Migration between Ghana’s Rural and Urban Areas

8/13/2019 Migration between Ghanas Rural and Urban Areas

1/42

Migration between Ghanas Rural and Urban Areas:

The Impact on Migrants Welfare

Louis Boakye-Yiadom*

Andrew McKay**

September 2006

*

Corresponding author. E-mail: [email protected]. PhD student, Department of Economics andInternational Development, University of Bath, UK.**

Professor, Department of Economics and International Development, University of Bath, UK.
mailto:[email protected]:[email protected]


2/42

2

Abstract

This paper examines the impact on migrants welfare of migration between rural

and urban sectors, using data from Ghana. Employing a consumption measure of

welfare and a model that corrects for selectivity bias, the analysis also captures factors

influencing migration decisions. Our findings highlight the importance of anticipated

welfare gains and personal attributes in migration decisions. We also find support for

the positive selectivity of urban-to-rural migrants. In addition, estimates of migration

gains suggest that although some migrants incur welfare losses, migration increases

on average the welfare of migrants, but would reduce the mean welfare of non-

migrants if they were to migrate. Finally, the average welfare increment derived by

rural-to-urban migrants is proportionately much higher than what accrues to their

urban-to-rural counterparts.

JEL Classification:O15; O18; I31; R23

Keywords:Migration; rural-to-urban; urban-to-rural; welfare; selectivity bias


3/42

3

1. Introduction

Over the years, the relevance of migration, the rationale for migrating, and the policy

response to migration patterns and magnitudes have dominated academic and policy

discussions. Ever since the seminal work of Ravenstein (1885), numerous studies

have explored various aspects of this pervasive phenomenon. The diversity of

disciplinary perspectives e.g., demography, economics, and geography of the

studies attests to migrations intriguing and complex nature. Many studies address

issues relating to the rationale for migrating (see Sjaastad, 1962; Todaro, 1969; and

Lucas and Stark, 1985), migration patterns (see Ravenstein, 1885; and Lee, 1966), or

the determinants of migration (for example, Caldwell, 1968; and Hay, 1980). Others,

however, examine the welfare impacts of these population movements (for example,

Falaris, 1987; and Litchfield and Waddington, 2003). Migration studies can,

nevertheless, be broadly categorised into two, namely, those focusing on internal

migration, and the set of studies examining migration across national borders, with the

latter body of studies somewhat dominating in volume.

Clearly, international migration has immense significance for many countries. For

typical developing countries especially those in sub-Saharan Africa however,

internal migration is of equal, if not, greater importance. Given that the rural-urban

categorization is the major spatial grouping in sub-Saharan African countries, and that

urbanisation in these countries is on the rise, it is hardly surprising that rural-to-urban

migration has dominated the countries internal migration research. The present study

augments the developing countries migration literature by examining for Ghana

both rural-to-urban and urban-to-rural migration.

The main purpose of this paper is to determine the impact of Ghanas inter-sectoral

migration1on migrants welfare. In pursuit of this, we also explore migration patterns

and factors that influence migration decisions. Within the context of migration

between rural and urban areas, our main research questions are as follows:

1We use inter-sectoral migration to refer to migration between rural and urban sectors.


4/42

4

i) What are the major influences on migration decisions?ii) What is the impact of migration on migrants welfare?

The studys analysis consists of the use of both descriptive statistics and econometric

modelling. In the next section, we discuss some salient issues emerging from the

literature. The third section consists of comments relating to data, and a list of

relevant definitions. Migration patterns and a profile of migrants are presented in

section four. The empirical modelling of migrations impact on migrants welfare is

the focus of the fifth section, whilst section six discusses the results of the empirical

analysis. We summarise and conclude in the seventh section.

2. Relevant insights from the migration literature

General literature

A major challenge in the estimation of migrants income gains is the determination of

what they would have earned if they had not migrated. Even though several methods

for estimating migrants income gains have been identified in the literature (see

Lucas, 1997), two approaches seem to dominate. These methods are the application of

migration dummies in earnings (or income) functions, and the estimation of separate

income equations for migrants and non-migrants.

The application of migration dummies (in income functions) has been employed by

Yap (1976) in a study of rural-to-urban migration and urban underemployment in

Brazil. Using census data, Yap estimated income functions to compare the incomes of

migrants with those of non-migrants residing in migrants areas of origin; this was

done for three rural areas (origins of migration) corresponding to the three major

regions in Brazil. For each rural area, the income gain to migrants (from that area)

was estimated by pooling observations on rural non-migrants, rural-to-rural migrants,

and rural-to-urban migrants. The regressors for Yaps income functions included

education level, age group, sex, race, and migration status, with each regressor

represented by a set of dummy variables. In particular, there were four dummy

variables for migration status, namely, rural non-migrant (omitted), rural-to-rural

migrant, recent rural-to-urban migrant, and less recent rural-to-urban migrant.


5/42

5

Yaps (1976) results suggest that rural-to-urban migrants derived significant income

benefits from migration, and that no significant income gains accrued to rural-to-rural

migrants. In spite of the fact that Yaps approach is insightful, it does not capture

convincingly the incomes migrants would have earned if they had not migrated. As

Lucas (1997) points out, there are concerns about Yaps assumption that the returns to

education and other estimated coefficients are identical in both origin and destination

localities. Lucas further observes that Yaps approach is potentially problematic

because of the possible correlation of the migration dummies with unobserved

attributes distinguishing migrants from non-migrants.

The issue of the counterfactual as it relates to migrants incomes has often been

addressed by estimating separate income equations for migrants and non-migrants

(see, for example, Nakosteen and Zimmer, 1980; Pessino, 1991; and Tunali, 2000).

Such methods typically incorporate an adjustment for selectivity bias by employing

techniques such as Lees (1978) two-step method, an approach that is commonly

known as Heckmans (1979) two-step procedure. For the sake of illustration, let on

and off describe the different sets of observations corresponding to individuals who

take a particular decision (such as, the decision to migrate), and those who do not,

respectively. Then, in brief, the Heckman two-step method entails deriving selectivity

variables from a probit (decision) regression, and inserting these variables into income

OLS regressions for the on and off samples. This approach yields consistent

estimates for the truncated that is, the on and off income regressions (see Lee,

1978; and Heckman, 1979). It is worth noting, however, that this popular two-step

method assumes that the error terms in the probit and OLS regressions are normally

distributed; a violation of this assumption might result in failure to detect selectivity

in the OLS regressions (see Lee, 1982).

Using data from Peru, Pessino (1991) applied Heckmans two-step procedure to

estimate wage equations for movers and stayers in three regions; Lima, other urban,

and rural localities. The regressors used in the study captured attributes, such as, work

experience, education, and marital status. Even though Pessino did not find evidence

in support of selectivity amongst movers in any of her samples, she found positive

selectivity amongst stayers in Lima, and negative selectivity amongst stayers in rural


6/42

6

areas. In other words, Pessinos evidence suggests that Lima stayers earned more than

movers would have earned if they had stayed, but contrary to expectations rural

stayers earned less at their location than movers would have earned if they had stayed.

In a more recent study, Agesa (2001) contributes to the discourse on rural-urban

migration by applying the Heckman two-step technique to Kenyan rural and urban

data. Agesas approach is, essentially, identical to that of Nakosteen and Zimmer

(1980); after using the two-step method to estimate migrant and non-migrant income

equations, these equations are then used to generate an urban-rural income gap

variable for inclusion in a migration structural equation. Regressors for the migration

(decision) and income equations include indicators (or proxies) for personal attributes

and human capital. In addition to obtaining a positive and statistically significant

coefficient for the urban-rural income gap variable, Agesas findings suggest that

those rural workers more likely to gain from migrating to the urban sector are those

who migrate. Other studies that have analysed the migration-income link using

corrections for selectivity bias include Falaris (1987; for Venezuela), Lanzona (1998;

for rural Philippines), and Tunali (2000; for Turkey).

The Ghana literature

Even though migration in Ghana has attracted very few econometric analyses, the

available studies lend some support to the view that migration enhances migrants

welfare. Using data from two waves of the Ghana Living Standards Survey (GLSS),

Litchfield and Waddington (2003) employ multivariate analyses to investigate the

impact of migration on welfare. They used standard OLS regressions (of equivalised

household consumption expenditure) and migration dummies to determine whether

migrants are better off than non-migrants. Poverty probits were also used to examine

the impact of migration on the likelihood of being poor.

