Implications of Migration on Employment and Occupational Transitions in Tanzania

Abstract

The overarching message from the growth literature is that a transition from the low-productivity agricultural sector to the high-productivity manufacturing sector is necessary for structural change. Although sub-Saharan Africa has experienced substantial economic growth rates, rural–urban migration contributed very little to this progress. Migration to peri-urban areas may offer prospects for diversification out of agriculture with lower moving costs and job-search frictions than urban centers. We document migration patterns and worker selection into peri-urban and urban areas in Tanzania. Standard spatial classifications mask the prominent phenomenon of peri-urban (rather than rural) to urban migration. Lacking job experience and social networks, many youth moving to urban destinations are more likely to be unemployed. However, conditional on being employed over the two-year period of the study, migration to peri-urban as well as urban areas allows workers to transition from low- to high-valued occupations.

Keywords

internal migration peri-urban employment transitions Africa

Introduction

A major component of the rural–urban growth linkages literature is the study of migration. As a country develops, laborers migrate from predominantly agricultural areas to urban areas in search of higher wages in the industrial service sector (Harris and Todaro 1970). In theory, not only is the well-being of migrants improved, but the increase in land/labor ratios enables nonmigrants to raise their labor productivity and income. Rural economic growth, in turn, generates employment in the rural nonfarm sector, which consequently reinforces growth in the rest of the economy (Mellor 1976; Hazell and Haggblade 1990).

Much of the research on African labor markets suggests that the inability to create sufficient labor demand in the urban sector resulted in underwhelming urban growth linkages via migration.¹ The new economic geography literature contends workers will earn higher returns on their human capital investments (i.e., education) where higher-skilled individuals are clustered, inducing potential migrants to move to economic agglomerations (see, e.g., Hirshman 1958; Pred 1966; Lucas Jr. 1988; Moretti 2004; Ciccone and Peri 2006). There is little evidence in sub-Saharan Africa (SSA) that agglomeration economies are, in fact, inducing rural–urban migration (Jedwab and Vollrath 2015). Rather than moving directly to urban areas, whereby lack of roads, liquidity constraints, education deficits, and other factors pose significant barriers to migration (Stark and Bloom 1985; Stark 1991; Rozelle, Taylor, and de Brauw 1999; Wouterse and Taylor 2008), we hypothesize migrants are moving to take advantage of blossoming peri-urban (in addition to urban) areas in an attempt to transition from rural economies, but because of misclassification of urban and rural areas, part of the moves are often hidden in the data.

Recent work has underlined the need to evaluate the peri-urban space to understand structural transformations in emerging economies. Peri-urban areas create an interdependent environment between rural and urban areas, offering lower land rents and within-area transportation costs than urban areas, while also providing better linkages to raw materials and consumers from either spatial spectrum (Sharma 2016). These features of peri-urban locations foster conditions that are conducive to the expansion of the nonagricultural sector, whether through the creation of service sector employment or employment in the manufacturing of agricultural inputs or in the processing of raw commodities. The extent to which public and private investments in the agricultural sector are sufficient to strengthen backward or final-demand linkages, promoting jobs that attract rural workers to the peri-urban nonagricultural sector, has yet to be established. However, qualitative evidence suggests, at least from the perspective of the migrant, secondary towns are more likely to lead to poverty reduction given its relative appeal to workers (Ingelaere et al. 2017). Rural workers are attracted to secondary towns due to its physical and cultural proximity to rural villages (Ingelaere et al. 2017). Moreover, employment opportunities are less risky, given the likelihood of having job contacts closer to one’s origin and the lower costs of returning home if the job search is unsuccessful (Ingelaere et al. 2017). The above suggests an alternative structural transformation process, whereby peri-urban areas play a greater role in the transition from agriculture to the nonfarm sector.

We use a nationally representative survey to describe the landscape for which rural surplus labor may be converted into more productive forms of employment in peri-urban and urban areas of Tanzania. These transitions are often overlooked when using standard surveys, as individuals leaving villages are not tracked over time and as administrative boundary definitions fail to distinguish peri-urban from rural locations. We exploit the Living Standards Measurement Study–Integrated Surveys on Agriculture (LSMS–ISA; World Bank 2016) to quantify employment transitions over space in Tanzania, as the panel survey tracked and georeferenced the locations of respondents over time. Utilizing geographic information systems (GIS) that record the exact location of households, we develop a methodology of assigning origin and destination locations to rural, peri-urban, and urban classifications. Migration patterns are further defined by the distance traveled to understand the extent to which the transformation process follows similar patterns of regional interdependence observed in India (Sharma 2016), where the majority of villages were within 20 kilometers of a nearby town. The survey data combined with our spatial typologies are applied to examine the role of peri-urban areas in facilitating employment transitions out of agriculture.

We examine our hypothesis in two stages. First, we use descriptive statistics to determine the profile of migrants by origin and location choice (rural–peri-urban, rural–urban, peri-urban–peri-urban, and peri-urban–urban). Given Tanzania’s high urban growth rate (approximately 5.3 percent per year), it is estimated that by 2027, a greater share of Tanzanians will live in urban areas than rural areas (Morisset, Cunningham, and Haji 2014; World Bank 2017). In order to keep up with this fast pace of urbanization, the government of Tanzania has prioritized economic transformation in its long-term policy plans with the goal of increasing manufacturing and medium-technology industries (Newman et al. 2016). We characterize how the spatial allocation of the workforce will conform with these development objectives, by documenting rural to urban migration, in addition to rural to peri-urban and peri-urban to urban transitions, and noting variations in demographic and skill profiles of migrant workers.

