Internal migration in Cameroon, Ghana, Rwanda and Zambia: How big and how beneficial? Abstract: This paper looks at inward migration propensities for four SSA countries and analyzes the extent to which households benefit from inter-regional moves. The paper documents that only about 2 percent of the working-age population migrates for job-related reasons across locations in Cameroon, Rwanda and Zambia, with even less migrating within Ghana. These migration rates pale against those in more developed countries. Based on household consumption and individual income regressions, households in Ghana that migrated over the prior 12 months consume between 11 and 15 percent more than households that have not moved; the corresponding figures for Cameroon are 0-10 percent and 4-5 percent in Rwanda (income) and Zambia. 2 I. INTRODUCTION A number of theories have argued that an important route out of poverty is migration from poorer to richer areas. This supposition is the basis for the Harris-Todaro model in which a gap between rural and expected urban earnings drives migration. However, the analysis of internal migration among Sub-Saharan African (SSA) countries is very much in its infancy. Little information exists on the extent of economic migration within SSA countries let alone whether individuals benefit financially following inter-regional moves although the revealed preference argument would assume that they do benefit. Recently, analysis has been conducted on the strength of structural transformation among SSA countries and the research finds that individuals have been moving out of sectors with low average productivity (agriculture) to sectors with higher levels of productivity (manufacturing and services). A new employment database has been made available based on household surveys for about 30 African economies (see Fox and Thomas, 2013). In figure 1 this dataset is combined with sector output levels in constant prices to generate sector labor productivity estimates and these are plotted against changes in sector employment shares. Points in the lower left quadrant indicate sectors with below average productivity and declining employment shares, while those in the upper right quadrant indicate sectors with above average productivity and rising employment shares. Figure 1. Selected SSA Countries: Labor Productivity and Changes in Employment Shares 2000–10 (In percent) 5 Agriculture Industry Services Change in relative employment share 4 3 NGA 2 1 GHA CMR MOZ UGABFA SEN ZMB 0 -1 CMR GHA SEN GHA RWA -2 SEN TZA RWA ZMB UGA CMR MOZ ZMB TZA RWA BFA UGA NGA MOZ BFA 1 1.5 TZA -3 -4 NGA -5 -2 -1.5 -1 -0.5 0 0.5 Sectoral labor productivity relative to average 2 2.5 3 The figure shows that workers have moved out of the low productivity agricultural sector into services and some industry although in some cases the agricultural share has failed to decline (Cameroon and Mozambique). This pattern is consistent with that observed in other regions of the world. However, the annual employment changes are quite small in Sub-Saharan Africa and we do not know the extent to which it is accomplished through internal migration versus differential population growth rates between urban and rural areas. This paper seeks to address this question by estimating internal migration propensities based on household surveys for four countries: Cameroon, Ghana, Rwanda and Zambia. These countries were chosen based on the inclusion of migration modules in their most recent household surveys. Little has been written on migration propensities for SSA countries. Indeed, estimates of internal migration only exist for South Africa and indicate that the propensity to migrate is highest among black households, but even among these households the propensity is only about 10 percent among adults 15 years and older. For the whole South African population, the percentage of households with at least one migrant worker is about 17-18 percent so with the average household size at 4.5 persons, this translates into a migration propensity of about 4 percent (Posel and Cassale, 2003). This internal migration propensity is comparable to results found for Indonesia and India although less than for China at 6 percent and far less than for Malaysia and Vietnam at over 20 percent (Wilson et al. 2012). This paper also analyzes the extent to which households benefit from inter-regional moves. Since the data do not track the same households over time, the effect of a migratory move is assessed by adding a dummy variable for those that moved over the past twelve months to a standard equation of household consumption/income determinants. Other work that has looked at financial benefits resulting from an inter-regional move include Beegle et al. (2011) who track the same households in Tanzania over a 13 year period. They find that migration added about 3 percentage points per annum to consumption growth over this period. Consumption and income are used as the measures of living standards and both have pluses and minuses. Since surveys can only hope to measure financial flows over a short period, consumption has advantages over income for a number of reasons. it is less volatile the concept of consumption is clearer to survey participants because the majority of people in low income countries (LIC) do not receive any income during their lifetime: they are paid in kind or are unremunerated employees in unincorporated family enterprises. 