Litchfield and Waddington observe, that even though the OLS regressions suggest

migrants have a higher standard of living than non-migrants, the migration premium

seemed to have halved between 1991/92 and 1998/99. The poverty probit for 1991/92

showed migrants having a lower probability of being poor (that is, relative to non-

migrants), but that for 1998/99 did not indicate any statistically significant difference


7/42

7

in the probabilities of being poor between migrants and non-migrants. It must be

noted that Litchfield and Waddington found an apparent zero welfare-gap (that is,

between migrants and non-migrants) when the analysis was extended to non-monetary

indicators of welfare. It is worth mentioning, however, that the absence of a correction

for selectivity bias implies that the parameter estimates of the studys linear

regressions are likely to be inconsistent, a limitation the authors acknowledge. Given

the paucity in the context of the Ghana literature of rigorous quantitative analyses

of migration, the study by Litchfield and Waddington is, nevertheless, an important

contribution.

The incorporation of a correction for selectivity bias is a key aspect of a recent

econometric migration study (Migration and Household, 2004). The study explores

in the context of Ghanas Volta Basin the determinants of the migration decision,

placing particular prominence on the role of income in migration decisions. In

accounting for the fact that migrants are non-randomly selected from the population,

the study utilises Heckmans procedure for selectivity bias correction. A major result

of the analysis is the evidence found for expected income gains in influencing

migration decisions. Furthermore, evidence was found to suggest that incomes of

migrant households are higher than those of their non-migrant colleagues. Since the

studys geographical focus was very localised, its findings cannot be generalised for

the entire country. Notwithstanding this limitation, the results and more importantly

the methodology employed constitute a valuable addition to the Ghana migration

literature.

3. Data and Definitions

The 1991/92 and 1998/99 Ghana Living Standards Surveys (GLSS) constitute the

data source for the descriptive analysis, whilst the econometric analysis employs the

latter surveys data2.The Ghana Living Standards Surveys are a series of nationally

representative household surveys, the first of which was carried out in 1987/88. The

1991/92 and 1998/99 surveys, being the third and fourth in the series, are often

2Owing to a difference in questionnaire design with respect to the migration section between the

two surveys, the 1991/92 data do not permit a precise identification of migrants (as defined in thisstudy). We consequently employ the 1991/92 data in the descriptive analysis only, and confine the

econometric analysis to the 1998/99 data.


8/42

8

referred to as GLSS3 and GLSS4, respectively. Apart from the demographic

information collected in the surveys, the GLSS data cover various aspects of living

conditions, such as, consumption, education, health, housing, employment, and

migration. The comprehensive coverage aside, the statistical methods employed in the

surveys make the GLSS data, presumably, the most widely used survey data on living

conditions in Ghana.

At this point, it is pertinent to address issues relating to the definition of a migrant, as

used in our study. Given that our empirical analysis is mainly based on GLSS4 data, it

is instructive to identify some definitions proposed by the GLSS4 Report. The

definitions all relating to persons aged 15 years or more are as follows (GSS,

2000):

In-migrant: a person born outside current place of residence;

Return-migrant: a person born at current place of residence, but who had lived

elsewhere for at least one year, and returned to place of birth;

Migrant: an in-migrant or a return-migrant;

Non-migrant: a person born at current place of residence, and who has never lived

elsewhere for a period lasting, at least, one year.

Whilst the above definitions seem appropriate, it is worth noting that they do not

capture the phenomenon of seasonal migration. Consequently, there is a chance of

classifying many seasonal migrants as non-migrants. This limitation is closely related

to the data collected in the survey. In other words, the surveys data do not permit an

examination of seasonal migration. As a result, the migration-related definitions used

in the present study similarly do not address this limitation.

A second problem with the above definitions relate to the importance placed on

birthplace. For example, the definition of an in-migrant can inappropriately classify

certain persons as in-migrants, as might occur with persons who have always lived in

a rural locality, but were born in a nearby town (possibly, the district capital). In

Ghana (and most likely, in many other developing countries), it is not uncommon for

expectant women to deliver their babies outside their localities of residence. This may

occur in cases where rural residents give birth in nearby towns owing to inadequate


9/42

9

health facilities in their own localities. An expectant woman may also deliver outside

her residential locality simply because of a decision to be with the mother just before

delivery, presumably to facilitate the transfer of knowledge and skill in the nursing of

babies. In the cited and similar instances, the nursing mother typically returns (with

her child) to her usual place of residence, where the child may live permanently or

until he or she becomes an adult. Furthermore, circumstances unrelated to baby

delivery may result in children moving (along with their parents or guardians) to some

other locality and residing there permanently. Thus, a desirable feature of migrant-

related definitions is the capacity for dealing with problems posed by the strict linkage

of migrant status to place of birth.

In the light of the preceding discussion, the following definitions are proposed:

In-migrant: an adult (aged at least 15 years) born outside current place of residence,

and who was, at least, six years old at the time of moving to current place of

residence;

Return-migrant: an adult born at current place of residence (or who moved to

current place of residence before sixth birthday) and who has lived elsewhere for

more than one year and returned to current place;

Migrant: an in-migrant or a return-migrant;

Non-migrant: an adult who has lived at current place of residence since birth (or

before sixth birthday), and has never moved and lived elsewhere for more than one

year.

4. Migration Patterns and Profile of Migrants

The available data lend support to the prevalence of migration in Ghana, since a

sizable proportion of the population are migrants or have migrated at some point in

their lives. In 1991/92, 54 percent of Ghanas population were migrants (that is, either

in-migrants or return-migrants), whereas the migrant share of the population in

1998/99 was 50 percent (see Table 4.1).


10/42

10

Table 4.1: Extent of migration in Ghana; 1991/92 and 1998/99

Migrant status

Share (%) of population

(1991/92)

Share (%) of population

(1998/99)

In-migrant 37.57 34.70

Return-migrant 16.51 15.08Non-migrant 45.91 50.22

Total 100.00 100.00Source: Authors computation using data from the Ghana Living Standards Survey, 1991/92 and

1998/99.

In terms of an origin-destination classification, the data suggest that rural-to-rural and

urban-to-rural forms of population movement dominate Ghanas internal migration.

For example, in 1991/92, 17.1 percent of Ghanas population were rural-to-ruralmigrants and 15.5 percent were urban-to-rural migrants, whereas urban-to-urban and

rural-to-urban migrants constituted 12.4 percent and 4.9 percent, respectively (see

Table 4.2). The pattern of internal migration in 1998/99 was not very different from

that of 1991/92; urban-to-rural migrants accounted for 16.6 percent of the population,

whilst rural-to-rural and urban-to-urban migrants represented 14.4 percent and 10.9

percent of the population, respectively. As reflected in Table 4.2, the rural-to-urban

migrant share (4.5 percent) of the population was the lowest (that is, amongst internal

migrants) in 1998/99. On the whole, and in both survey years, the migrant category

with the lowest proportion of the population was foreign-to-urban migrants, followed

by foreign-to-rural migrants. It is important to note, however, that in both 1991/92 and

1998/99, the majority of persons migrating to Ghana from other countries were

return-migrants (see Tables A1 and A2 in the Appendix).


11/42

11

Table 4.2: Distribution of types of migrants in Ghana; 1991/92 and 1998/99

Migrant Category

Proportion (%) of

population; 1991/923

Proportion (%) of

population; 1998/99

Urban-to-urban 12.40 10.94

Urban-to-rural 15.48 16.55Rural-to-urban 4.90 4.47

Rural-to-rural 17.11 14.42

Foreign-to-urban 1.78 1.28

Foreign-to-rural 2.18 2.12

Non-migrant 46.15 50.22

Total 100.00 100.00Source: Authors computation using data from the Ghana Living Standards Survey, 1991/92 and

1998/99.

The characteristics of migrants have been fairly similar between 1991/92 and

1998/99. With regard to the gender distribution of migrants, females had a higher

share than males in each of the survey years. In 1991/92, 53.6 percent of migrants

were females, whereas the corresponding proportion in 1998/99 was 53.8 percent. It is

worth pointing out, that in 1991/92, even though female migrants outnumbered their

male colleagues, the migration rate amongst males (54.4 percent) was slightly higher

than the rate (53.8 percent) amongst females (see Table A3 in the Appendix). In

1998/99, however, the migration rate amongst females (50 percent) was higher than

that of males (49.53 percent).