Second, we assess whether migration to peri-urban (and urban) destinations improves employment prospects and occupational transitions out of agriculture in Tanzania. The number of businesses operating in urban areas has increased by approximately 15 percent per year, reflecting the growing urbanization witnessed over the last decade (Morisset, Cunningham, and Haji 2014). However, 93 percent of these businesses report two or less employees, and a majority are informal with little specialization. Kweka and Fox (2011) evaluated small, urban enterprises in Tanzania and report that a majority of the businesses formed due to “push” factors such as requiring additional income for family well-being rather than dynamic entrepreneurship seeking to fill business opportunities. Although alternative employment options in urban settings may be filling a gap for required family income needs, it is not clear that rural–urban (rather than rural–peri-urban) migrants will be contributing to the creation of new enterprises or the expansion of existing ones. In other words, for those rural out-migrants with prospects to diversify out of agriculture via moving to urban areas, we remain uncertain of their ability to secure employment and/or experience productivity gains.²

In the first stage, our descriptive statistics suggest that peri-urban to urban migration rates are three times higher than rural to urban migration rates in Tanzania. Furthermore, peri-urban to urban migrants (rather than rural–urban migrants) are more skilled in terms of having a secondary education. In addition, we find that migrants originating from both rural and peri-urban areas who travel above 20 and 50 kilometers are uniformly wealthier in terms of an asset index than nonmigrants originating from their respective areas. These findings bare two implications. First, skilled labor is more likely to originate from neighboring peri-urban areas. Second, liquidity tends to be an important factor in enabling migrants to overcome mobility costs (similar results have been reported in Bazzi (2017) and Munshi and Rosenzweig (2016)). Qualitative evidence provided by migrant interviews in Tanzania suggests the existence of liquidity constraints may also explain why the average rural worker in search of employment in peri-urban areas tends to migrate rather than commute on a regular basis (Ingelaere et al. 2017).

As mentioned above, in a second stage, we focus on whether migration patterns change labor market participation and facilitate transitions out of the agricultural sector, by exploiting rich information on the occupations and locations of migrants and nonmigrants over two periods of the panel survey. Propensity and nearest neighbor matching (NNM) approaches are applied to quantify the effects of migration on employment and occupational transitions while accounting for the nonrandom selection of workers into migration (McKenzie, Gibson, and Stillman 2010; Ham, Li, and Reagan 2011). Migrants are found to be more likely to move from low- to high-valued occupations in the nonagricultural sector. Yet hints of transition into unemployment also emerge, which can deter future migration.

The remainder of the article is organized as follows. The second section describes data sources as well as the analysis used to define rural, peri-urban, and urban locations of sampled households. The third section explains the empirical strategy. The fourth section discusses results and explains overall trends. The fifth section concludes.

Data Sources

We use a nationally representative panel survey collected in Tanzania (2008/2009, 2010/2011) in our analysis. The survey was administered to approximately 3,200 households. Diligent tracking of individuals and households over time allows for the development of a migration measure in wave 2 (hereafter referred to as the endline) and documentation of employment and occupational transitions from baseline to endline.

Migration and Spatial Classifications

We use the tracking of individuals and georeferenced information to create an individual permanent migration variable. An individual migrant is defined as a baseline household member who permanently left the household or left with his entire household by endline. The complexity of measuring migration patterns becomes immediately apparent when attempting to untangle urban definitions. For example, the urban definition in Tanzania follows the 2002 Population and Housing Census designations, whereby urban areas are localities defined as such by the district authority. With rapid urbanization and city expansion happening in Tanzania in the past decade, the urban definition from 2002 is likely to misrepresent the realities on the ground. We present an alternative classification method and include a peri-urban category that captures a transitional space between rural and urban domains. Among other challenges, analysis of household surveys based on official definitions of urban and rural areas would tend to underestimate moves out of rural and into the outskirts of cities (peri-urban), as administrative classifications often do not keep pace with urban expansion. A household moving out of a rural area and into a peri-urban area classified as rural would then be understood as a rural to rural migration move rather than a move out of a rural area and into the periphery of a city.

We categorize household location using GIS by adopting the concept of agglomeration as defined in the 2009 World Development Report on economic geography (World Bank 2009). In our implementation of the agglomeration index, location is defined using estimated travel time to cities of 20,000 or more inhabitants, population density, and percentage of built-up area. In this context, urban is defined as an area that has at least 150 people per square kilometer within one- hour travel time of a city of at least 20,000 people, and within a location that is classified as at least a 50 percent built-up area.³ Peri-urban areas follow the same criteria as urban areas with the exception that the built-up area can be less than 50 percent. Rural areas include households that are not defined within the urban and peri-urban categories (i.e., are not within one-hour travel time of a city of 20,000 people or do not meet a requisite density threshold of 150 people per square kilometer, regardless of built-up area values). Thus, each individual within their household is defined as rural, peri-urban, or urban in both waves of the survey, given that household location is recorded at baseline and individual movers are tracked (including a GPS location) at endline.

Figure 1 maps the rural, peri-urban, and urban locations covered by our survey in Tanzania in its right panel. The left panel of the figure provides an enlarged vision of Dar-es-Salaam to illustrate the different classifications with an example. The urban administrative boundary of the city (represented by the blue outline in the map) would mark the official urban–rural definitions, whereas the red and yellow areas mark the urban and peri-urban definition used in this article (and described above). It becomes apparent the degree of peri-urban space that exists outside of the urban boundary (shown in Figure 1 in yellow) that would normally be classified as rural based on administrative definitions. Figure 1 also illustrates that Tanzania has a large number of peri-urban areas and cities; however, given the country’s geographic size, connective infrastructure from city to city is limited, and therefore, the country is more defined by numerous, independent urban agglomerations.

Figure 1.

Peri-urban space and urban administrative boundary in Tanzania. Detail map depicts Dar es Salaam urban agglomeration.