4 respondents are generally more reluctant to share information about their income than about their consumption. Since income is usually taxable, it may be hard for respondents to be persuaded that their income information will not be passed to the tax authorities. As a drawback, focusing on household consumption misses a lot of information because the determinants of consumption are generally based on the characteristics of the household head rather than on the whole family. Since the income concept is based on individuals, individual characteristics can be used to explain its movements. The basic idea of the paper is that standard consumption/income determinants capture the fixed characteristics of individuals and the migration dummy captures the added benefit of a move. Of course, the dummy variable also captures unobserved attributes of those that move and therefore one cannot distinguish whether the consumption/income effect of migration captures higher consumption in the new location because of the move or specific attributes of the movers that are not reflected in the other determinants of consumption. A partial control for specific characteristics of migrants is obtained by comparing the coefficient of heads of households and individuals that moved for job reasons versus the coefficient of those that moved for non-job related reasons. Only about 2 percent of the working-age population migrates for job-related reasons across locations in Cameroon, Rwanda and Zambia, with even less migrating within Ghana (1½ percent of the population). These migration rates do not compare favorably with much higher rates in more developed countries (e.g. 5-6 percent migration rate for inter-county moves in Canada and the USA and 20 percent in Malaysia and Vietnam). In terms of the financial benefit from migrating, households in Ghana that migrated over the prior 12 months consume between 11 and 15 percent more than households that have not moved; the corresponding figures for Cameroon are 0-10 percent and 4-5 percent in Rwanda (income) and Zambia. II. DATA AND INTERNAL MIGRATION PROPENSITY The analysis is based on two household surveys for Ghana (1998, 2005), Cameroon (2001, 2007), Rwanda (2001 and 2005) and Zambia (2004, 2010). The household samples consist of about 6000-9000 for Ghana, around 11000 for Cameroon and 17000 for Zambia. The individual sample of wage earners for Rwanda is about 2500 in 2001 and 5000 in 2005. All surveys use a two-stage stratified sampling approach in which primary sampling units (census areas) are first defined and then subdivided in a second stage into a specific number of households per area. The migration modules include questions on whether an individual has always lived in this village/town and if the answer is no, the question is asked about the timing of the locational move. In order to focus on recent moves, we identify moves that took place during the past 12 months. 5 Figure 2. Migration propensities for a sample of SSA and Asian countries 25 20 15 10 5 0 Vietnam Cambodia India Indonesia Philippines Malaysia South Africa Rwanda Zambia Ghana Cameroon Sources: SSA household surveys and Wilson and others (2012) Figure 2 shows that except for Rwanda, the propensity to migrate in the most recent survey for the sample of SSA countries is low at between 2-5 percent per annum, and much lower than corresponding totals for most low income (Cambodia and Vietnam) and middle income countries in Asia (Malaysia, Philippines). However, they are comparable to the migration rates in India and Indonesia. It should be noted that these migration rates include all types of migratory moves and not only those associated with the labor market. For the SSA sample, it is possible to identify the reason for the move (job related or otherwise). The propensity of the unemployed to move is comparable between Cameroon, Ghana, Rwanda and Zambia at between 0.3-0.7 percent per annum whereas the propensity of the employed to move is more variable across countries. It is currently lowest in Ghana at about 0.7 percent per annum, and slightly higher in Cameroon, Rwanda and Zambia at between 1.2-1.4 percent per annum (Table 1). It appears therefore for Rwanda that migratory movements for job reasons are much less than for other reasons. 