An examination of Table 4.3 shows that the majority of Ghanas migrants are less

than forty years old, and that little change occurred in migrants age distribution

between 1991/92 and 1998/99. In each of the two survey years, migrants aged

between fifteen and forty-five years accounted for more than 60 percent of the

migrant population. It must be stressed though, that whilst this age distribution is not

surprising, it appears to be largely due to a generally large share of young persons in

Ghanas population. This point is supported by the fact that amongst the non-migrant

population, the young still dominates (see Table 4.3).

3Information on migrants previous place of residence was missing for 0.5% of migrants.


12/42

12

Table 4.3: Age distribution of migrants and non-migrants; 1991/92 and 1998/99

Age group (in

years)

Share (%)

amongst

migrants;

1991/92

Share (%)

amongst

migrants;

1998/99

Share (%)

amongst non-

migrants;

1991/92

Share (%)

amongst non-

migrants;

1998/99

15 age < 25 18.64 17.38 47.95 43.12

25 age < 35 24.92 24.69 20.07 21.29

35 age < 45 21.41 22.16 11.84 14.30

45 age < 55 16.90 16.24 8.66 9.04

55 age < 65 9.83 9.52 5.11 5.61

Age 65 8.29 10.01 6.36 6.63Total 100.00 100.00 100.00 100.00Source: Authors computation using data from the Ghana Living Standards Survey, 1991/92 and

1998/99.

The main reasons for migrating as indicated by migrants are generally similar in

the two surveys. In 1991/92, other family reasons was the response category that

accounted for the largest share (42.8%) of reasons for migrating. This was followed

by marriage, own employment, spouses employment, other, schooling, and

drought/war in that order. In 1998/99, other family reasons was again the response

category accounting for the largest share (46%) of reasons for migrating. The second

most cited reason in 1998/99 was own employment, with marriage, other, spouses

employment, schooling, and drought/war following in that order.

While broad similarities were evident in the cited reasons for migrating, it is worth

highlighting some gender-related differences. For males, and in both 1991/92 and

1998/99, own employment accounted for the second largest share of reasons for

migrating. In the case of females, however, own employment ranked sixth and fourth

in 1991/92 and 1998/99, respectively. For both 1991/92 and 1998/99, males

accounted for more than 80 percent of persons who migrated because of own

employment. Furthermore, in each of the two survey years, marriage accounted for

the second highest share of reasons for female migration, whereas for males, it ranked

sixth. In each of 1991/92 and 1998/99, females again had a huge share (more than 90

percent) of persons who migrated because of marriage. Another insightful observation

relates to migration induced by spouses employment. In GLSS3, females constituted

about 60 percent of persons who migrated because of spouses employment; thecorresponding figure for GLSS4 was much higher (81.8 percent).


13/42

13

On the whole, other family reasons was the most cited reason for migrating,

irrespective of gender and age group. This notwithstanding, it accounted for a

particularly high share of reasons amongst persons aged less than twenty-five years

(but, at least fifteen years old); the shares were 60.1 percent in 1991/92, and 73.3

percent in 1998/99. Amongst this same age group (that is, fifteen-to-twenty years)

in both survey years schooling was the second most stated reason for migrating, an

unsurprising result.

In order to obtain a rough measure of living standards across migrant status, we

examine for both rural and urban sectors the mean consumption welfare for in-

migrants, return-migrants, and non-migrants. To this end, we define an individuals

consumption welfare as the total consumption expenditure per adult equivalent of that

individuals household, measured in real terms. In 1991/92, return-migrants had the

highest mean consumption welfare, followed by in-migrants and non-migrants, in that

order. In 1998/99, in-migrants had the highest mean consumption welfare, with no

other clear pattern emerging (see Tables A5 and A6 in the Appendix).

A comparison of the mean consumption welfare of internal migrants indicates that

urban-to-urban migrants had the highest consumption welfare in each of the two

survey years, with rural-to-urban, urban-to-rural, and rural-to-rural migrants following

in that order (see Table 4.4). Furthermore, in both 1991/92 and 1998/99, urban non-

migrants had, on the average, a higher level of consumption welfare than urban-to-

rural migrants, whilst rural non-migrants had a lower level of consumption welfare

than rural-to-urban migrants.

Table 4.4: Mean consumption welfare of internal migrants; 1991/92 and 1998/99

Migrant category 1991/92 1998/99

Urban-to-urban 1,845,521.8 1,994,174.1

Urban-to-rural 1,104,548.5 1,206,615.9

Rural-to-urban 1,637,388.2 1,632,757.1

Rural-to-rural 1,017,758.9 1,067,828.3Source: Authors computation using data from the Ghana Living Standards Survey, 1991/92 and

1998/99.


14/42

14

5. Modelling of Migrations Impact on Migrants Welfare

This section outlines a general model for estimating the impact of migration on

migrants welfare, since only a slight modification is required to extend the analysis to

the specific cases of rural-to-urban migration and urban-to-rural migration. The basic

modelling strategy follows very closely Lee (1978) and Nakosteen and Zimmer

(1980), and is summarised as follows:

1. A simultaneous estimation of the following three equations:

a) A migration decision equation, defined over both migrants and non-migrants;b) A welfare equation for migrants; andc) A welfare equation for non-migrants.

2. The use of the two welfare equations and data on both migrants and non-migrants

to estimate the average impact of migration on migrants welfare.

a) Theoretical Framework

Sjaastads (1962) human capital framework constitutes the theoretical underpinning

for the model employed. By viewing migration as an investment in human capital,

Sjaastad suggests that prospective migrants aim to maximise the present value of the

net gains resulting from locational change. For any potential migrant, suppose the

present value of the migration generated net gain is given by:

m

T

ntmtm Cdt-pt

eWW(t)PV -]-[0= (1)

where

Wmtrepresents anticipated welfare at mth prospective destination locality at time t;

Wntrepresents anticipated welfare at origin locality at time t;

Cmdenotes a one-time cost4of migrating to locality m;

T: duration of migration status;

p: implicit discount rate.

4Even though costs associated with migration are not incurred once, recurring costs of locational

change are subsumed in the welfare measure.


15/42

15

The individual does not migrate ifPVm0 for all m;

The individual migrates, if there exists an m for whichPVm> 0, and where this

condition is satisfied by more than one prospective destination, the individual selectsthe destination yielding the maximumPVm.

In order to adapt the above theoretical framework for empirical analyses, we

(following Nakosteen and Zimmer, 1980) make the following assumption:

At any given time, individual i will choose to migrate if the anticipated welfare gain

exceeds the corresponding migration costs.

This implies that at any given time, an individual will migrate if his/her proportionate

welfare gain exceeds the migration costs, as a proportion of welfare. Thus, denoting

the individuals migration costs (as a proportion of welfare) by Qi, individual iwill

migrate if:

[(WmiWni)/Wni] Qi> 0, (2)

and will not migrate if:

[(Wmi Wni)/Wni] Qi 0, (3)

where

Wmidenotes individual is welfare as a migrant; and

Wnidenotes individual is welfare as a non-migrant.

It may be argued that the costs of migration depend on individual attributes (for

example, age, sex, and marital status) and community-level characteristics, such as

the cost of living and the unemployment rate. Thus, the decision to migrate as

indicated in inequalities (2) and (3) may be expressed as a function of (anticipated)

welfare gains, individual attributes, and community characteristics. In the tradition of

similar methodologies (see, for example, Lee, 1978; and Nakosteen and Zimmer,

1980), we adopt a linear functional form for the migration decision equation as

follows:

Individual imigrates if:

Ii = 0+ 1[(Wmi- Wni)/Wni] + .Gi i> 0, (4)

and does not migrate if

Ii = 0+ 1[(Wmi- Wni)/Wni] + .Gi i 0, (5)


16/42

16

where

1: Coefficient of the welfare gain variable

Gi: Vector of variables representing appropriate individual and community

characteristics: Vector of coefficients of the variables in Gi:

i: An error term; and

0: A constant term

It is also reasonable to postulate that an individuals welfare level depends on

personal characteristics (such as educational attainment) and community attributes

(for example, the availability of socio-economic amenities). An individuals welfareequation can consequently be expressed as a function of variables representing

individual and community characteristics. Invoking the argument of Lee (1978) that

(LnWmi-LnWni) and (Wmi- Wni)/Wniare approximately equal, the empirical model is

specified below, with the welfare equations formulated in logarithmic form:

=Ii 0+ 1(LnWmi-LnWni) + .Gi- i

LnWmi= am+ m.Xi+ mi

LnWni= an+ n.Xi+ ni

where

Ii is not observed, but we rather observe

Ii= 1 if Ii> 0, andIi= 0 if Ii

0;

LnWmi: log of migrant welfare

LnWni: log of non-migrant welfare

Xi:Vector of variables representing relevant individual and community characteristics

m: Migrant vector of coefficients of the variables inXi

n: Non-migrant vector of coefficients of the variables inXi

i, mi, and niare all Normally distributed error terms with zero mean and constant

variance;

All other variables retain their definitions.