Table 1 shows migration rates conditional on moving over the two-year period in rural and peri-urban areas. While rural–rural and peri-urban–peri-urban migration rates are quite high, these rates are sensitive to the assumption of what distance qualifies a person to be considered a migrant. We allow for different definitions based on distance-traveled thresholds to show how assumptions on what constitutes a migrant may affect our understanding of migration rates. Our first definition defers to a 1-kilometer threshold to acknowledge computing errors in the GPS coordinates. Interestingly, under this basic definition of migration, rural individuals are more likely to move to peri-urban (1.9 percent) than urban (1.0 percent) areas. In contrast, 3.4 peri-urban residents tend to move to urban areas.

Table 1.

Migration Rates by Origin and Destination Spatial Classifications.

Origin	Rate		Rate
Origin	Rural	N	Peri-urban	N
Destination
>1 km
Rural	9.37	382	2.53	42
Peri-urban	1.85	78	8.31	128
Urban	0.99	46	3.38	58
>10 km
Rural	6.25	244	2.06	36
Peri-urban	1.51	62	5.09	69
Urban	0.97	45	2.85	44
>20 km
Rural	5.66	216	1.64	30
Peri-urban	1.34	54	4.97	65
Urban	0.93	41	2.57	38
>50 km
Rural	4.92	185	1.27	23
Peri-urban	0.92	34	4.46	58
Urban	0.78	35	2.57	38
N	4,333		1,944

Note: Migrants are original household members who reside in a new household and live at least 1 kilometer from their original residence. Additional restrictions are imposed for various distances traveled: greater than 10 kilometers, greater than 20 kilometers, and greater than 50 kilometers. Migration rates are computed using the baseline survey sampling weights.

We next compare the migration rates with distance cutoffs above 10 kilometers, 20 kilometers, and 50 kilometers. Restricting by 10 kilometers generates similar migration rates, with the exception that we observe much lower rates of rural–rural and peri-urban–peri-urban migration. Long-distance moves (distances traveled above 20 and 50 kilometers) reduce the migration rates further. Rural to rural migration is between 4.9 and 5.7 percent compared to the original 9.4 percent. Rural to peri-urban migration rates are halved for the long-distance measure (from 1.9 to 0.9). In contrast, rural to urban migration is only slightly lower at a rate of 0.8, indicative of the distances traveled being much higher than to other destinations.

Perhaps an interesting result of this exercise is to highlight how standard classifications may misattribute the rate of urbanization to rural out-migration. Moving to a focus of long-distance moves obviates the implications of misclassifications. Peri-urban to urban migration rates triple observed rural to urban migration rates. Without adequate standardizations across space, many of the migration patterns considered rural to urban moves are likely falsely classified.

Employment Status and Sectoral Employment Outcomes

We monitor how migrants’ employment status and occupational mobility varies from those who remain in their original location. Here, we consider a person’s efforts over the last twelve months as recorded in the farm labor, wage labor, nonfarm enterprise, and education modules. We consider a person to be employed, that is, assign a value of one to a binary employment variable, if he or she engaged in one or more of the following activities in a given round: agriculture (on-farm or wage labor), nonagriculture (nonfarm enterprise or wage labor), or schooling. From this binary variable developed using information at baseline and endline, we create four binary employment transition variables: (i) whether a person was employed at baseline and endline, (ii) whether a person was employed at baseline and unemployed at endline, (iii) whether a person was unemployed at baseline and employed at endline, and (iv) whether a person was unemployed at baseline and endline.

To document sectoral transitions, we restrict the analysis to those who were actively engaged in agriculture, nonagriculture, or both at baseline and endline. This part of the analysis omits those who were attending school or unemployed at any point in the study, leaving us a smaller sample of migrants to monitor occupational transitions. A binary variable distinguishes those who were exclusively involved in agricultural work from those who diversified into nonagricultural work (accounting for exclusive nonagricultural labor and mixed forms of employment) in each round. We then evaluate the differences in four occupational transition binary variables among migrants and nonmigrants: employed in the agricultural sector at baseline and endline, employed in the agricultural sector at baseline and the nonagricultural sector at endline, employed in the nonagricultural sector at baseline and the agricultural sector at endline, and employed in the nonagricultural sector at baseline and endline.

Individual and Household Characteristics

To understand how individual worker attributes and household economic conditions influence migration, we create a common set of lagged (baseline) individual, demographic, and wealth characteristics across surveys. We include the individual’s gender, age, education (no schooling, primary school completion, secondary school completion, and missing education), and marital status (married, separated/divorced, widowed, and never married). Demographic and wealth dimensions of the household are captured using household size, number of working-age adults (aged fifteen to sixty-five) present in the household, and total landholdings owned (acres). We also construct an asset index using the first principal component of the following variables (Filmer and Pritchett 2001): whether the housing has a durable roof or wall, the size of the house (number of rooms), and whether the household has access to piped water or a flush toilet.

Empirical Strategy

Descriptive Analysis

In the time frame for which we observe migration, our aim is to first use the results from descriptive statistics to understand the age and skill profile of individuals drawn to specific locations (rural, peri-urban, and urban). We simply compare the baseline individual and household characteristic averages across the samples of nonmigrants and long-distance migrants distinguishing by spatial origin and destination classifications and distances traveled (>20 and 50 kilometers). T statistics are used to determine whether the differences in characteristic averages are meaningful in terms of statistical significance.

Matching Approach

The second stage of the analysis studies what forms of long-distance migration support the employment and occupational mobility of rural and peri-urban workers. We posit skilled workers who move longer distances, which is widely reflective of moves to urban locations, are more inclined to diversify into or work exclusively in the nonagricultural sector than skilled migrants moving to nearby areas. This is because the demand for nonagricultural goods and services is likely greater in population-dense and wealthier areas creating jobs outside of agriculture. Moreover, relative to the unskilled, there may be a greater tendency for the skilled to secure a nonagricultural wage occupation or initiate their own nonfarm enterprise at any destination.