6 Table 1. Population weighted propensity to migrate Cameroon 2001 2007 Employed 1.4 1.2 Unemployed 0.7 0.8 Ghana 1 1998 2005 1.4 Zambia 2004 2010 Rwanda 2001 2005 0.7 2.3 1.25 0.4 1.5 0.7 0.5 0.3 0.3 0.5 Notes: 1 In the Ghana 1998 survey no distinction is made between those searching for work and those moving to new job opportunities III. DETERMINANTS OF CONSUMPTION AND INCOME To assess whether the welfare gain is higher for those that migrate relative to those that stay behind, we estimate a standard equation of the determinants of household consumption for Cameroon, Ghana and Zambia and individual incomes for Rwanda. The variables included in the consumption equations are age of household head, gender, education level, employment status, sector of employment, dummy for urban or rural location and regional dummy variables. The consumption aggregate includes home produced and purchased food and nonfood expenditure. The non-food component includes clothing and footwear, housing (imputed services), electricity, water and gas, furnishings, health, transport, communication, recreation, education and miscellaneous goods. Households with heads of households aged between 16 and 65 are analyzed, closely corresponding to the ILO definition of the working age population. The individual-based income dataset for Rwanda is a little richer than the other datasets. The most recent salary after deducting taxes is the concept of income that is used in the Rwanda dataset and since individuals record their incomes on a daily/weekly/monthly/annual basis, all estimates are expressed as a monthly aggregate. The list of income determinants has many commonalities with the consumption variables except that individual rather than characteristics of the household head are used. Moreover, the income dataset includes additional variables for on-the-job experience, vocational and firm-specific training, incidents of repeat schooling, non-wage employment benefits, firm size and hours worked. 7 A. Consumption determinants The explanatory power for the consumption regressions is high (R squared estimates between 0.63 and 0.72) and compares favorably with log consumption regressions for other regions. 1 Household size has a positive coefficient which is significantly below unity indicating that larger families consume more but at a declining rate, likely because more children are represented in the household. The age variable is significantly positive indicating a positive age-earnings profile up to 65. Since the variable is expressed in logarithms, the consumption benefit levels out as the head of household ages. Male heads of household consume a little more in Ghana and Zambia but the difference is negligible for Cameroon (Table 2). The consumption benefit of employment is positive in all three countries with household data. It is considerably higher in Ghana and Zambia (between 15-20 percent) than in Cameroon (about 5 percent) and its effect has increased over time in Ghana and Zambia. The coefficient corresponds to non-government workers in the service sector because separate coefficients are estimated for agricultural, manufacturing and government workers. Families with household heads employed in agriculture consume significantly less than families whose household heads are employed in other sectors but this disparity has declined over time. Indeed, in Zambia, families of agricultural workers now consume as much as those of non-government service workers. Perhaps surprisingly, household consumption of heads of household employed in manufacturing and mining has deteriorated over time in both Cameroon and Ghana so that household heads employed in this sector now earn less than those employed in non-government services in both countries. The consumption premium for government sector employees is consistently positive in Cameroon and Zambia whereas in Ghana, families whose head is a government employee consume less than families whose head is employed in non-government services. Turning to education characteristics of the household head, a distinct rising educationconsumption profile is identified. The education-consumption profile has become steeper over time in Ghana (across the board) and in Cameroon (at the bottom end) but has narrowed in Zambia. The range of log consumption coefficients between household heads with primary and college level education in Ghana has risen from 0.05-0.3 in 1998 to 0.09-0.68 in 2005 with the biggest increase occurring for those with college level or equivalent education. In Cameroon the range below upper secondary schooling has increased from 0.05-0.14 to 0.10.19 and, in Zambia, while the education coefficients have declined, their levels have converged to those in the other two countries. 1 Unweighted estimates are shown in the paper but population weighted estimates are comparable except where indicated. 