change in the share ofvillage households with telephones and withtap water, change in the number of long-distance buses that run through the village,and the 1999 number of village enterpriseswere also included in equation [3].17

Though not critical for identification inthis specification, the use of good instru-mental variables (zi) helps to ensure that themodel is well specified (Maddala 1983;Wooldridge 2002). Household autonomyin the participation decision was identifiedas an excellent candidate. It is not onlysignificantly correlated with program par-ticipation in the sample, but uncorrelatedwith any of the income change variables.18

Furthermore, in the logit analysis of house-hold autonomy status discussed above,

program participation was found to havea large and highly significant positivemarginal effect as an explanatory variable,while household per capita income wasinsignificant. To be able to compare modelestimates between the full sample and thesubsample of nonautonomous households,two additional variables were also selectedas candidates for instruments. Based onobservations made in the field, we suspectthat (1) whether the household had amember in the Communist party in 2002and (2) whether the household had amember in village government in 2002 arealso related to the participation decisionbut not to income change.19 Note that whenestimating the model on the subsample ofnonautonomous households, selection biascan still come into play if program officialshave selected certain types of individualsbased on unobserved characteristics thatare correlated with outcomes. As such, wethink that (1) and (2) could also be goodinstruments in this type of selection process.

Tables 9 and 10 present model estimates.The first pair of models uses only variables(1) and (2) and their interaction as instru-ments and was estimated both on the fullsample and the nonautonomous subsamplefor comparison. The results indeed provideevidence that selection bias exists in the fullsample but not in the subsample. Inparticular, for cropping income the corre-

16 We compared mean 1999 household characteristicsbetween participants and nonparticipants for both the fullsample and the subsample of nonautonomous house-holds. For the full sample, household labor populationwas 0.227 larger for participants, significant at 10%,household total arable land was 0.242 ha larger and percapita land was 0.042 ha larger for participants,significant at 0.01% and 10%, respectively, and per capitanumber of plots affected by disaster was 0.317 larger forparticipants, significant at 0.01%. For the nonautono-mous subsample, number of workers in ‘‘outside area’’wage labor was 0.205 larger for participants, significant at5%, total arable land was 0.254 larger for participants,significant at 5%, and per capita number of plots affectedby disaster was 0.265 larger for participants, significant at0.01%.

17 We argue that 2002 party membership is a validproxy for 1999 membership (and the underlying unob-served characteristics it captures); entering the communistparty in rural China involves a long process of letter-writing and application. We make similar arguments forthe 2002 village official variable, since often villagegovernment posts rotate regularly and have both costsand benefits associated with them. Thus, it is also likely tobe associated with probability of participation, but notwith outcome.

18 Its pairwise correlation with participation is 0.3043,significant at 0.01%. In the probit model explainingautonomy status, mentioned above, program participa-tion was estimated to have a large and highly statisticallysignificant marginal effect, while household per capitaincome was not statistically significant.

19 (1) and (2) are included as proxies for thesecharacteristics in 1999, since the survey collected dataon these only for 2002. Note that both party membershipand village government positions generally are theoutcomes of relatively long term processes, and so weare comfortable using the 2002 levels as proxies forpreprogram status. In rural China, communist partymembership does not provide any clear financial benefitsto members or their families, who come from a wide rangeof backgrounds and income strata. Similarly, villagegovernment membership encompasses a variety ofpositions with a wide variation in actual power andresponsibility and also can entail significant opportunitycosts in areas where village government finances arestrained, suggesting no clear relationship with householdper capita income. In the sample, this variable is in factuncorrelated with any income component changes.However, neither (1) nor (2) is significantly correlatedwith program participation in the sample, while (1) isnegatively correlated with change in husbandry income,significant at 0.5%.


lation between the error terms is large,negative, and highly statistically significantin both specifications for the full sample,while insignificant for the nonautonomoussubsample. This makes sense, since itsuggests that households with lower oppor-tunity cost for retiring their land—in termsof their foregone cropping income—aremore likely to participate when given thefreedom to choose. It thus also suggests thatensuring that households have autonomy inparticipation choice could improve pro-gram cost-effectiveness.