To validate the above assertions with empirical evidence, we employ a matching approach to compare how the employment and occupational paths vary with long-distance migration, accounting for selection into migration. We evaluate the effects of multiple patterns of migration (rural–peri-urban, rural–urban, peri-urban–peri-urban, peri-urban–urban) distinguishing between moves over 20 and 50 kilometers separately. The four employment and four occupational transition outcomes referred to in Empirical Strategy section are used to quantify the effect of these migration patterns on one’s ability to maintain or gain employment and transition into a more productive nonagricultural sector. We are particularly interested in verifying whether long-distance migrants, who typically are assumed to be more skilled, have a greater probability of diversifying from agriculture or exiting agriculture altogether.

We apply the nearest neighbor and propensity score matching approaches to estimate the average treatment on the treated (ATT) effect of migration on the probability of engaging in each of eight employment and occupational transition outcomes (Rosenbaum and Rubin 1983, 1984; Abadie and Imbens 2008; Busso, DiNardo, and McCrary 2014). These techniques have been used to reduce the selection bias present in parameter estimates caused by inherent differences in migrants and nonmigrants. A distance metric or propensity score is used to match each migrant with a similar nonmigrant before calculating the impacts of migration on outcomes (McKenzie, Gibson, and Stillman 2010; Ham, Li, and Reagan 2011). In our example, the ATT effect of long-distance migration is computed by taking the average of the differences in outcomes calculated from each migrant and matched nonmigrant pair. Bias is reduced by purging the sample analyzed of nonmigrants who are most dissimilar than migrants in characteristics known to explain selection.

Although we offer estimates from two matching approaches to compare to estimates computed by simply subtracting the averages of the outcomes from the migrant and nonmigrant samples, we place greater emphasis on inferences based on the NMM approach for two reasons. First, the analytical standard errors for the NMM algorithm have been derived (Abadie and Imbens 2008). This means our inferences that are based on the standard errors are much more reliable with the NMM approach than the propensity score matching approach. Second, covariate matching estimators have been deemed superior to other matching estimators, particularly when overlap between the treatment and the control groups is limited. We are particularly concerned with this latter point when making comparisons made between rural–urban migrants and nonmigrants. In the Appendix, we graph the kernel density distributions of the propensity scores predicted from a logit model for each migration pattern, focusing on the definitions that restrict migration to those who travel more than 20 kilometers (Figures A1 –A4).⁴ Common support is weakest for the comparison made between rural–urban migrants and nonmigrants, which is partially attributable to the relatively low concentration of migrants in this category. Due to the limited sample size, we are unable to provide estimates that focus on areas of common support in the analysis. However, because of this weakness, we exercise caution in the interpretation of the rural–urban migration patterns.

Poor-quality matches can reinforce the bias that we attempt to eliminate (Morgan and Harding 2006). We try to improve the quality of our matches in three ways. First, the variables used in the construction of the distance and propensity score metrics are associated with selection into long-distance migration. We verify the individual and household characteristics statistically explain selection into long-distance migration in our descriptive analysis. We further force matching between migrants and nonmigrants who originated from the same district, to compare individuals exposed to similar economic factors that can influence migration and employment patterns but are unobservable to the econometrician (e.g., development projects that are complementary to the nonagricultural sector). Second, we match each migrant with a single (rather than multiple) nonmigrant to ensure that our estimates are based only on differences among the most similar nonmigrants according to the distance metric and propensity scores applied. Minimizing the potential for bias by focusing on one-to-one comparisons of outcomes is crucial for this study, given the relatively small sample sizes for some of the migrant categories. Third, we perform an additional bias adjustment to the calculated ATT effect of migration on employment transitions. Balancing tests performed using t statistics to test the statistical difference of covariate averages across treatment and control groups validate the procedure (Tables A2 –A5). We nevertheless estimate a regression to correct for any remaining imbalances after the matching procedure. In practice, this involves using the sample of migrants and matched nonmigrants in a regression that includes the migration indicator and variables used to construct the distance metric and propensity score as explanatory variables. We report the bias-corrected ATT, the parameter estimate on the migration indicator in the regression, controlling for the additional covariates to adjust for any remaining imbalances across pairs.

Results

Selection into Long-distance Migration

Table 2 presents the average values of the individual and household characteristics of survey respondents who resided in rural areas at baseline. We compare the average values across samples of nonmigrants and migrants by destination and distance traveled. The results from the t statistics testing the difference in the characteristic averages between the nonmigrant and migrant samples are also reported.

Table 2.

Summary Statistics of Rural Sample.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	Nonmigrant	Migrant	Migrant	Migrant	Migrant	(2)–(1)	(3)–(1)	(4)–(1)	(5)–(1)
Destination		Peri-urban	Peri-urban	Urban	Urban
Distance		>20 km	>50 km	>20 km	>50 km
Female	0.50	0.62	0.56	0.56	0.49	0.12*	0.06	0.05	−0.01
Primary education	0.51	0.59	0.56	0.55	0.62	0.08	0.05	0.04	0.11
Secondary education	0.03	0.07	0.10	0.04	0.00	0.04	0.08	0.01	−0.03***
Missing education	0.02	0.12	0.10	0.10	0.12	0.10**	0.08*	0.08*	0.10*
Married	0.35	0.28	0.36	0.09	0.08	−0.07	0.01	−0.26***	−0.28***
Separated or divorced	0.03	0.05	0.07	0.08	0.05	0.02	0.04	0.05	0.01
Widowed	0.03	0.09	0.06	0.05	0.00	0.05	0.03	0.02	−0.03***
Age ÷ 10	3.33	2.70	2.71	2.19	2.06	−0.62***	−0.61***	−1.13***	−1.26***
Age squared ÷ 100	0.13	0.09	0.08	0.05	0.05	−0.04***	−0.05***	−0.08***	−0.08***
Household size	6.75	6.12	5.99	7.01	7.31	−0.62	−0.76*	0.26	0.56
Number of adults	7.21	7.14	6.99	8.48	8.84	−0.07	−0.22	1.27*	1.63**
Owned land (acres)	7.22	5.15	3.75	3.79	4.12	−2.06	−3.47*	−3.43***	−3.10**
Asset index	−0.60	−0.47	−0.35	−0.05	−0.15	0.13	0.24	0.54***	0.44**
N	3,594	54	34	41	35

*p < .1.