8 Table 2. Log Household Consumption Determinants Ghana 1998 2005 2001 Cameroon 2007 Zambia 2004 2010 Household size (log) 0.31 *** 0.37 *** 0.29 *** 0.29 *** 0.17 *** 0.28 *** Age (log) Experience (log) Firm size (log) Hours worked (log) Male head of household 0.09 *** 0.13 *** 0.2 *** 0.18 *** 0.04 *** 0.12 *** 0.03 *** 0.03 ** 0.01 0.02 * Employment dummy Agriculture sector dummy Manufacturing sector dummy Government sector dummy 0.12 *** 0.02 *** 0.04 *** 0.63 *** 0.09 *** 0.05 *** 0.06 *** 0.27 *** 0.06 *** 0.13 *** 0.06 *** -0.17 *** 0.05 *** 0.16 *** 0.04 *** -0.13 *** -0.03 ** 0.18 *** 0.07 ** 0.15 *** -0.04 *** -0.01 0.03 *** 0.12 *** 0.03 *** 0.06 *** -0.38 *** 0.02 -0.14 ** -0.11 *** 0.35 *** -0.04 0.05 *** 0.14 *** 0.29 *** 0.6 *** 0.1 *** 0.19 *** 0.31 *** 0.6 *** 0.05 ** -0.07 *** 0.13 *** 0.01 0.47 *** 0.23 *** 1.03 *** 0.76 *** 0.03 0.3 *** 0.75 *** 1.08 *** 0.11 *** 0.47 *** 0.95 *** 1.89 *** training vocational training primary school repeat secondary school repeat 0.15 ** 0.07 ** 0.19 *** -0.04 ** -0.13 ** other benefits public holidays retiree benefits 0.1 0.22 *** 0.18 *** 0.12 *** 0.19 *** 0.44 *** Primary schooling Lower secondary schooling Upper secondary schooling College/nursing/teacher training Urban dummy 0.15 *** 0.19 *** -0.24 *** -0.2 *** 0.01 -0.08 *** 0.02 -0.1 *** 0.01 Rwanda 1/ 2001 2005 0.05 *** 0.11 *** 0.29 *** 0.3 *** 0.24 *** 0.09 *** 0.17 *** 0.38 *** 0.68 *** 0.23 *** 0.09 *** 0.1 *** 0.18 *** 0.11 *** 0.21 *** 0.29 *** 0.15 *** 0.05 * 0.04 0.04 0.02 0.05 * Employed migrant (<12 months) Employed migrant (>12 months) Unemployed migrant Migrant (other reasons) -0.06 0.11 * 0.08 *** -0.01 0.03 0.03 * -0.01 -0.05 0.04 -0.01 0.03 Consumption of lowest quartile -0.97 *** -1.02 *** -0.81 *** -0.76 *** 0.1 -0.01 0.1 *** 0.04 -1.34 *** -1.13 *** -0.2 *** 0.12 *** 0.04 -0.03 -1.44 *** -1.03 Diagnostic statistics Number of observations 5144 7280 10001 10416 15320 17864 2653 5025 R-squared 0.68 0.68 0.66 0.69 0.63 0.67 0.7 0.72 Notes: 1/ For Rwanda, the dependent variable is the log of monthly wage income ***,**,* indicate statistical significance at the 99, 95 and 90 percent levels 9 The regression includes regional dummy variables (not shown), but even with these included, the dummy variable for urban locations is strongly significant at about 20 percent in all three countries. The large negative coefficient on the consumption of the lowest quartile shows that characteristics alone are unable to replicate the large differences between consumption levels for those in the bottom quartile versus those in higher quartiles.2 B. Income Determinants For Rwanda, most of the income coefficients have the same sign as the consumption coefficients for the other countries. The two exceptions are the dummy variables for manufacturing and public sector employment. The manufacturing wage premium is very large in Rwanda in 2005 and may reflect the changing nature of production in this country with greater demand for manufacturing workers. The coefficient on public sector employment is negative in contrast to the positive coefficient for most of the other countries. The main reason for this is that the income equation includes variables for non-wage benefits and these are positive and highly correlated with employment in the public sector. When the non-wage benefit variables are excluded from the specification the public sector dummy variable becomes positive and significant in the regression using 2005 data. In general, coefficient estimates are considerably larger in the income equation. This is especially true for the education variables with the coefficient on university education almost twice as large as any university education coefficient in the consumption equations. It appears that incomes are more unequal than consumption aggregates and this phenomenon is consistent with those earning wage incomes saving more than the rest. The larger education premiums could also reflect a closer mapping between individual education characteristics and individual wages than between household head characteristics and household consumption. The gender premium is also very large in Rwanda although it has declined over time, consistent with the work of Ezemenari and Wu (2005) An advantage of the analysis of income over consumption is that it is based on individual data and the explanatory variables associated with it are richer. The two variables that capture whether individuals have attended short-term training courses or apprenticeships are significantly positive suggesting that training has been successful in raising income levels. Similarly, variables that capture whether individuals were required to repeat an education year at primary or secondary level are also significant but negative so that repeating a year may indicate a slow learner that is reflected in worker performance. Finally, variables that capture wage income supplemented by non wage benefits appear to signal remunerative jobs 2 When separate coefficients for the characteristic variables for those in the bottom quartile of the consumption distribution are included most are insignificant and only provide a little additional explanatory power. 