In the selection equations for the fullsample, the variable capturing whether thehousehold has a member in village govern-ment is positive, and its interaction withhousehold having a communist party mem-ber is negative—both significant at 10%—inthe majority of specifications. When esti-mated on the subsample, the variablecapturing whether the household has acommunist party member is negative and

significant at 5% in the selection equationfor the majority of specifications. Thesecond pair of models for the full sampleadds autonomy status to the list of instru-ments. Autonomy status is indeed found tobe positive and significant at 0.01% in theselection equation for all specifications.Similar to the first pair of models, theseresults also provide evidence of selectionbias in the sample for cropping income,since the correlation between errors isstatistically significant in these models,though only weakly so in the model thatdoes not include an impact lag variable.

Overall, model estimates for the fullsample provide evidence that the programis inducing farmers to intensify agriculturalproduction on their remaining plots, whilealso shifting into husbandry. Estimatedimpacts on cropping income are positiveand significant at 0.01% in all specificationsrun on the full sample, and range fromCNY 194 to CNY 263 per capita for cro-

TABLE 9

PROGRAM IMPACT ON PARTICIPANT INCOME, TREATMENT EFFECTS MODEL: ALL HOUSEHOLDS (N 5 345)

Income Component

Without Autonomy Status as an Instrument

Impact Impact with Lag

Estimate ra Estimate r

Program Impact (d)

Cropping before subsidy 263.07*** (45.62) 20.851*** (0.043) 341.83*** (45.62) 20.827*** (0.051)Cropping with subsidy 303.17*** (48.73) 20.819*** (0.05) 371.58*** (48.73) 20.808*** (0.054)Husbandry 311.30* (185.86) 20.0806999 (0.139) 389.22* (185.86) 20.081 (0.139)Total agricultural before subsidy 272.67 (199.36) 20.149 (0.139) 477.19** (199.36) 20.148 (0.138)Total agricultural with subsidy 318.57 (203.52) 20.144 (0.141) 491.47** (203.52) 20.143 (0.141)Off-farm 230.56 (153.91) 20.013 (0.204) 243.68 (153.91) 20.013 (0.205)Noncropping 296.83 (238.79) 20.088 (0.156) 361.1 (238.79) 20.087 (0.156)Other 26.81 (87.27) 20.039 (0.195) 37.22 (87.27) 20.040 (0.193)Total before subsidy 280.88 (273.08) 20.167 (0.158) 510.48 (273.08) 20.163 (0.156)Total with subsidy 325.75 (277.86) 20.162 (0.16) 524.43 (277.86) 20.159 (0.159)

Program Lagged Impact (dL)

Cropping before subsidy — 239.49** (16.19) —Cropping with subsidy — 230.87* (16.43) —Husbandry — 230.76 (49.08) —Total agricultural before subsidy — 281.68 (52.25) —Total agriculture with subsidy — 268.95 (52.75) —Off-farm — 5.12 (30.21) —Noncropping — 225.56 (57.83) —Other — 217.23 (17.82) —Total before subsidy — 293.39 (64.97) —Total with subsidy — 280.69 (65.42) —


pping income without subsidy.20 Averageimpacts on husbandry range from CNY 299to CNY 390. Model results for the fullsample also suggest that the program couldbe having a positive impact on total income,though the estimated impact is statisticallysignificant only for the model with programlag variables using household autonomystatus as an instrument. Estimates ofaverage program impact on total incomewithout subsidies range from CNY 238 toCNY 335 in the full sample.

As a comparison with the results for thefull sample, model estimates from thenonautonomous subsample provide littleevidence of positive impact on participanttotal and cropping income, and only weak

evidence of positive impact on husbandryincome. These are presented in Table 10.Estimates of average impact on croppingincome without subsidy for the subsamplerange from CNY 54 to CNY 284 percapita. Estimated average impacts on totalincome without subsidy range from CNY27 to CNY 227. Although estimates forthe full sample attempt to correct forselection bias, they do not allow for thepossibility that selection bias could alsocome into play through systematic differ-ences in the parameters for householdcharacteristics. If that were the case, andtaking the tests for selection bias at facevalue, it would suggest that the nonauton-omous subsample, as a sample of individ-uals that are ‘‘randomly’’ assigned treat-ment, might provide better estimates ofprogram impact within this modelingframework. Though further exploration ofthis is beyond the scope of this paper, these