**p < .05.

***p < .01.

The findings in Table 2 demonstrate that the profile of rural–peri-urban migrants compared to rural–urban migrants differs along a few dimensions. First, three percentage points fewer long-distance rural–urban migrants had a secondary education prior to their move than a rural nonmigrant. On average, the rates of having a secondary education are qualitatively greater for the other types of migrants. Rural–urban migrants are also younger and much less likely to be married. The average long-distance, rural–urban migrant is twenty-one to twenty-two years old compared to a rural–peri-urban migrant who is on average twenty-seven years old and a rural nonmigrant who is thirty-three years old. Contrary to the story of skilled individuals sorting to urban areas, our analysis suggests the opposite: rural–urban migrants are remarkably single, young, and unskilled.

In Table 3, we display similar statistics as shown in Table 2 for the peri-urban sample. Here, movers from peri-urban to urban areas are much more likely to have completed a secondary education. Thirty-two percent of migrants that traveled greater than 50 kilometers from a peri-urban to urban area are likely to have a secondary education compared to 9 percent of the nonmigrant peri-urban population. Peri-urban to urban migrants are also much younger than their peri-urban nonmigrant counterparts (twenty-five compared to thirty-four years old) but still slightly older than the average rural–urban migrant who was twenty-one years old.

Table 3.

Summary Statistics of Peri-urban Sample.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	Nonmigrant	Migrant	Migrant	Migrant	Migrant	(2)–(1)	(3)–(1)	(4)–(1)	(5)–(1)
Destination		Peri-urban	Peri-urban	Urban	Urban
Distance		>20 km	>50 km	>20 km	>50 km
Female	0.51	0.61	0.59	0.57	0.57	0.10	0.08	0.06	0.06
Primary education	0.61	0.60	0.63	0.41	0.41	−0.00	0.02	−0.20**	−0.20**
Secondary education	0.09	0.22	0.23	0.32	0.32	0.14**	0.14*	0.23***	0.23***
Missing education	0.02	0.05	0.02	0.09	0.09	0.02	−0.00	0.07**	0.07**
Married	0.31	0.27	0.27	0.22	0.22	−0.05	−0.04	−0.09*	−0.09*
Separated or divorced	0.04	0.02	0.02	0.05	0.05	−0.02	−0.02	0.01	0.01
Widowed	0.04	0.06	0.04	0.00	0.00	0.02	−0.00	−0.04***	−0.04***
Age ÷ 10	3.36	2.85	2.89	2.53	2.53	−0.51**	−0.47*	−0.83***	−0.83***
Age squared ÷ 100	0.13	0.09	0.10	0.07	0.07	−0.04**	−0.04**	−0.06***	−0.06***
Household size	6.29	6.85	6.82	6.30	6.30	0.56	0.53	0.02	0.02
Number of adults	7.77	7.84	7.82	7.87	7.87	0.07	0.05	0.10	0.10
Owned land (acres)	8.81	1.81	1.67	1.81	1.81	−7.00	−7.14	−7.00	−7.00
Asset index	0.23	0.63	0.71	0.96	0.96	0.39**	0.48**	0.73**	0.73**
N	1,609	65	58	38	38

*p < .1.

**p < .05.

***p < .01.

One effect that persists across samples is the importance of asset wealth in migration (Tables 2 and 3). Migrants are generally wealthier than nonmigrants whether comparisons are being made using the rural or peri-urban samples. Individuals that resided in peri-urban areas at baseline generally have a much higher baseline asset index value than the average values generated from individuals who resided in rural areas at baseline.⁵ However, when comparing the asset wealth values of individuals who originate from rural (or peri-urban areas), we surmise access to liquidity may be an important factor determining whether an individual can realistically incur the cost of moving long distance and risk entering an uncertain labor market environment.

Employment Transitions

The previous section demonstrated peri-urban individuals are positively selected into long-distance migration to urban areas, in their endowments of human capital and wealth. We next evaluate how long-distance migration aids workers in obtaining employment and/or exiting agriculture or diversifying from agriculture through the relocation to peri-urban and urban areas. By using the matching techniques described earlier, we compare the outcomes of migrants to those of nonmigrants most similar in demographic and location characteristics to mitigate positive selection bias on the ATT effects of migration on employment transitions.

Table 4 displays the employment transitions by long-distance migration status. There is no robust evidence that migration to peri-urban and urban areas increases the job prospects of rural or peri-urban inhabitants. Rather, the probability of long-distance migrants working continuously in baseline and endline rounds declines for all migrant types (rural–peri-urban migrants, rural–urban migrants, peri-urban–peri-urban migrants, and peri-urban–urban migrants). Focusing on the NNM estimates, the probability of maintaining employment given a rural–peri-urban (urban) move declines 20 for any distance threshold (24 and 32 for those who travel above 20 and 50 kilometers). Fewer migrants who move from peri-urban to peri-urban (urban) locations remain employed, amounting to a decline of 30 and 33 for distances traveled above 20 and 50 kilometers, respectively (33 for any distance cutoff).

Table 4.