10 since the presence of non-wage benefits, paid holidays and retirement benefits are associated with higher incomes. A few other individual and firm attributes deserve mention. The log of experience is significantly positive so that staying in the same firm/job provides learning by doing benefits that are reflected in higher wages. Larger firms offer higher incomes and more monthly hours worked also raises incomes although the coefficient is far below unity and pretty stable over time. IV. CONSUMPTION AND INCOME BENEFITS FROM MIGRATION We now turn to the second major strand of the paper and identify whether migrants are associated with higher consumption/incomes in their new location. To make this assessment, migrants are distinguished based on whether the change in location was job related or not. Ghana For Ghana, the question is asked whether the move took place over the past 12 months or earlier with results differing according to the timing of the move. 3 Since no distinction is made between employed and unemployed labor moves in the Ghana 1998 questionnaire, the insignificant coefficient for job related moves could reflect the fact that many moves were those of unemployed workers (Table 2). This is supported by the insignificant coefficient on unemployed migrants in the 2005 sample. 4 Interestingly, migrants in the 1998 sample who had moved at least one year before the sample was taken experienced almost a 10 percent increase in household consumption while the results are reversed in the 2005 sample. For the most recent year, job moves within the previous 12 months are associated with a 10-15 increase in household consumption (depending on whether population weights are used) while moves that took place at least one year prior show no increase in consumption. A possible explanation for the difference in results between the two years for Ghana is that the country may have been in a different cyclical position in 1998 versus 2005 making it take longer to find suitable employment in 1998. This view is partially supported by looking at the historical profile of Ghana’s growth rate. The initial survey took place between April 1998 and March 1999 and during this time Ghana’s growth rate was decelerating to its cyclical trough of 4.2 percent in 2000, a decline of about 1 ½ percentage points from its peak in 2007 (see figure 3). The year 2005, by contrast, was a cyclical peak with the economy registering a 3 4 For Cameroon and Zambia, the question on place of residence is only asked for the previous 12-month period. Separating out unemployed migrants who later found work from those that remained unemployed has no bearing on the results because the coefficients of both groups are insignificant, mainly relating to the small sample size. 11 growth rate of 6.2 percent. While the different cyclical positions of the economy provide some support for the increased difficulty of finding suitable employment in the late 1990s, the annual differences in growth rates are not that large compared to the cyclical growth variability among advanced economies. Figure 3. Real GDP growth for Cameroon and Ghana 8 Ghana 6 4 2 0 -2 Cameroon -4 -6 -8 1990 1992 1994 1996 1998 2000 2002 2004 2006 Sources: IMF World Economic Outlook Since we are not tracking the same households over time, it is not possible to distinguish whether migrants receive higher consumption in the new location because of job opportunities or because they have individual features that differ from the rest of the population. A partial control for this can be gained by comparing the coefficient on migrants who move for non-job related reasons versus those who move for job reasons. In the 1998 sample, the coefficient on migrants who move for reasons other than employment is significantly positive at 0.03. This supports the view that migrants have some identifying characteristics that make them different from the rest of the population. However, this coefficient is considerably below the coefficient of those that moved for job related reasons so that job moves do give rise to higher incomes. The difference is even larger using the 2005 sample since the coefficient on non job movers is zero. Cameroon For Cameroon we can only focus on moves over the past 12 months. In 2001, those that moved because of a job transfer received 10 percent higher household consumption compared to those that did not move while those that moved for non job related reasons 12 received 4 percent more than non-moving households. In 2007, those that moved because of a job transfer received 5 percent higher household consumption compared to those that did not move although this difference was not present in the weighted regression. 