Context

Migration patterns from Senegalese regions result from a combination of historical events and the changing demographic landscape, notably the growing prominence of the Dakar region. In the early 20th century, the first Senegalese migrants comprised sailors, traders, and demobilized soldiers.Footnote ⁵ Sailors were driven to depart due to the decline in traffic on the Senegal River, often finding employment in the merchant navy and the French war navy [Robin (Reference Robin2000)]. However, significant international emigration did not truly emerge until the 1960s, spurred by the establishment of recruitment centers in Senegal by the French Office National de l'Immigration, in collaboration with French companies and supported by the French Ministry of Labor and Population. This move was prompted by a labor demand surge in the French automobile industry. They recruited extensively from the Senegal River valley, the Tambacounda region, and Casamance, including all départements located south of the Gambia. During this period, rural Senegal was grappling with a series of droughts [Robin (Reference Robin1996)].

In terms of African destinations, Senegalese migrant farmers and traders were drawn to countries like Ivory Coast and Ghana, which were experiencing economic success. These migrants also sought opportunities in Gabon, Zaire, and Cameroon. However, the crisis in groundnut cultivation, compounded by successive droughts from the 1970s onwards, led to a significant increase in international emigration towards Europe. Until the 1980s, a majority of these migrants originated from the Senegal River valley and increasingly settled in France, despite the formal end of immigration programs in 1974. France's immigration policy in 1975 and 1976 shifted traditional labor immigration, predominantly composed of single men, towards family immigration [Robin (Reference Robin2000)].

Since the 1980s, the capital, Dakar, has become a significant hub for international emigration, particularly due to its rising share in the national population. Although there has been a gradual increase in the proportion of young adults in Dakar who have taken some initial steps toward international emigration (reaching one-third during 2000–2008), their actual likelihood of emigrating from Dakar remained relatively stable until the late 2000s, at around one in ten young adults [Lessault and Flahaux (Reference Lessault and Flahaux2013); Beauchemin et al. (Reference Beauchemin, Flahaux and Schoumaker2020)]. Starting in the 1990s, Senegalese emigration dynamics were also influenced by emigration from the Diourbel, Thiés, and Louga regions (formerly known as Baol, Cayor, and Djambour). African destinations gradually declined to the benefit of a more diversified range of high-income and transit countries.Footnote ⁶ Senegalese emigration to the United States dates back to the early 1980s and reflects this shift away from Africa. Additionally, two new destinations, Italy and Spain, gained prominence [Robin (Reference Robin1996)].

Data sources

Our dataset is derived from the 2013 Senegalese census, which encompasses the entire population, totaling approximately 13.2 million individuals. This dataset is comprehensive and includes a wealth of individual-level attributes, such as age, educational attainment, birthplace, and current residence, among others. Our analysis specifically focuses on individuals aged 15 and older, categorizing them into two skill groups: college graduates (comprising individuals with tertiary/higher education degrees) and the less educated. In doing so, our primary emphasis lies on the accumulation of upper-tail human capital, a crucial factor in modeling disparities in growth and productivity across countries and time periods [Mokyr and Voth (Reference Mokyr and Voth2009); Castelló-Climent and Mukhopadhyay (Reference Castelló-Climent and Mukhopadhyay2013)].

Senegal's administrative structure comprises four levels of divisions: 14 régions, 45 départements, 133 arrondissements, and 548 communes or municipalities. Our quantitative analysis in Section 4 is conducted at the département level. This choice is primarily driven by the availability of income data stratified by education level and its relevance in formalizing residential mobility between administrative units, as well as spatial disparities in access to tertiary education. It is worth noting that commuting flows between départements are generally limited, with the exception of the Dakar region, where départements tend to be smaller in size, and the transportation infrastructure is more developed.

The Senegalese census offers a crucial feature by providing comprehensive information regarding households’ migratory backgrounds and exposure to international migration. This is achieved through two key elements:

• • Measurement of Internal Migration: The census includes detailed inquiries about each respondent's department of birth. By comparing birthplaces and current residences, we can precisely gauge the stock of internal migrants between départements over the course of their lives. We adopt the concept of lifetime migration, considering individuals who live in a département different from their birth département. This approach views migration as a life-course trajectory, focusing on individuals’ long-term location choices. As of 2013, Senegal had 1,546,378 internal migrants, constituting approximately 24.1% of the working-age population. Among these, 471,527 internal migrants had relocated within the last five years. Notably, this group includes 106,747 college graduates, making up roughly 47.2% of the college-educated working-age Senegalese population.