17/42

17

The observed variables in the model are the limited dependent variables, Wmiand Wni,

the dichotomous migration decision variable Ii, and the variables contained in vectors

Gi,andXi. In general, the ordinary least squares (OLS) technique is inappropriate for

estimating the welfare equations due to its failure to account for selectivity bias (see

Nakosteen and Zimmer, 1980; and Lee, 1978). As a result, we employ Lees (1978)

proposed solution5; the welfare equations are modified by incorporating appropriate

selectivity variables, and adding error terms with zero means. It is worth

emphasising that even though the variables contained in GiandXirepresent individual

and community characteristics, the two vectors need not contain identical variables.

b) Estimation procedure

In order to estimate all the parameters of the model, the following estimation

technique is used:

i. Probit estimation of the reduced-form migration decision equation

The regressors in this equation consist of the exogenous variables in all the three

structural equations. Fitted values ( i ) obtained from this (first) stage are used to

construct variables u1iand u2i, where:

u1i= -f ( i ) /F( i )is the selectivity variable for the migrant welfare equation;

u2i=f ( i )/ [1-F( i )] is the selectivity variable for the non-migrant welfare

equation;

f: the density function of a standard normal random variable; and

F: the cumulative distribution function of a standard normal random variable.

ii. Insertion of u1iand u2iinto the appropriate welfare equations and estimating the

welfare equations by OLS

Estimates obtained using the above two-step procedure are known to be consistent

(see Lee, 1978).

5As noted earlier, this technique is often referred to as the Heckman two-step method.


18/42

18

iii. Probit estimation of the structural migration decision equation

In this step, the consistent parameter estimates of the welfare equations are used to

obtain fitted values of the logarithm of welfare, which are in turn, used to compute

estimates of (LnWmi-LnWni). Together with other exogenous variables, the estimates

of (LnWmi-LnWni) are inserted into the structural decision equation to obtain the

probit estimates of the structural migration decision equation.

c) Determination of migrations impact on migrants welfare

In order to estimate migrations impact on migrants welfare, we employ simulations

of counterfactual scenarios, coupled with the estimation of an index of welfare gain

due to migration. This is accomplished by computing an index of welfare differential

between migration and non-migration scenarios as follows:

For any sub-group () of the entire sample, letN() be the corresponding population

size;

Then, for any sub-group (), the average proportionate welfare increment attributable

to migration may be expressed as:

=

i Wni

WniWmi

Ng

)(

1

Where, Wmiand Wni individual is welfare levels as a migrant, and as a non-

migrant, respectively are proxied by their corresponding fitted values for cases

where they are unobserved.

6. Empirical Results

a) Introductory note and discussion of regressors

In applying the described model to an analysis of migration between Ghanas rural

and urban sectors, a number of adjustments are made. These adjustments stem from


19/42

19

an explicit recognition of two separate migratory movements, namely, urban-to-rural

migration and rural-to-urban migration. Two separate analyses are consequently

carried out, one for each migratory movement. It should be noted also, that a migrant

as used in the outlined analytical model is equivalent to an in-migrantas defined

earlier. Furthermore, we employ a consumption measure of welfare, where, an

individuals consumption welfare is defined as the total consumption expenditure per

adult equivalent of that individuals household, measured in real terms.

For the analysis of urban-to-rural migration, the three structural equations consist of a

migration decision equation (defined over a pooled sample of urban non-migrants and

urban-to-rural in-migrants), a welfare equation for urban-to-rural in-migrants, and a

welfare equation for urban non-migrants. Similarly, in analysing rural-to-urban

migration, the relevant structural equations comprise a migration decision equation

(defined over a pooled sample of rural non-migrants and rural-to-urban in-migrants), a

welfare equation for rural-to-urban in-migrants, and a welfare equation for rural non-

migrants.

The entire data contained 14,196 observations on persons aged 15 years or more. The

set of regressors for each of the two migration status equations includes variables for

highest educational attainment, age group, marital status6,location, and (anticipated)

welfare gain. The choice of these variables is informed by theory and by a preliminary

analysis that explored various combinations of regressors. In particular, in order to

ensure that the parameters of the structural (migration) decision equation are

identified, the welfare equations contain at least one exogenous variable that is

excluded from the structural migration equation (Nakosteen and Zimmer, 1980).

On a priori grounds, we expect education to have a positive impact on rural-to-urban

migration, and to have a negative effect on urban-to-rural migration. With regard to

age, the literature suggests that young adults tend to have a higher propensity to

migrate than the elderly. It is therefore expected that the tendency to migrate amongst

lower age groups will be higher than that of higher age groups. The impact (on

migration) of anticipated welfare gain is expected to be positive; an expectation

6Here, what is used is a variable that places emphasis on whether (or not) a person has always been

single.


20/42

20

rooted in both theoretical and intuitive considerations. Finally, apart from serving as

control variables, the locational variables can provide spatial-related insights into

population movements between rural and urban localities.

Theory and preliminary data examination were again crucial in the selection of

regressors for the welfare equations. Most of the regressors were common to all four

welfare equations. These common regressors include variables for highest educational

attainment, employment category, selectivity, and location, as well as variables that

capture household characteristics, such as size, access to pipe-borne water for

drinking, and the use of electricity for lighting. In the analysis of urban-to-rural

migration however, a gender variable and regressors for other household

characteristics were included7(see Table 6.1 for a list of variables used in the

analysis).

7An initial analysis suggested the relevance of these regressors is confined to the urban-to-rural model.


21/42

21

Table 6.1: List of regressors for the econometric analysis

Variable Definition

hhsize Household size

agegp1 1 if 15 age (in years) < 25, 0 otherwise

agegp2 1 if 25 age (in years) < 35, 0 otherwiseagegp3 1 if 35 age (in years) < 45, 0 otherwiseagegp4 1 if 45 age (in years) < 55, 0 otherwiseagegp5 1 if 55 age (in years) < 65, 0 otherwiseagegp6 (the omitted category) 1 if age (in years) 65sex1 1 if male, 0 otherwise

mar5 1 if never married, 0 otherwise

hiedq1 (the omitted category) Dummy for highest educational qualification; 1 if no

educational qualification, 0 otherwise

hiedq2 Dummy for highest educational qualification; 1 ifMSLC/BECE, 0 otherwise

hiedq3 Dummy for highest educational qualification; 1 if

vocational, commercial, O or A levelhiedq4 Dummy for highest educational qualification; 1 if T/T,

nursing, or Tech/Prof, 0 otherwise


University degree holder, 0 otherwise


unspecified qualification, 0 otherwise

empcat1 (the omitted category) 1 if unemployed, 0 otherwise

empcat2 1 if employed in agriculture, 0 otherwise

empcat3 1 if employed in industry, 0 otherwise

empcat4 1 if employed in services, 0 otherwise

empcat5 1 if unspecified employment category, 0 otherwise

farmliv1 1 if household owns/operates a farm, keeps livestock, or is

engaged in fishing, 0 otherwisefoodpr1 1 if household is engaged in food processing, 0 otherwise

othbus1 1 if household is engaged in some other non-farm business,

0 otherwise

pbw1 1 if household drinks pipe-borne water, 0 otherwise

eg1 1 if households main lighting source is electricity or

generator

ez1 (the omitted category) Dummy for ecological zone; 1 if Coastal, 0 otherwise

ez2 Dummy for ecological zone; 1 if Forest, 0 otherwise

ez3 Dummy for ecological zone; 1 if Savannah, 0 otherwise

reg1 Regional dummy; 1 if Western, 0 otherwise

reg2 Regional dummy; 1 if Central, 0 otherwise

reg3 Regional dummy; 1 if Greater Accra, 0 otherwisereg4 Regional dummy; 1 if Eastern, 0 otherwise

reg5 Regional dummy; 1 if Volta, 0 otherwise

reg6 Regional dummy; 1 if Ashanti, 0 otherwise

reg7 Regional dummy; 1 if Brong Ahafo, 0 otherwise

reg8 Regional dummy; 1 if Northern, 0 otherwise

reg9 Regional dummy; 1 if Upper West, 0 otherwise

reg10 (the omitted category) Regional dummy; 1 if Upper East, 0 otherwise

diflnWh Estimated migrant-non-migrant gap in log of welfare

sel selectivity variable

_cons Constant term


22/42

22

b) Factors influencing urban-to-rural migration

The results of the analysis (see Table 6.2) generally conform to a priori expectations.