Including Autonomy Status as an Instrument


Estimate r Estimate r

233.67*** (49.88) 20.830* (0.054) 321.65*** (61.58) 20.802*** (0.065)266.74*** (54.33) 20.787* (0.066) 345.93*** (66.75) 20.777*** (0.07)389.63** (160.71) 20.159 (0.118) 472.52** (204.24) 20.161 (0.117)347.95** (174.96) 20.224 (0.12) 561.36** (219.09) 20.227 (0.119)397.68** (177.81) 20.221 (0.121) 579.12*** (222.54) 20.224 (0.12)

217.55 (140.54) 20.034 (0.194) 230.81 (162.68) 20.033 (0.195)397.22* (204.64) 20.173 (0.132) 466.78* (252.36) 20.173 (0.132)

261.7 (89.79) 0.101 (0.212) 215.45 (100.04) 0.091 (0.206)335.21 (242.3) 20.215 (0.141) 578.19** (291.11) 20.217 (0.139)384.39 (245.66) 20.213 (0.143) 595.66** (295.11) 20.215 (0.141)

— 243.71*** (16.17) —— 234.96** (16.48) —— 232.16 (49.07) —— 283.6 (52.21) —— 270.9 (52.71) —— 4.93 (30.24) —— 227.1 (57.81) —— 216.84 (17.83) —— 295.6 (64.91) —— 282.91 (65.36) —

Note: Standard errors are in parentheses. Selection and impact equations were estimated simultaneously via maximum likelihood.a Significant levels for the correlation coefficient are for chi-squared tests.* Significant at 10%; **significant at 5%; ***significant at 1%.

TABLE 9

(Extended)

20 The sample average number of years in the programof 2.92 and 2.57 for the full sample and nonautonomoussubsample, respectively, were used for calculations usingthe models with lagged impact variables.


results suggest at minimum that lack ofautonomy in participation choice could bedampening program impacts.

In addition to these findings, it is alsointeresting to note that for both the fullsample and subsample, estimated averageprogram impact on change in off-farmincome is small and not statistically oreconomically significant in any specifica-tion, which is contrary to governmentclaims regarding program impacts. In fact,farmers in the sample reported they had tospend considerable time planting and caringfor trees on enrolled land so as to guaranteethat they received subsidies. Combined withthe model results, this suggests that ratherthan freeing up labor, program participa-tion could be tying up labor to manageretired land, at least over the short term.Furthermore, though production risk islikely a significant concern for householdsin the sample, participation in the SLCP

might simply be replacing one type of riskfor another.

Specifically, program stipulations requir-ing a minimal survival rate of trees or grassplanted on retired land in order forsubsidies to be delivered, combined withthe uncertainties and lag time involved withimplementation and inspection by villageand higher-level authorities, suggest thatparticipation itself entails a form of non-negligible income risk. Our survey tells usthat in only 5 of the 18 sample townshipswere survival rates consistently above thegovernment standard (70% in Gansu andShaanxi, 85% in Sichuan). In Li county inSichuan, survival rates were below thestandard in all but one inspection in onetownship, and this had been declining inrecent years. According to farmers sur-veyed, survival rates were even below 40%in many places, and significant replantinghas had to be done. Pervasive replanting

TABLE 10

PROGRAM IMPACT ON PARTICIPANT INCOME, TREATMENT EFFECTS MODEL: HOUSEHOLDS WITHOUT AUTONOMY

(N 5 161)

Income Component


Estimate ra Estimate r

Program Impact (d)