Employment Transitions by Migration Status.

Employment Transition Outcomes		20 km				50 km
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	Stays	Moves	(2)–(1)	NNM	PWR	Moves	(6)–(1)	NNM	PWR
Sample: rural
Destination: peri-urban
Employed to employed	.88	.57	−.31***	−.20***	−.22**	.58	−.29***	−.20**	−.18
Employed to unemployed	.04	.28	.24***	.22***	.18**	.27	.23	.23***	.17
Unemployed to employed	.06	.07	.01	−.08*	−.06	.10	.04	−.06	.06
Unemployed to unemployed	.02	.08	.06	.06*	.10*	.04	.02	.03	−.04
N	3,533	47				30
Destination: urban
Employed to employed	.88	.58	−.30***	−.24***	−.23*	.53	−.34***	−.32***	−.34**
Employed to unemployed	.04	.25	.21**	.12**	.25***	.25	.21**	.11	.22*
Unemployed to employed	.06	.01	−.05***	−.04	−.15*	.02	−.05**	.01	−.05
Unemployed to unemployed	.02	.16	.14**	.16***	.13*	.20	.18**	.20***	.18**
N	3,533	33				27
Sample: peri-urban
Destination: peri-urban
Employed to employed	.79	.51	−.29***	−.30***	−.17**	.49	−.30***	−.33***	−.13
Employed to unemployed	.07	.34	.28***	.27***	.21***	.34	.28***	.27***	.15*
Unemployed to employed	.09	.06	−.03	−.01	−.10*	.07	−.02	.01	−.02
Unemployed to unemployed	.05	.09	.03	0.04	.05	.10	.04	.04	.00
N	1,555	63				57
Destination: urban
Employed to employed	.79	.46	−.33**	−.41***	−.14	.46	−.33**	−.41***	−.14
Employed to unemployed	.07	.29	.22***	.26***	.12	.29	.22***	.26***	.12
Unemployed to employed	.09	.11	.03	.07	−.02	.11	.03	.07	−.02
Unemployed to unemployed	.05	.14	.08	.08	.04	.14	.08	.08	.04
N	1,555	32				32

*p < .1.

**p < .05.

***p < .01.

Note: Inferences on mean differences and NNM estimates based on sampling weights. Mean differences account for enumeration area clustered standard errors. A single control observation is matched with each treated observation in NNM and PWR. NNM inferences based on robust standard errors using two observations. NNM forces matches of individuals within the same district at baseline. NNM = nearest neighbor matching; PWR = Propensity score weighted regression.

Migration to peri-urban and urban areas is more likely to lead to unemployment. For moves of 20 kilometers or above, we observe robust positive effects of each form of migration on the tendency to move from employment to unemployment. One explanation for these results may be the two-year transition period over which we observe employment outcomes may be too short for securing migrant employment in this economic environment particularly for youth with limited job experience and potential job contacts.⁶ A final striking result which is isolated to the case of rural–urban migration is that the probability of remaining unemployed over the two-year period is much higher, ranging from 16 for migrants of at least 20 kilometer to 20 to migrants of at least 50 kilometers (using NNM).

We next examine whether those who remain employed over the two-year period use migration to diversify out of the agricultural sector. Table 5 details the matching estimates for the effect of migration on occupational transition outcomes: agriculture to agriculture, agriculture to nonagriculture, nonagriculture to agriculture, and nonagriculture to nonagriculture. For brevity, the first word defines the occupation at baseline and the second word refers to the occupation at endline as described in Empirical Strategy section. We focus on long-distance moves defined by distance traveled above 20 kilometers to maximize our migrant samples but report results for the alternative migration definition in the Appendix (Table A6).

Table 5.

Sectoral Transitions by Migration Status (>20-kilometer Distances Traveled).

Sectoral Transition Outcomes	(1)	(2)	(3)	(4)	(5)
Sectoral Transition Outcomes	Stays	Moves	(2)–(1)	NNM	PWR
Sample: rural
Destination: peri-urban
Agriculture to agriculture	.69	.45	−.24	−.18	−.35
Agriculture to nonagriculture	.06	.12	.06	.12**	.02
Nonagriculture to agriculture	.20	.35	.15	−.01	.18
Nonagriculture to nonagriculture	.06	.08	.02	.07	.14
N	1,973	14
Destination: urban
Agriculture to agriculture	.69	.04	−.64***	−.67***	−.70***
Agriculture to nonagriculture	.06	.48	.43**	.48***	.33**
Nonagriculture to agriculture	.20	.04	−.16***	−.05	−.08
Nonagriculture to nonagriculture	.06	.43	.37**	.24	.44***
N	1,973	13
Sample: peri-urban
Destination: peri-urban
Agriculture to agriculture	.51	.15	−.36***	−.48***	−.33**
Agriculture to nonagriculture	.08	.10	.03	.01	.08
Nonagriculture to agriculture	.24	.27	.03	.11	.02
Nonagriculture to nonagriculture	.18	.47	.30	.36***	.23
N	593	21
Destination: urban
Agriculture to agriculture	.51	.00	−.51***	−.40**	−.68**
Agriculture to nonagriculture	.08	.41	.33**	.41***	.43*
Nonagriculture to agriculture	.24	.00	−.24***	.20**	—
Nonagriculture to nonagriculture	.18	.59	.42**	−.21	.25
N	593	8

*p < .1.