The significant and positive coefficient of diflnWh (the welfare gap variable)

suggests that the anticipated welfare gain is a major influence on urban-to-rural

migration decisions in Ghana. The results further show, that in comparison with their

colleagues who are (or have ever been) married8,urban residents who have never

been married are less likely to migrate to rural areas. This finding is contrary to what

one might expect, given the perception that persons who are less encumbered will be

more likely to migrate. This notwithstanding, it is possible that the result is a

reflection of a tendency for unmarried migrants to marry within a few years after

migrating. In this connection, it is worth noting that almost 60 percent of all in-

migrants in the entire sample have lived at their current place of residence for at least

ten years. Thus, it is plausible that most of those migrants who had never been

married at the time of migrating had experienced a change in marital status by the

time of the survey.

The role of education in urban-to-rural migration decisions is insightful. The evidence

suggests that the attainment of education beyond the vocational, commercial,

Ordinary (O), or Advanced (A) level tends to reduce the probability of migrating

from an urban area to the rural sector. This evidence is not surprising; it is common

knowledge that persons with higher levels of education tend to have a preference for

settling in urban areas, and have a better chance (relative to the less educated) of

finding employment in urban centres. Our findings further suggest, that urban

residents in the Upper East Region are more likely (relative to their counterparts in

other Regions, except the Upper West) to migrate to rural areas. Given that the Upper

East Region is one of the poorest in Ghana, this finding is reasonable.

8Marriage, as used here, includes informal or loose unions.


23/42

23

c) Factors influencing consumption levels

With regard to living standards that is, in terms of consumption of urban-to-rural

in-migrants, there is strong support for a negative effect of household size on welfare.

The results show a strong negative association between household size and the

welfare of urban-to-rural in-migrants. Moreover, the magnitude of this link between

household size and welfare is the same for both urban-to-rural in-migrants and urban

non-migrants.


24/42

24

Table 6.2: Results of the urban-to-rural migration model

welf1: OLS welfare regression for urban-to-rural in-migrants

welf0: OLS welfare regression for urban non-migrants

probit: Probit regression for the pooled sample of urban non-migrants and urban-to-

rural in-migrants

---------------------------------------------------------Variable welf1 welf0 probit

---------------------------------------------------------sex1 -0.06** -0.10***

(0.03) (0.02)hhsize -0.10*** -0.09***

(0.01) (0.00)

hiedq2 0.14*** 0.09*** -0.05(0.03) (0.02) (0.06)hiedq3 0.27*** 0.31*** 0.09

(0.07) (0.03) (0.10)hiedq4 0.43*** 0.28*** -0.27**

(0.08) (0.05) (0.13)

hiedq5 1.20*** 0.46** -2.13***(0.27) (0.22) (0.48)

hiedq6 0.23 0.00 -0.52(0.31) (0.16) (0.47)

empcat2 0.17*** 0.02(0.06) (0.04)

empcat3 0.21*** 0.07**(0.07) (0.04)

empcat4 0.26*** 0.04(0.06) (0.03)

empcat5 -0.07 0.10(0.11) (0.06)

farmliv1 0.00 -0.07**(0.05) (0.03)

foodpr1 -0.06 -0.06**

(0.04) (0.03)othbus1 0.05 0.09***

(0.03) (0.02)pbw1 -0.13* 0.06

(0.07) (0.04)eg1 0.14** 0.36***

(0.06) (0.04)ez2 0.18*** 0.04 0.33***

(0.06) (0.04) (0.09)ez3 0.13 -0.01 -0.18

(0.08) (0.06) (0.14)reg1 (Western) 0.77*** 0.25** -1.22***

(0.15) (0.12) (0.25)

reg2 (Central) 0.65*** -0.19 -2.22***(0.15) (0.12) (0.27)

reg3 (Gt. Accra) 0.75*** 0.35*** -1.29***(0.16) (0.12) (0.25)

reg4 (Eastern) 0.45*** 0.05 -1.09***(0.14) (0.11) (0.24)

reg5 (Volta) 0.67*** -0.01 -1.87***(0.15) (0.11) (0.25)


25/42

25

Table 6.2: continued

---------------------------------------------------------Variable welf1 welf0 probit

---------------------------------------------------------

reg6 (Ashanti) 0.69*** 0.23** -1.43***(0.15) (0.11) (0.25)

reg7 (B.-Ahafo) 0.57*** 0.18* -1.43***(0.14) (0.11) (0.24)

reg8 (Northern) 0.25 0.12 -1.16***(0.15) (0.10) (0.23)

reg9 (Upper West) -0.02 0.10 0.56**(0.15) (0.12) (0.24)

sel -0.33*** -0.02(0.10) (0.07)

agegp1 0.18(0.13)

agegp2 0.19*

(0.11)agegp3 0.15

(0.11)agegp4 0.22*

(0.12)agegp5 0.09

(0.13)mar5 -0.46***

(0.08)diflnWh 3.41***

(0.13)_cons 13.31*** 14.11*** 2.01***

(0.21) (0.12) (0.27)

----------------------------------------------------------Number of obs. 1360.00 2720.00 4080.00R-squared 0.39 0.46Adj. R-squared 0.38 0.45F-statistic 30.53 81.80Pseudo R-squared 0.38

Chi-squared 1974.61----------------------------------------------------------

Legend: * p


26/42

26

As expected, strong support is found amongst both urban-to-rural in-migrants and

urban non-migrants for a welfare-enhancing role of education. In both welfare

equations, virtually all the education dummies have significantly positive effects on

welfare. The results further suggest that being employed (whether in agriculture,

industry, or services) enhances the welfare of urban-to-rural in-migrants, with the

services sector generating the biggest impact. In comparison with urban-to-rural in-

migrants, urban non-migrants specific employment sector appears to be less

important in influencing welfare. Nevertheless, working in the industrial sector exerts

a positive effect on the welfare of urban non-migrants.

The results also suggest that having electricity (or a generator) for lighting enhances

the welfare of both urban-to-rural in-migrants and urban non-migrants. This finding is

tenable, given the health hazards associated with some other forms of lighting. An

intriguing finding of our analysis is the apparent higher welfare of females relative to

males. This finding holds for both welfare equations. In other words, it does appear

that female urban-to-rural in-migrants are slightly better off than their male

counterparts, and also, that the welfare of male urban non-migrants is lower (on

average) than that of their female counterparts. It must be stressed however, that this

result is not conclusive, since the welfare measure is equal for all members of a given

household, irrespective of gender.

Another notable finding of this study relates to self-selectivity. For the urban-to-rural

welfare equation, the coefficient of the selectivity variable (sel) is statistically

significant and negative. This provides support for the positive selectivity of urban-to-

rural in-migrants. Other results relating to urban non-migrants are worth noting; all

things being equal, individuals whose households are engaged in an agricultural or

food processing enterprise are likely to have lower levels of welfare relative to those

whose households are not engaged in these activities. On the other hand, the results

suggest that an urban non-migrants welfare tends to be enhanced by the households

ownership of some other business.

All things being equal, urban-to-rural in-migrants living in the Upper East Region

tend to have lower welfare levels than their colleagues in every other Region, with the

exception of the Upper West and Northern Regions. This finding reflects the


27/42

27

relatively low standards of living in the three northern (that is, Northern, Upper West,

and Upper East) Regions of Ghana (see GSS, 2000). Roughly similar results, but to a

much lesser degree, are found for urban non-migrants; living standards amongst urban

non-migrants residing in the Upper East Region tend to fall below those of their

counterparts in the Western, Greater Accra, Ashanti, and Brong-Ahafo Regions9.