Cropping before subsidy 54.54 (252.76) 20.559 (0.621) 52.55 (214.78) 20.191 (0.679)Cropping with subsidy 141.22 (141.63) 20.622 (0.299) 199.54 (211.59) 20.512 (0.536)Husbandry 176.97* (106.8) 20.098 (0.18) 194.43 (136.61) 20.098 (0.18)Total agricultural before subsidy 78.42 (141.63) 20.138 (0.208) 220.35 (171.86) 20.136 (0.206)Total agriculture with subsidy 131.3 (146.18) 20.141 (0.216) 257.67 (175.82) 20.137 (0.214)Off-farm 245.86 (189.18) 20.025 (0.236) 246.54 (226.64) 20.025 (0.236)Noncropping 135.61 (219.67) 20.081 (0.219) 152.35 (267.29) 20.081 (0.218)Other 294.18 (251.23) 0.118 (0.573) 2.53 (215.58) 0.093 (0.454)Total before subsidy 27.29 (303.84) 20.101 (0.269) 217.62 (344.28) 20.097 (0.263)Total with subsidy 45.79 (316.07) 20.102 (0.281) 254.92 (354.49) 20.097 (0.274)

Program Lagged Impact (dL)

Cropping before subsidy — 252.47** (21.71) —Cropping with subsidy — 244.31* (23.24) —Husbandry — 27.25 (35.37) —Total agricultural before subsidy — 259.78 (41.84) —Total agricultural with subsidy — 253.51 (42.09) —Off-farm — 0.28 (51.62) —Noncropping — 26.96 (63.31) —Other — 235.77 (30.71) —Total before subsidy — 295.34 (73.7) —Total with subsidy — 289.1 (74.06) —

Note: Standard errors are in parentheses. Selection and impact equations were estimated simultaneously via maximum likelihood.a Significant levels for the correlation coefficient are for chi-squared tests.*Significant at 10%; **significant at 5%; ***significant at 1%.


has led to shortfalls in government seedlingsubsidies of around CNY 50 per muaccording to our survey. Moreover, inmany regions even replanting cannot guar-antee survival rates due to lack of water,and in many cases is simply done to passgovernment inspections so as to obtainprogram subsidies (Du and Xu 2003).

Households, as the core agents engagedby the government to implement the SLCP,are in the best position to know whetherprogram participation will benefit them.Lack of choice could thus be increasing thelikelihood that households not well posi-tioned to benefit from the SLCP, or whoseincome prospects would suffer as the resultof participation, are being selected into theprogram.

VII. CONCLUSION

Though the representativeness of thedata set used for this analysis is open todebate in light of the huge diversity of localinstitutional, economic, and ecological con-ditions encompassed by the program, it isnonetheless the best available to evaluatethe program’s implementation over a rela-tively long period. Overall, though ourresults provide some support for the gov-ernment’s positive view of the program,they are not strong enough to justify the fastexpansion of the SLCP after the pilotperiod. Our targeting analysis, while indi-cating that land enrollment choice hasindeed been strongly influenced by programgoals, also points to significant mistargetingof fertile flatland for retirement. Thissuggests some inconsistencies with theSLCP’s emphasis on retirement of highlysloped land so as to reduce soil erosion. Theincome impact analysis results, thoughproviding evidence of positive programimpacts on cropping, husbandry, and totalincome, fall short of the claims made by thegovernment about the huge success of thepilot phase, and it is such claims that haveserved to justify the breakneck speed ofexpansion seen during full implementation.

Most troubling is the finding that, thoughon paper a payment for environmental

services scheme, the SLCP appears to bein practice just another in a long line oftopdown, campaign-style programs imple-mented by China’s central government. Theimportance of farm households as the keylong-term actors in implementation makesparticipant willingness and choice necessaryconditions to program success. Both theplot targeting and income impact analysisresults, in fact, suggest that significant gainsin the cost-effectiveness of the programcould be achieved by ensuring that house-holds have autonomy of participationchoice. In particular, the results indicatethat when given the choice, households withlower opportunity costs are more likely toselect into the program, and plots withlower opportunity costs for participanthouseholds are more likely to be enrolled.