**p < .05.

***p < .01.

Recognizing our limited ability to make inferences based on such small sample sizes, the results in Table 5 qualitatively highlight the relevance of migration for diversifying out of the agricultural sector. Rural–urban, peri-urban–peri-urban, and peri-urban–urban migrants are much less likely to have worked in agriculture at baseline and endline compared to nonmigrants. Specifically, the likelihood of working in the agricultural sector over time declines by 67 for rural–urban migrants, 48 for peri-urban–peri-urban migrants, and 40 for peri-urban–urban migrants. Moreover, there is a marked shift of engagement out of the agricultural sector to the nonagricultural sector. The probability of rural–urban and peri-urban–urban migrants exiting the agricultural sector increases by 48 and 41. Overall, these findings are suggestive that long-distance migration to peri-urban and urban areas may be acting as a conduit for structural transformation in these countries. However, it is important to note that not all long-distance moves were fruitful since many migrants remained or were left unemployed.

Conclusion

In this work, we exploit the LSMS-ISA in Tanzania to document migration and employment patterns during the current economic transformation in SSA. Further disaggregating spatial classifications into rural, peri-urban, and urban classifications highlight the rising importance of rural to peri-urban migration in the allocation of skilled workers across space. Rural–peri-urban migration rates triple rural–urban migration rates. Moreover, rural–peri-urban migrants are much more positively selected in skills than rural–urban migrants.

Given the notion that peri-urban areas will increasingly serve as a hub for rural out-migrants, we attempt to provide evidence of the extent that any transformation process in Tanzania will be driven by the movement of the workforce to peri-urban and urban areas over time. We learn that liquidity constraints are likely to remain an obstacle for the skilled interested in moving long distances. If nonfarm enterprises are serving as the primary employer of migrants entering the nonagricultural sector in peri-urban and urban areas, then liquidity is necessary and sufficient for market entry (e.g., to purchase a motorbike or to establish a stall in the local market for commerce and retail).

Our ability to generalize how migration facilitates employment transitions is limited by the sampling frame of the panel data, where tracking a census of individuals would have been cost-prohibitive. A significant fraction of our migrant sample consists of youth who although might have been partially employed in a particular sector were also studying at baseline. For those migrants who were not students over the course of the study, migration appears responsible for a divergence in employment patterns. First, migrants to peri-urban and urban areas are more likely to move from employment to unemployment. Given the short period of time covered by our survey, this could indicate a short-term effect. However, it points to the need for policy that facilitates entry into local labor markets for migrants. Second, rural to peri-urban, rural to urban, and peri-urban to urban migrants who were able to secure employment were more likely to transition out of the agricultural sector. These figures suggest a different transformation process than the one that occurred in China, and an alignment with India’s economic growth pattern. As in India, nonfarm employment may serve as a primary driver of rural labor diversification in peri-urban areas. Standard rural panel surveys may miss an aspect of the transformation process by failing to track individuals over time and by failing to differentiate peri-urban locations from rural and urban classifications.

Footnotes

Appendix

Table A6.

Sectoral Transitions by Migration Status (>50-kilometer Distances Traveled).

Sample	Rural					Peri-urban
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Stays	Moves	(2)–(1)	NNM	PWR	Stays	Moves	(7)–(6)	NNM	PWR
Destination: peri-urban
Agriculture to agriculture	.69	.46	−.22	−.19	−.39	.79	.15	−.36***	−.43***	−.27*
Agriculture to nonagriculture	.06	.12	.07	.12**	.12	.07	.04	−.04	−.07***	−.04
Nonagriculture to agriculture	.20	.41	.21	.06	.40	.09	.30	.06	.11	.05
Nonagriculture to nonagriculture	.06	.00	−.06***	.00	−.12	.05	.51	.34***	.39***	.26*
N	1,973	8				1,555	19
Destination: urban
Agriculture to agriculture	.69	.00	−.69***	−.60***	−.72***	.51	.51	.00	−.40**	−.68**
Agriculture to nonagriculture	.06	.56	.50**	.56***	.62**	.08	.08	.41	.41***	.43*
Nonagriculture to agriculture	.20	.00	−.20***	−.12	−.28	.24	.24	.00	.20**	—
Nonagriculture to nonagriculture	.06	.44	.38**	.15	.38	.18	.18	.59	−.21	.25
N	1,973	9				593	8

***p < .01.

**p < .05.

*p < .1.

Acknowledgment

This work was funded by UK DFID through the MDTF for Sustainable Urban Development TF071544. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, nor those of the Executive Directors of the World Bank nor the governments they represent. All errors are our own. We thank Channing Arndt and participants of the STAARS Conference in Addis Ababa for their feedback on the initial version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Abadie

Imbens

. 2008. “On the Failure of the Bootstrap For Matching Estimators.” Econometrica 76:1537–57.

Azam

J. P.

Lesueur

J. Y.

. 1997. “Efficiency-wage and Supervision: Theory and Application to the Ivorian Manufacturing Sector.” Journal of African Economies 6 445–62.

Bazzi

2017. “Wealth Heterogeneity and the Income Elasticity of Migration.” American Economic Journal: Applied Economics 9:219–55.

Beegle

De Weerdt

Dercone

. 2011. “Migration and Economic Mobility in Tanzania: Evidence from a Tracking Survey.” Review of Economics and Statistics 93:1010–33.

Bigsten

Soderbom

. 2006. “What have We Learned from a Decade of Manufacturing Enterprise Surveys in Africa?” World Bank Research Observer 21:241–65.

Busso

DiNardo

McCrary

. 2014. “New Evidence on the Finite Sample Properties of Propensity Score Reweighting and Matching Estimators.” Review of Economics and Statistics 96:885–97.

Ciccone

Peri

. 2006. “Identifying Human Capital Externalities: Theory with Applications.” Review of Economic Studies 73:381–412.

Fafchamps

Minten

. 2001. “Social Capital and Agricultural Trade.” American Journal of Agricultural Economics 83:680–85.

Fafchamps

Söderbom

. 2006. “Wages and Labour Management in African Manufacturing.” Journal of Human Resources 41:346–79.

10.