Urban-to-rural in-migrants located in the forest ecological zone also tend to have

higher living standards than their coastal zone counterparts, a result that is apparently

linked to the concentration of cocoa farming in the rural Forest ecological zone (see

Coulombe and McKay, 2003).

d) Welfare-gain from urban-to-rural migration

The average proportionate welfare gap between in-migrant and non-migrant scenarios

is very informative. On the whole, migration has a positive impact on the welfare of

urban-to-rural in-migrants, as they incurred (on average) a 27.7 percentage welfare

gain (see Table 6.3). Further examination shows that 56.3 percent of urban-to-rural in-

migrants gained from migrating, their average gain being 72.2 percent. These findings

suggest that although gains are not guaranteed from urban-to-rural migration,

participants often benefit considerably.

A similar simulation for urban non-migrants shows that if they were to migrate to the

rural sector, they would incur a 30.5 percent welfare loss on average. As shown in

Table 6.3, migrating to the rural sector would leave the overwhelming majority (83.2

percent) of urban non-migrants un-rewarded. Indeed, amongst the three categories of

individuals shown in the Table, only urban-to-rural in-migrants gain from migration.

Whilst the findings summarised in Table 6.3 are not a direct test for selectivity bias,

they lend support to the notion that migrants and non-migrants are often non-

randomly selected from the population. Migrants tend to be those who have a better

chance (compared with non-migrants) of gaining from migration.

9The coefficients of the other regional dummies were not statistically significant.


28/42

28

Table 6.3: Migration-generated welfare gains; urban-to-rural migration

(selectivity bias adjusted)

Number of

persons

Mean

percentage

welfare gain

Percentage

with welfare

gain

Percentage

without

welfare gainUrban-to-rural

in-migrants

1,360 27.66 56.25

(Mean % gain

= 72.23)

43.75

(Mean % loss

= 29.64)

Urban non-

migrants

2,720 -30.49 16.84

(Mean % gain

= 48.68)

83.16

Mean % loss =

(46.52)

Urban-to-rural

in-migrants

and urban non-

migrants

4,080 -11.10 29.98

(Mean % gain

= 63.41)

70.02

(Mean % loss

= 43.00)

Since our findings are based on a correction for selectivity bias, it is worth indicating

the outcome of an analysis that does not make such an adjustment. In the absence of a

correction for selectivity bias, the analysis of welfare-gains (that is, from urban-to-

rural migration) produces considerably different results from that shown in Table 6.3

(see Table A7 in the Appendix). In failing to correct for selectivity bias, the analysis

generally shows gains from urban-to-rural migration for all categories of individuals

considered. It would be recalled from our discussion of the welfare regression

estimates for urban-to-rural in-migrants that there is evidence for the positive

selectivity of urban-to-rural in-migrants. Thus, failure to correct for selectivity bias

leads to an overestimation of migrant welfare, and consequently, to an overestimation

of gains from urban-to-rural migration.

e) Factors influencing rural-to-urban migration

Many of the results from the analysis of rural-to-urban migration conform to that of

the urban-to-rural analysis. The findings provide strong support for the importance

in rural-to-urban migration decisions of anticipated welfare gains, as reflected by

the performance of the variablediflnWh (see Table 6.4). The results further point to

an absence of a strong influence of age in rural-to-urban migration decisions.Compared with rural dwellers who are (or have ever been) married, rural residents


29/42

29

who have never been married are less likely to migrate to the urban centres. As

suggested earlier, it might be misleading to draw strong conclusions from this finding,

owing to the long duration of stay at current residence of most in-migrants in our

sample.

In comparison with rural residents whose educational attainment is either lower than

MSLC/BECE10or unspecified, rural dwellers with MSLC/BECE (that is, as their

highest educational attainment) are less likely to migrate to urban centres. Similarly,

rural dwellers with the highest educational attainment being vocational, commercial,

O, or A level, are less likely to migrate to the urban sector, that is, relative to the

reference group (those with a lower-than-MSLC/BECE or with an unspecified

educational attainment). Although the underlying reasons for these findings are

unclear, the significant and positive coefficient of hiedq5 (the dummy for the

holding of a university degree) provides strong support for a tendency for university

graduates to settle in urban centres. Our results further show that the Upper East

Regions rural residents are less likely (compared to their counterparts in other

Regions) to migrate to urban centres. Considering the Upper East Regions status as

one of the poorest in Ghana, this finding is consistent with the view that very poor

individuals or households are often unable to invest in migration.

f) Factors influencing consumption levels

One of our robust results is the strong support established for a negative relationship

between welfare and household size. This relationship exhibits similar orders of

magnitude amongst in-migrants and non-migrants. All other things being equal, rural-

to-urban in-migrants whose households drink pipe-borne water enjoy a higher level of

welfare, relative to their counterparts lacking this amenity. Furthermore, the results

for both welfare equations suggest that living standards are enhanced by the use of

electricity (or a generator) for lighting. The welfare equations for both rural non-

migrants and rural-to-urban in-migrants further reflect a high and positive link

between education and welfare. This positive association between education and

10

MSLC is the Middle School Leaving Certificate (no longer awarded), and the BECE is the BasicEducation Certificate Examination. Each of these represents the highest qualification at the basic

education level.


30/42

30

welfare conforms to a priori expectations, and highlight the need to attach immense

importance to education in national development policies.

Table 6.4: Results of the rural-to-urban migration model

welf1: OLS welfare regression for rural-to-urban in-migrants

welf0: OLS welfare regression for rural non-migrants

probit: Probit regression for the pooled sample of rural non-migrants and rural-to-

urban in-migrants

----------------------------------------------------------

Variable welf1 welf0 probit----------------------------------------------------------hhsize -0.09*** -0.07***

(0.01) (0.00)hiedq2 0.11** 0.04** -0.18**

(0.05) (0.02) (0.08)hiedq3 0.34*** 0.14*** -0.51***

(0.08) (0.04) (0.15)hiedq4 0.35*** 0.20** -0.09

(0.09) (0.09) (0.19)hiedq5 0.30 0.94** 1.74***

(0.34) (0.37) (0.67)empcat2 -0.14* 0.00

(0.07) (0.02)empcat3 0.06 0.06

(0.09) (0.05)

empcat4 0.09 0.08**(0.07) (0.03)

empcat5 0.02 -0.07*(0.14) (0.04)

othbus1 0.07 0.12***(0.05) (0.02)

pbw1 0.34*** 0.05(0.11) (0.03)

eg1 0.38*** 0.11***(0.08) (0.03)

ez2 0.24*** 0.11*** -0.58***(0.08) (0.02) (0.10)ez3 0.05 0.21*** 0.64***

(0.11) (0.04) (0.15)reg1 (Western) 0.46*** 1.03*** 6.87***

(0.13) (0.05) (0.46)reg2 (Central) (dropped) 0.81*** 7.31***

(0.05) (0.48)reg3 (Gt. Accra) 0.51*** 1.00*** 7.76***

(0.16) (0.08) (0.46)reg4 (Eastern) 0.20 0.76*** 7.36***

(0.16) (0.04) (0.45)reg5 (Volta) -0.02 0.89*** 8.07***

(0.14) (0.05) (0.47)reg6 (Ashanti) 0.23 0.91*** 7.70***

(0.16) (0.05) (0.46)


31/42

31


----------------------------------------------------------Variable welf1 welf0 probit

----------------------------------------------------------reg7 (B.-Ahafo) 0.32* 0.74*** 6.87***

(0.17) (0.04) (0.45)reg8 (Northern) 0.16 0.37*** 6.48***

(0.20) (0.04) (0.44)reg9 (Upper West) 0.36 0.15*** 3.78

(0.49) (0.05) (.)sel -0.03 0.12

(0.11) (0.08)agegp1 -0.26*

(0.15)agegp2 -0.20

(0.13)agegp3 0.19

(0.12)

agegp4 0.18(0.13)

agegp5 0.21(0.15)

mar5 -0.43***(0.12)

diflnWh 3.23***(0.14)