These results are important, since inabsence of significant and sustainable gainsin the development of sources of noncrop-ping and off-farm income, participants whooriginally did not wish to participate, whoare not being adequately compensated fortheir opportunity costs of participation,may easily return land to cultivation uponthe end of the subsidy period. And resultsfrom the survey indicate that this is not asmall share of participants.21 Survey evi-dence, in fact, suggests that farmers couldface significant added income risk due toprogram participation. For participants inthe sample, per capita cropland decreasedby 43% on average overall, and by 57% forhouseholds in Shaanxi Province in particu-lar (from 0.24 ha in 1999 to only 0.09 ha in2002). Considering the adverse climateconditions and frequent natural disastersin this region, reducing both household percapita land area and the number of plotsthat households have distributed acrosslocal microclimates could significantly im-pact households’ ability to hedge against

21 Bennett (2008) calculates from the survey data thatonly about 20.8% of participant households (representing23.7% of retired area) indicated that they would allowprogram-planted trees to reach maturity. Another 36.7%of participant households (representing 30.5% of retiredarea) said that they could maintain their livelihood withrevenue from current activities.


significant production risk in absence ofalternative sources of income.

The stability of such alternative sourcesof income, furthermore, is questionable.First, the future value and shorter-termincome-generating capacity from timberforests (i.e., ‘‘ecological forests’’) and or-chards (i.e., ‘‘economic forests’’) plantedunder the SLCP does not look promising.For timber forests, this is due to low timberforest survival and growth rates in manyregions as a result of lack of rainfall andunsuitable conditions for timber trees (es-pecially in the arid northwest provinces ofGansu and Shannxi), the uncertainties inthe future of China’s forest sector reforms,and the potential oversupply of timber dueto large-scale plantations in the south.Regarding economic forests, the fast ex-pansion in the SLCP has led to manydifferent regions in China planting similarorchard crops, raising concerns aboutfuture oversupply and dampened economicvalue of these forests.

Overall, the SLCP, a program imple-mented in a period of grain surplus, wasexpanded too fast as a result of the interplaybetween the central government’s underly-ing goal of reducing State Grain Bureaugrain stocks, and local government interestin increasing subsidy inflows. Lack ofsuccessful experiences in the past andexcessively fast program expansion havecreated significant fiscal risk for govern-ments, income risks for participants, andecological risks for program-covered are-as.22 In fact, due to such risks, thegovernment significantly slowed the rate

of expansion in 2005 and is currentlydiscussing how to scale back the program.

Though the government’s growing lar-gesse toward environmental initiatives isencouraging, large-scale campaign-styleprograms may not be the way to reverseadverse environmental outcomes stemmingfrom a complex combination of factors.Improvements in program design that allowfor greater local initiative, flexibility, andthe use of more market-based instrumentscould help to improve program cost-effec-tiveness by better matching participantopportunity cost with program subsidies,thus increasing the possibility of programsuccess. At the same time, ironically, thelack of household autonomy seen in oursample suggests that the lessons learned andapproach toward program evaluationadopted here could also be applied to other,more compliance based conservation pro-grams internationally, especially as worldpolicy moves toward greater emphasis onenvironmental protection via public regula-tory schemes. That said, voluntary partic-ipation remains an important principle toinsure the cost-effectiveness and eventualsuccess of the SLCP.

Finally, the results of the analysis—whichfind no statistically or economically signif-icant positive program impacts on off-farmincome growth—argue against the validityof a key underlying assumption of theSLCP: retiring land from cultivation wouldautomatically lead to desirable rural eco-nomic restructuring that in turn wouldreduce the risk of returning retired land tocultivation after the subsidy period. In thefast expansion of the program, neither thecentral nor the local governments havedesigned and installed sufficient coordinat-ed policies to induce effective incomerestructuring within and permanent migra-tion out of the program target regions. Infact, effective government support for thenoncropping agricultural sectors, balancedwith policies to promote nonagriculturalsectors and permanent migration out ofecologically fragile regions, may well bemore effective in relieving pressure onsloping, marginal cropland than the cam-

22 Most of the SLCP-covered area in northwesternChina is arid or semi-arid land suitable only for grass orshrub plantation. Timber tree plantation would indeedhamper the water conservation function of soil or evenlead to further land desertification because timber treesneed much more water than grass or shrubs. The lowsurvival rates of timber trees has been very apparent inChina’s other ecological programs implemented in earlyperiods, such as the Northeast, North, and NorthwestChina Green Belt Program. One example is MingqinCounty, Gansu Province, where the area of governmentafforestation was as high as 87,000 ha, but only 20,000 hasurvived (Jiang 2003).


paign-style government program we havewitnessed. Only by positioning the SLCPwithin an array of coordinated policies canthe dual goals of environmental ameliora-tion and poverty alleviation be realized.