Filmer

Pritchett

. 2001. “Estimating Wealth Effects without Expenditure Data—or Tears: An Application to Educational Enrollments in States of India.” Demography 38:115–32.

11.

Ham

Reagan

. 2011. “Matching and Semi-parametric IV Estimation, A Distance-based Measure of Migration, and the Wages of Young Men.” Journal of Econometrics 161:208–27.

12.

Harris

Todaro

. 1970. “Migration, Unemployment and Development: A Two-sector Analysis.” American Economics Review 60:126–42.

13.

Hazell

P. B.

Haggblade

. 1990. “Rural-Urban Growth Linkages in India.” Working Paper Series 430. Washington, DC: World Bank.

14.

Hicks

Joan Hamory

Kleemans

Marieke

Nicholas Y.

Miguel

Edward

. 2017. “Reevaluating Agricultural Productivity Gaps with Longitudinal Microdata.” NBER Working Paper No. 23253.

15.

Hirshman

A. O.

1958. The Strategy of Economic Development. New Haven, CT: Yale University Press.

16.

Ingelaere

Christiaensen

De Weerdt

Kanbur

. 2017. “Why Secondary Towns Can Be Important for Poverty Reduction: A Migrant’s Perspective.” International Growth Centre Working Paper No. C-40300-TZA-1.

17.

Jedwab

Vollrath

. 2015. “Urbanization without Growth in Historical Perspective.” Explorations in Economic History 58:1–21.

18.

Kweka

Fox

. 2011. “The Household Enterprise Sector in Tanzania: Why it Matters and Who Cares.” World Bank Policy Research Working Paper No. 5882.

19.

Lucas

R. E.

Jr . 1988. “On the Mechanics of Economic Development.” Journal of Monetary Economics 22:3–42.

20.

McKenzie

Gibson

Stillman

. 2010. “How Important Is Selection? Experimental vs. Non-experimental Measures of Income Gains from Migration.” Journal of the European Economic Association 8:913–45.

21.

Mellor

1976. The New Economics of Growth. Ithaca, NY: Cornell University Press.

22.

Moretti

2004. “Estimating the Social Return to Higher Education: Evidence from Longitudinal and Repeated Cross-sectional Data.” Journal of Econometrics 121:175–212.

23.

Morgan

Harding

. 2006. “Matching Estimators of Causal Effects: Prospects and Pitfalls in Theory and Practice.” Sociological Methods and Research 35:3–60.

24.

Morisset

Cunningham

Haji

. 2014. Tanzania Economic Update: Who Wants a Job. The Magnetic Power of Cities. Tanzania Economic Update. Accessed online at December 15, 2017 at: http://documents.worldbank.org/curated/en/393381468122079213/pdf/PRIORITY00000005th0Issue01900602014.pdf.

25.

Munshi

Rosenzweig

. 2016. “Networks and Misallocation: Insurance, Migration, and The Rural-urban Wage Gap.” American Economic Review 106:46–98.

26.

Newman

Page

Rand

Shimeles

Söderbom

Tarp

. . Eds. 2016. Manufacturing Transformation: Comparative Studies of Industrial Development in Africa and Emerging Asia. Oxford, UK: Oxford University Press.

27.

Pesaresi

Ehrilch

Florczyk

A. J.

Freire

Julea

Kemper

Soille

Syrris

. 2015. GHS Built-up Grid, Derived from Landsat, Multitemporal (1975, 1990, 2000, 2014) European Commission, Joint Research Centre (JRC). PID. Accessed December 1, 2015. http://data.europa.eu/89h/jrc-ghsl-ghs_built_ldsmt_globe_r2015b.

28.

Pred

A. R.

1966. The Spatial Dynamics of US Urban-Industrial Growth, 1800–1914: Interpretive and Theoretical Essays. Cambridge, MA: MIT press.

29.

Rosenbaum

Rubin

. 1983. “The Central Role of the Propensity Score in Observational Studies for Causal Effects.” Biometrika 70:41–55.

30.

Rosenbaum

Rubin

. 1984. “Reducing Bias in Observational Studies Using Subclassification on the Propensity Score.” Journal of the American Statistical Association 79:516–24.

31.

Rozelle

Taylor

J. E.

de Brauw

. 1999. “Migration, Remittances, and Agricultural Productivity in China.” American Economic Review 89:287–91.

32.

Sharma

2016. “Urban Proximity and Spatial Pattern of Land use and Development in Rural India.” The Journal of Development Studies 52:1593–611.

33.

Sharma

Das

. (2017). “Migrant Networks in the Urban Labor Market: Evidence from India.” Journal of Development Studies. doi: https://dx-doi-org.web.bisu.edu.cn/10.1080/00220388.2017.1342815.

34.

Söderbom

Teal

Wambugu

. 2004. “Does Firm Size Really Affect Earnings?” Development and Comp Systems Paper 0409011. St. Louis, MO: EconWPA.

35.

Stark

1991. The Migration of Labor. Boston, MA: Basil Blackwell.

36.

Stark

Bloom

. 1985. “The New Economics of Labor Migration.” American Economic Review 75:173–78.

37.

World Bank. 2009. World Development Report 2009: Reshaping Economic Geography. Washington, DC: World Bank. Accessed December 1, 2015. https://openknowledge.worldbank.org/handle/10986/5991.

38.

World Bank. 2016. Living Standards Measurement Study—Integrated Surveys on Agriculture, Tanzania. Accessed December 1, 2015. http://go.worldbank.org/OOLZL0UIR0.

39.

World Bank. 2017. World Bank Development Indicators. Accessed online November 8, 2017. Available at: https://data.worldbank.org/data-catalog/world-development-indicators.

40.

Wouterse

Taylor

J. E.

. 2008. “Migration and Income Diversification: Evidence from Burkina Faso.” World Development 36:625–40.