_cons 13.56*** 13.13*** -8.00***(0.35) (0.05) (0.47)

----------------------------------------------------------Number of obs. 523.00 4447.00 4970.00R-squared 0.54 0.35

Adj. R-squared 0.52 0.34F-statistic 25.63 97.58Pseudo R-squared 0.41chi-squared 1374.95----------------------------------------------------------Legend: * p


32/42

32

With the exception of being employed in the agricultural sector (which is negatively

linked to welfare), the specific sector of ones employment exerts no significant

impact on the welfare of rural-to-urban in-migrants. In the case of rural non-migrants,

being employed in the services sector relative to being jobless is positively

associated with a higher level of welfare. Rural non-migrants whose households are

engaged in business enterprises (other than agricultural or food processing ventures)

have higher welfare levels compared with those whose households do not engage in

such activities.

g) Welfare-gain from rural-to-urban migration

On the basis of our results, rural-to-urban migration is generally very rewarding for

rural-to-urban in-migrants. By migrating to urban localities, rural-to-urban in-

migrants increased their welfare by 46.3 percent, on average (see Table 6.5). This

percentage increment is considerably higher than that of their urban-to-rural

counterparts (see Table 6.3). Furthermore, the majority (68.5 percent) of rural-to-

urban in-migrants gained by migrating, the mean welfare-gain of the gainers being

80.1 percent. For the minority (31.6 percent) of rural-to-urban in-migrants who did

not gain, the average welfare-loss was 26.9 percent. Interestingly, our findings

suggest that for most rural dwellers, rural-to-urban migration is not necessarily

profitable. As shown in Table 6.5, if rural non-migrants were to migrate to urban

areas, they would incur, on average, a welfare-loss of 2 percent, and 38.4 percent of

such migrants would gain.


33/42

33

Table 6.5: Migration-generated welfare gains; rural-to-urban migration

(selectivity bias adjusted)

Number of

persons

Mean

percentage

welfare gain

Percentage

with welfare

gain

Percentage

without

welfare gainRural-to-urban

in-migrants

523 46.30 68.45

(Mean % gain

= 80.05)

31.55

(Mean % loss

= 26.92)

Rural non-

migrants

4,447 -1.97 38.36

(Mean % gain

= 55.52)

61.64

(Mean loss =

37.75)

Rural-to-urban

in-migrants

and rural non-

migrants

4,970 3.11 41.53

(Mean % gain

= 59.77)

58.47

(Mean % loss

= 37.13)

For the purpose of comparison, Table A8 (in the Appendix) shows welfare-gains from

rural-to-urban migration when no correction is made for selectivity bias. As can be

seen, although the two sets of results are different, the disparities are not as large as

those found for the urban-to-rural analysis. This gives credence to the lack of

significance of the selectivity variables in the welfare regressions for rural-to-urban

in-migrants and rural non-migrants (see Table 6.4). The results in Table 6.5 do

suggest however, that rural residents who migrate to the urban areas tend to be those

individuals who have a better chance of reaping a welfare-gain.

h) Impact of return-migration on migrants welfare

Our multivariate analysis has so far focused on in-migrants. Given the availability of

data on return-migrants, it is instructive to investigate the impact of migration on the

welfare of persons who have returned to their origin localities after engaging in either

urban-to-rural or rural-to-urban migration. Thus, as an ancillary exercise, welfare

regressions (with return-migrant dummies) are used to examine the impact of

migration on the welfare of those return-migrants whose previous form of migration

was urban-to-rural or rural-to-urban. We therefore estimate the following two welfare

equations:


34/42

34

i) urbwelf: OLS welfare equation for a pooled sample of urban non-migrantsand rural-to-urban return-migrants (that is, urban-to-rural-to-urban

migrants);

ii) rurwelf: OLS welfare equation for a pooled sample of rural non-migrantsand urban-to-rural return-migrants (that is, rural-to-urban-to-rural

migrants);

In each of the two equations, the sample consists of non-migrants and return-migrants

residing in the same locality. This feature non-existent in samples for our main

model facilitates the use of dummy variables for return-migrant status. The dummy

variables are defined as follows:

du1: 1 if urban-to-rural-to-urban migrant, 0 if urban non-migrant;

dr1: 1 if rural-to-urban-to-rural migrant, 0 if rural non-migrant.

Table 6.6: Impact of inter-sectoral migration on return-migrants welfare

-------------------------------------------Variable urbwelf rurwelf

-------------------------------------------du1 -0.03

(0.05)

sex1 -0.09*** -0.02(0.02) (0.01)

hhsize -0.09*** -0.08***(0.00) (0.00)

hiedq2 0.09*** 0.05***(0.02) (0.02)

hiedq3 0.31*** 0.11***(0.03) (0.04)

hiedq4 0.29*** 0.23***

(0.05) (0.07)hiedq5 0.45** 0.63**

(0.22) (0.30)hiedq6 0.00 0.06

(0.16) (0.15)empcat2 0.02 -0.01

(0.03) (0.02)

empcat3 0.07** 0.07*(0.03) (0.04)

empcat4 0.03 0.09***(0.02) (0.03)

empcat5 0.09 -0.06*(0.06) (0.04)

farmliv1 -0.07** -0.02(0.03) (0.03)

foodpr1 -0.05** -0.03(0.02) (0.02)

othbus1 0.10*** 0.12***(0.02) (0.02)


35/42

35


-------------------------------------------Variable urbwelf rurwelf

-------------------------------------------pbw1 0.06** 0.03

(0.03) (0.02)eg1 0.36*** 0.14***

(0.03) (0.02)ez2 0.05 0.08***

(0.04) (0.02)ez3 0.01 0.17***

(0.06) (0.03)reg1 (Western) 0.27** 1.03***

(0.11) (0.05)reg2 (Central) -0.18 0.78***

(0.11) (0.05)reg3 (Gt. Accra) 0.38*** 1.12***

(0.11) (0.06)

reg4 (Eastern) 0.08 0.74***(0.10) (0.04)

reg5 (Volta) -0.00 0.91***(0.11) (0.04)

reg6 (Ashanti) 0.25** 0.92***(0.11) (0.04)

reg7 (B.-Ahafo) 0.19* 0.78***(0.11) (0.04)

reg8 (Northern) 0.12 0.40***(0.10) (0.03)

reg9 (Upper West) 0.11 0.18***(0.11) (0.04)

dr1 0.05**

(0.02)_cons 14.07*** 13.25***

(0.11) (0.05)-------------------------------------------Number of obs. 2839.00 5437.00R-squared 0.46 0.37

Adj. R-squared 0.45 0.37F-statistic 85.07 114.00

-------------------------------------------Legend: * p


36/42

36

Our findings generally conform to those of the main model. The strong negative link

between household size and welfare is confirmed. The strength of this link is about

the same in both urban and rural areas (see Table 6.6). The positive welfare impact of

education is given further support in both rural and urban samples. In urban areas,

industrial sector workers tend to have a significantly higher level of welfare than the

unemployed, whereas in the rural areas it is the services sector workers who are

significantly better off than the unemployed. The results further highlight regional

welfare disparities, especially in rural areas; the three northern Regions (Northern,

Upper West, and Upper East) are shown to have the lowest rural welfare levels, whilst

the Greater Accra and Western Regions have the highest rural welfare levels.

We now focus attention on the coefficients of the return-migrant dummies, which are

the main variables of interest. Even though the coefficient of du1 (the return-

migrant dummy for the urban sample) has a negative sign, it is not statistically

significant, thus providing little insight into the welfare impact of migration on return-

migrants. The coefficient of dr1 (the return-migrant dummy for the rural sample),

on the other hand, is positive and significant, with a p-value of 0.015. This provides

evidence in support of return-migrants (that is rural-to-urban-to-rural migrants) being

better off than rural non-migrants. This finding is consistent with our earlier result

suggesting that rural-to-urban migration is, on average, very profitable for participants

(see Table 6.5).

7. Conclusion

This paper has examined the impact of migration between rural and urban sectors

on migrants welfare, using data from Ghanas 1998/99 Living Standards Survey.

Employing a consumption measure of welfare and a model that corrects for selectivity

bias, the analysis has also highlighted factors influencing migration decisions between

Ghanas rural and urban areas.