References

Bennett, Michael T. 2008. ‘‘China’s Sloping LandConversion Program: Institutional Innovationor Business as Usual?’’ Ecological Economics 65(4): 699–711.

Carter, Michael, and Yang Yao. 1998. ‘‘PropertyRights, Rental Markets, and Land in China.’’Working Paper. Madison: Department ofAgricultural and Applied Economics, Univer-sity of Wisconsin–Madison.

China Green Times. 2003. ‘‘The Sloping LandConversion Program Provides More ThanCNY 60 Million in Benefits to Farmers.’’ [InChinese.] Beijing: China Green Times Publish-ing. Available at www.greentimes.com (ac-cessed October 16, 2003).

———. 2004. ‘‘The Six Big Forestry Programs:The Heroic Feat of Recovering the BeautifulLandscape.’’ [In Chinese.] Beijing: China GreenTimes Publishing. Available at www.greentimes.com (accessed September 10, 2004).

———. 2005. ‘‘The Sloping Land ConversionProgram Is Key for Building the Ecosystem.’’[In Chinese.] Beijing: China Green TimesPublishing. Available at www.greentimes.com(accessed April 22, 2005).

Cooper, Joseph C., and Tim Osborn. 1998. ‘‘TheEffect of Rental Rates on the Extension ofConservation Reserve Program Contracts.’’American Journal of Agricultural Economics 80(1): 184–94.

Du, Jishan, and Jintao Xu. 2003. ‘‘A Socio-Economic Evaluation of the Sloped LandConversion Program.’’ [In Chinese.] Workingpaper. Center for Chinese Agricultural Policy,IGSNRR, CAS, Beijing.

Economic Daily. 2007. ‘‘State Forest Administra-tion Issues Report on the Overall Development ofthe Sloping Land Conversion Program.’’ Febru-ary 4, 2007.

Heimlich, Ralph. 2003. ‘‘Land Retirement.’’ InAgricultural Resources and Environmental Indi-cators, ed. Ralph Heimlich and William Ander-son, Chap. 6.2. Washington, DC: EconomicResearch Service, U.S. Department of Agricul-ture.

Huang, Jikun. 2000. ‘‘Land Degradation in China:Erosion and Salinity Component.’’ CCAP

Working Paper WP-00-E17. Beijing: Centerfor Chinese Agricultural Policy.

Hyde, William, Brian Belcher, and Jintao Xu, eds.2003. China’s Forests: Global Lessons fromMarket Reforms. Washington, DC: Resourcesfor the Future.

Jiang, Gaoming. 2003. ‘‘More Emphasis onPlantation Than Conservation: Ecological Pro-tection Is Not Equal to Tree Plantation.’’ [InChinese.] China Youth Daily, Sept. 24.

Kung, James Kai-Sing. 1995. ‘‘Equal Entitlementversus Tenure Security under a Regime ofCollective Property Rights: Peasants’ Prefer-ence for Institutions in Post-Reform ChineseAgriculture.’’ Journal of Comparative Econom-ics 21 (1): 82–111.

Liu, Shouying, Michael Carter, and Yang Yao.1998. ‘‘Dimensions and Diversity of PropertyRights in Rural China: Dilemmas on the Roadto Further Reform.’’ World Development 26(10): 1789–1806.

Lu, Feng. 1998. ‘‘Grain versus Food: A HiddenIssue in China’s Food Policy Debate.’’ WorldDevelopment 26 (9): 1641–52.

———. 1999. ‘‘Three Grain Surpluses: Evolutionof China’s Grain Price and Marketing Policies(1978–1999).’’ Presented at the InternationalAgricultural Trade Research Consortium, Chi-na’s Agricultural Trade and Policy: Issues,Analysis, and Global Issues. San Francisco,California, June 25–26.

Maddala, G. S. 1983. Limited-Dependent andQualitative Variables in Econometrics. NewYork: Cambridge University Press.