5 The other migration variables are insignificant in the 2007 sample. Zambia The migration coefficients for Zambia are comparable to those of the other two countries in terms of the unemployed and non work-related migrants. Although the coefficient on the unemployed migrants is quite large in the 2004 sample (0.1) it is insignificant, similar to the findings for Cameroon and Ghana. In the 2010 sample the coefficient is almost zero. In the 2004 sample, a significant consumption premium of about 10 percent is identified for non work-related migrants. This falls and becomes insignificant in the 2010 sample. Finally, in contrast to the other two countries, the premium for workers that move for employmentrelated reasons is insignificant in both surveys although positive at 4 percent. Rwanda Rwanda is the only country where the unemployed migrants earn significantly less than the rest of the population in the destination region, although this is only true for the earlier period. For the other migrants, the profiles are similar to that of Zambia. Migrants moving for non-job reasons receive an income premium in 2001 but not in 2005 while the premium for job movers is fairly constant at 2-5 percent. It may be thought that the benefit of migrating would be higher for a rural-urban or urbanurban move given the tendency for incomes to be higher in urban areas. However, when the sample is restricted to those that live in urban areas, the coefficients are comparable to those of the full sample (results not shown). The main difference is that the coefficients on the education variables are considerably higher for the urban sample than for the whole population. Although the consumption of unemployed migrants does not change after a move, most have changed employment status as a result. Between 90 and 95 percent of those moving to find work actually obtain work within one year of the move (Table 3). Moreover, except for Rwanda in 2001, it does not appear that unemployed migrants were forced to take work below what their individual characteristics would suggest. Rather, given that most migrants join the workforce after a move, they benefit from the employment premium. 5 Both the coefficients on employed migrants and non-job related migrants are significantly positive in a regression that excludes the education variables, but once these are introduced the migrant consumption premiums vanish. 13 Employed Unemployed Table 3. Employment Status of Unemployed Migrants (in percent of population weighted sample) Ghana Cameroon Zambia 1998 2005 2001 2007 2004 2010 98.7 92.2 89.5 94.7 92.5 93.5 1.3 7.8 10.5 V. 5.3 7.5 6.5 Rwanda 2001 2005 90 86.6 10 CONCLUSION This paper has shown that controlling for standard consumption determinants that proxy fixed characteristics of individuals, households in Ghana that migrated over the prior 12 months earn between 11 and 15 percent more than households that did not move. The corresponding figures for Cameroon are 0-10 percent and 4-5 percent in Rwanda and Zambia. This finding shows that economic welfare improves for those individuals that move for job related reasons. While some of this welfare improvement likely captures unobserved attributes of those that moved, the consumption premium received by job movers outweighs the premium received by non-job movers in Ghana and Cameroon and Rwanda (in the most recent period) although not in Zambia. The paper also documents internal job-related migration rates for the four countries and finds that they are all small at less than 1½ percentage points per annum. This rate does not compare favorably with much higher rates in more developed countries and countries in South East Asia with similar initial economic structures (Vietnam). The share of agriculture in employment has fallen by over 20 percentage points between 1995 and 2010 in Vietnam and its high internal migration rate has been a key factor in this reduction. Indeed, if we assume that population in the agricultural region grows by 2 ½ percent per annum while the rest of the population grows by 2 percent per annum, we can mimic the 20 percent decline in the agricultural employment share assuming a migration propensity of 20 percent. If we now translate these assumptions into an African context, a migration rate of 2 percent per annum would keep the agricultural employment ratio fairly constant. Therefore, the extent of internal migration can have an important bearing on the relative speeds of structural transformation across countries. Understanding the low internal migration rates in SubSaharan Africa would be an interesting topic for future research. 13.4 14 REFERENCES Beegel, K. J. de Weerdt, and S. Dercon, 2012, “Migration and economic mobility in Tanzania: evidence from a tracking survey,” forthcoming in Review of Economics and Statistics Day, K. and S. Winer, “Interregional Migration and Public Policy in Canada,” mimeo Department of Human Resources Canada, July 2001 Ezemenari, K. and R. Wu, “Earnings differences between men and women in Rwanda,” World Bank Africa Region working paper no. 81, January 2005 Fox, L., A. Thomas, C. Haines, and J. Huerta-Munoz, 2013 “Africa has work to do: Employment prospects in the new century,” forthcoming IMF working paper IMF, Regional Economic Outlook for Sub-Saharan Africa, September 2012 Molloy, R., Smith, C. and A. Wozniak, 2011, “Internal migration in the US: updated facts and recent trends,” Federal Reserve Board, USA Posel, D. and D. Casale, 2003, “What has been happening to internal labour migration in South Africa, 1993-1999?”, South African Journal of Economics, Vol 71,3, pp. 455-479 Wilson, E., Jayanthakumaran, K., and R. Verma, 2012, “Demographics, Labor Mobility and Productivity,” ADB Institute working paper no. 387, October