Our findings underscore the importance of anticipated welfare gains and personal

attributes in migration decisions. We also find support for the positive selectivity of

urban-to-rural migrants. In addition, estimates of migration gains suggest that


37/42

37

although some migrants incur welfare losses, migration enhances on average the

welfare of migrants, but would reduce the mean welfare of non-migrants if they were

to migrate. Finally, the average welfare increment derived by rural-to-urban migrants

is proportionately much higher than what accrues to their urban-to-rural counterparts.


38/42

38

Appendix

Table A1: Distribution of migrant type by in- or return-migrant status; 1991/92

Share (%) of

in-migrants

Share (%) of

return-migrants Total

Urban-to-urban 74.10 25.90 100.00

Urban-to-rural 58.75 41.25 100.00

Rural-to-urban 85.30 14.70 100.00

Rural-to-rural 77.13 22.87 100.00

Foreign-to-urban 44.68 55.32 100.00

Foreign-to-rural 44.78 55.22 100.00Source: Authors computation using data from the Ghana Living Standards Survey, 1991/92.

Table A2: Distribution of migrant type by in- or return-migrant status; 1998/99Share (%) of

in-migrants

Share (%) of

return-migrants Total

Urban-to-urban 81.25 18.75 100.00

Urban-to-rural 59.58 40.42 100.00

Rural-to-urban 83.02 16.98 100.00

Rural-to-rural 75.51 24.49 100.00

Foreign-to-urban 50.29 49.71 100.00

Foreign-to-rural 33.46 66.54 100.00Source: Authors computation using data from the Ghana Living Standards Survey, 1998/99.

Table A3: Extent of migration, by gender; 1991/9211

Sex Non-migrant share

(%) of population

Migrant share (%)

of population

Row Total

Male 45.60

45.79

54.40

46.38

100.00

Female 46.19

54.21

53.81

53.62

100.00

Column Total 100.00 100.00Source: Authors computation using data from the Ghana Living Standards Survey, 1991/92.

11For each cell in the Table, the first value represents the row percentage, whereas the second

represents the column percentage.


39/42

39

Table A4: Extent of migration, by gender; 1998/9912

Sex Non-migrant share

(%) of population

Migrant share (%)

of population

Row Total

Male 50.47

46.70

49.53

46.22

100.00

Female 50.00

53.30

50.00

53.78

100.00

Column Total 100.00 100.00Source: Authors computation using data from the Ghana Living Standards Survey, 1998/99.

Table A5: Mean consumption welfare (in constant cedis) across migrant status

and locality of residence; 1991/92

Migrant status Urban Rural

In-migrant 1,779,164.7 1,049,810.4

Return-migrant 1,984,938.8 1,104,967.7

Non-migrant 1,541,151.7 870,465.35Source: Authors computation using data from the Ghana Living Standards Survey, 1991/92.

Table A6: Mean consumption welfare (in constant cedis) across migrant status

and locality of residence; 1998/99

Migrant status Urban RuralIn-migrant 2,085,238.2 1,341,066.6

Return-migrant 1,848,170.6 1,215,428.4

Non-migrant 1,892,204.3 1,053,309.6Source: Authors computation using data from the Ghana Living Standards Survey, 1998/99.

12For each cell in the Table, the first value represents the row percentage, whereas the second

represents the column percentage.


40/42

40

Table A7: Migration-generated welfare gains; urban-to-rural migration

(selectivity bias unadjusted)

Number of

persons

Mean

percentage

welfare gain

Percentage

with welfare

gain

Percentage

without

welfare gainUrban-to-rural

in-migrants

1,360 30.88 57.87

(Mean % gain

= 74.56)

42.13

(Mean loss =

29.17)

Urban non-

migrants

2,720 11.86 46.36

(Mean % gain

= 60.84)

53.64

(Mean % loss

= 30.48)

Urban-to-rural

in-migrants

and urban non-

migrants

4,080 18.20 50.20

(Mean % gain

= 66.13)

49.80

(Mean % loss

= 30.11)

Table A8: Migration-generated welfare gains; rural-to-urban migration

(selectivity bias unadjusted)

Number of

persons

Mean

percentage

welfare gain

Percentage

with welfare

gain

Percentage

without

welfare gain

Rural-to-urban

in-migrants

523 36.93 63.67

(Mean % gain= 73.32)

36.33

(Mean % loss= 26.85)

Rural non-

migrants

4,447 4.82 42.86

(Mean % gain

= 59.37)

57.14

(Mean % loss

= 36.09)

Rural-to-urban

in-migrants

and rural non-

migrants

4,970 8.20 45.05

(Mean % gain

= 61.44)

54.95

(Mean % loss

= 35.45)


41/42

41

Bibliography

Agesa, R.U. (2001). Migration and the Urban to Rural Earnings Difference: A

Sample Selection Approach,Economic Development and Cultural Change,

Vol. 49, No. 4, pp. 847-865.

Caldwell, J.C. (1968). Determinants of Rural-Urban Migration in Ghana,

Population Studies: A Journal of Demography, 22(3).

Coulombe, H. and A. McKay (2003), Selective Poverty Reduction in a

Slow Growth Environment: Ghana in the 1990s, Paper Prepared for the

World Bank, Human Development Network.

Falaris, E.M. (1987). A nested logit migration model with selectivity,International

Economic Review, Vol. 28, No. 2, pp. 429-443.

Ghana Statistical Service (GSS) (2000), Ghana Living Standards Survey Report of the

Fourth Round (GLSS 4), Accra, Ghana.

Hay, M.J. (1980). A Structural Equations Model of Migration in Tunisia,Economic

Development and Cultural Change, Vol. 28, No. 2, pp. 345-358.

Heckman, J.L. (1979). Sample selection bias as a specification error,Econometrica,

Vol. 47, No. 1, pp. 153-161.

Lanzona, L.A. (1998). Migration, self-selection and earnings in Philippine rural

communities,Journal of Development Economics, Vol. 56, pp. 27-50.

Lee, E.S. (1966). A Theory of Migration,Demography, Vol.3, No.1, pp.47-57.

Lee, L.-F. (1978). Unionism and wage rates: a simultaneous equations model with

qualitative and limited dependent variables,International Economic Review,

Vol. 19, No. 2, pp. 415-433.

Lee, L.-F. (1982). Some Approaches to the Correction of Selectivity Bias, The

Review of Economic Studies, Vol. 49, No.3, pp. 355-372.

Litchfield, J. and H. Waddington (2003), Migration and Poverty in Ghana: Evidence

from the Ghana Living Standards Survey, Sussex Migration Working Paper

no. 10. Sussex Centre for Migration Research.

Lucas, R.E.B. and O. Stark (1985). Motivations to Remit: Evidence from Botswana,

Journal of Political Economy, 93 (5), pp. 901-918.

Lucas, R.E.B. (1997). Internal Migration in Developing Countries. In M.R.

Rosenzweig and O. Stark (eds.),Handbook of Population and Family

Economics, Vol. 1B.


42/42

Migration and Household Income Differentials in the Volta Basin of Ghana: A

Sample Selection Approach, http://economics.ca/2004/papers/0199.pdf,

accessed on 29 December, 2004.

Nakosteen, R.A. and M. Zimmer (1980). Migration and Income: The Question of

Self-Selection, Southern Economic Journal, Vol. 46, Issue 3, pp. 840-851.

Pessino, C. (1991). Sequential migration theory and evidence from Peru,Journal of

Development Economics, Vol. 36, No. 1, pp. 55-87.

Ravenstein, E.G. (1885). The Laws of Migration,Journal of the Royal Statistical

Society, Vol.44, pp.420-23.

Sjaastad, L.A. (1962). The Costs and Returns of Human Migration, The Journal of

Political Economy, Vol. 70 (Supplement).

Todaro, M.P. (1969). A Model of Labour Migration and Urban Unemployment in

Less Developed Countries,American Economic Review, 59, pp.138-149.

Tunali, I. (2000). Rationality of migration,International Economic Review, Vol. 41,

No. 4, pp. 893-920.

Yap, L.Y.L. (1976). Rural-urban migration and urban underemployment in Brazil,

Journal of Development Economics, Vol. 3, No. 3, pp. 227-243.
http://economics.ca/2004/papers/0199.pdfhttp://economics.ca/2004/papers/0199.pdf

Migration between Ghana’s Rural and Urban Areas

Documents

Transcript of Migration between Ghana’s Rural and Urban Areas