Pagiola, Stafano, Natasha Landell-Mills, andJoshua Bishop. 2002. ‘‘Making Market-BasedMechanisms Work for Forests and People.’’ InSelling Forest Environmental Services: Market-Based Mechanisms for Conservation and Devel-opment, ed. Stafano Pagiola, Joshua Bishop,and Natasha Landell-Mills. London: Earth-scan.

Park, Albert, Sangui Wang, and Guobao Wu.2002. ‘‘Regional Poverty Targeting in China.’’Journal of Public Economics 86 (1): 123–53.

Rozelle, Scott, Loren Brandt, Li Guo, and JikunHuang. 2002. ‘‘Land Rights in China: Fact,Fiction, and Issues.’’ China Journal 47:67–97.

State Forestry Administration (SFA). 2003. Planfor Sloping Land Conversion Program (2001–2010). [In Chinese.]. Beijing: State ForestryAdministration.

Uchida, Emi, Jintao Xu, and Scott Rozelle. 2005.‘‘Grain for Green: Cost-Effectiveness andSustainability of China’s Conservation Set-Aside Program.’’ Land Economics 81 (2):247–64.


Uchida, Emi, Jintao Xu, Zhigang Xu, and ScottRozelle. 2007. ‘‘Are the Poor Benefiting fromChina’s Land Conservation Program?’’ Envi-ronment and Development Economics 12 (4):593–620.

Wooldridge, Jeffrey M. 2002. Econometric Analysisof Cross-Section and Panel Data. Cambridge,MA: MIT Press.

World Bank. 2001a. China: Air, Land, and Water:Environmental Priorities for a New Millenni-um. Washington, DC: World Bank.

———. 2001b. China: Overcoming Rural Poverty.Washington, DC: World Bank.

World Wildlife Fund. 2003. Report SuggestsChina’s ‘‘Grain-to-Green’’ Plan is Fundamentalto Managing Water and Soil Erosion. Availableat www.wwfchina.org/english/loca.php?loca5

159 (accessed March 31, 2003).

Xu, Jintao, and Yiying Cao. 2001. ‘‘The Socioeco-nomic Impacts and Sustainability of theSLCP.’’ In Implementing the Natural ForestProtection Program and the Sloping LandConversion Program: Lessons and Policy Rec-ommendations, ed. Jintao Xu, Eugenia Katsi-gris, and Thomas A. White, CCICED-WCFGTF (China Council for InternationalCooperation on Environment and Develop-ment, Western China Forests and GrasslandTask Force). Beijing: China Forestry PublishingHouse.

Xu, Jintao, Eugenia Katsigris, and Thomas A.White, eds. 2001. Implementing the NaturalForest Protection Program and the Sloping LandConversion Program: Lessons and Policy Impli-cations. CCICED-WCFGTF (China Councilfor International Cooperation on Environmentand Development, Western China Forests andGrassland Task Force). Beijing: China ForestryPublishing House.

Xu, Zhigang, Michael Bennett, Ran Tao, andJintao Xu. 2004. ‘‘China’s Sloping Land Con-version Program Four Years On: CurrentSituation, Pending Issues.’’ International For-estry Review 6 (3–4): 317–27.

ZGTJNJ (China Statistical Yearbook). 2001.Beijing: China Statistical Press.

Zhong, Funing. 2001. ‘‘The Political Economy ofthe Chinese Grain Marketing System.’’ Techni-cal Report ACIAR Project 9721, WorkingPapers, APSEM. Canberra: Australian Nation-al University.

Zuo, Ting. 2001. ‘‘Implementation of the SLCP.’’In Implementing the Natural Forest ProtectionProgram and the Sloping Land ConversionProgram: Lessons and Policy Recommendations,ed. Jintao Xu, Eugenia Katsigris, and ThomasA. White, CCICED-WCFGTF (China Councilfor International Cooperation on Environmentand Development, Western China Forests andGrassland Task Force), Part 2, Chap. 3. Beijing:China Forestry Publishing House.


China’s Sloping Land Conversion Program: Does Expansion ... · tious land...

Documents

Transcript of China’s Sloping Land Conversion Program: Does Expansion ... · tious land...