• • Tracking International Emigration: The census features questions about household members who have left the country in the past five years. This information allows us to compile data on migration flows, categorized by place of residence, destination country, and education level. Between 2008 and 2013, the census recorded 150,370 international emigrants of working age, with 11.1% being college graduates. We utilize this data on emigrant flows to estimate the composition of the stock of international Senegalese migrants by départements of origin. More specifically, we use the five-year migration flows to calculate the skill-specific proportions of emigrants to OECD countries by department of origin. Then, we apply these proportions to the total stock of Senegalese migrants residing in an OECD member state in 2015. Data on immigration to OECD countries are sourced from Arslan et al. (Reference Dumont, Arslan, Kone, Moullan, Ozden, Parsons and Xenogiani2015), which provides insights into the characteristics of 311,066 Senegalese migrants.

For reference, Table A1 in the appendix contains the country codes utilized throughout this paper, along with descriptive statistics regarding the population structure for the 45 Senegalese départements.

Stylized facts

The 2013 census enables us to approximate the skill-specific stocks of lifetime internal and international migrants, categorized by département of origin and département or country of destination. Aggregating dyadic mobility at the région level, Figure 1 illustrates the lifetime movements of international (top panel) and internal (bottom panel) migrants, with a focus on college graduates in panels (a) and (c) and the less educated in panels (b) and (d).

Figure 1. Stock of international and internal migrants by education level, département of origin and destination, (a) International – College graduates, (b) International – Less educated, (c) Internal – College graduates, (d) Internal – Less educated.

The pivotal role of Dakar as a primary source of international migrants and a major destination for internal movers is prominently depicted. This is particularly pronounced among college-educated migrants. A substantial majority of highly skilled international migrants originate from the Dakar region, closely followed by the neighboring Thiés region, and they predominantly reside in OECD member states. In the realm of internal migration, Dakar emerges as the predominant destination, drawing migrants from across the country. For both college-educated international and internal migration, the largest régions of origin include Thiés, Saint-Louis, Ziguinchor, Diourbel, Kaolack, and Fatick.

Turning our attention to less educated migrants, the Dakar region remains the primary region of origin. However, owing to the persistence of historical migration routes used by Senegalese sailors, traders, and soldiers in the early 20th century, other traditional regions of emigration, such as Matam, Tambacounda, and Saint-Louis along the Senegal River, contribute significantly to the total stock of international migrants. Concerning low-skilled internal migrants, Dakar hosts 349,610 of them, while Thiés and Diourbel are also attractive regions, collectively attracting 377,275 movers.

These disparities in the magnitude of lifetime movements are partially attributed to variations in native population size among regions. Notably, the four départements of the Dakar region (Dakar, Rufisque, Pikine, and Guédiawaye) constitute 18% of the Senegalese adult population but account for 45% of the college-educated population. To better illustrate the diversity in exposure to international and internal movements, we rely on skill-specific emigration rates, which define the emigration stock as a percentage of the corresponding native population. The spatial distribution of these emigration rates reflects discrepancies in the international and internal connectivity of high- and low-skilled populations. Figure 2 maps these disparities in connectivity between départements.

Figure 2. International and internal emigration rates by education level and département of origin, (a) College graduates, (b) Less educated, (c) College graduates, (d) Less educated.

Regarding international migration (top panels), the highest emigration rates for college-educated individuals are observed in five départements: Dakar, Guédiawaye, Pikine, Rufisque (all within the Dakar region), and Kanel (in the Matam region). Moderate emigration rates are also noted in the less affluent départements of Mbacké, Matam, Goudiry, and Bakel. Other regions exhibit limited international connectivity. For less-educated individuals, the highest emigration rates are observed in the same départements and in the Senegal River regions bordering Mali and Mauritania, which still maintain relatively strong international connections. Nevertheless, these rates are considerably lower than those for college-educated individuals.

Shifting our focus to internal movements (bottom panels), the distribution of high-skilled emigration rates is more evenly spread across the country, although the eastern départements exhibit lower internal connectivity, in contrast to international migration patterns. The département of Salémata, in the Kédougou region, stands out with high emigration rates. Conversely, the distribution of emigration rates for low-skilled individuals suggests a lack of internal connectivity in many départements. Except for Salémata, distance from Dakar appears to be a reliable predictor of this isolation.

RUM structure

– Migration and education decisions are modeled as outcomes of a RUM with two sources of heterogeneity between individuals [as in Delogu et al. (Reference Delogu, Docquier and Machado2018); Cha'ngom et al. (Reference Cha'ngom, Deuster, Docquier and Machado2023)] – in the ability to acquire higher education and in preferences for various destinations. The utility of an individual λ born in region i, choosing education type-s and moving to a destination j is given by:

(1)$$U_{ij, s}^{\lambda} = \left[\ln \omega_{i}^{\lambda} + \ln \left(1-{e_{s}^{\lambda }\over G_{i}}\right)\right] + \left[\ln w_{\,j, s} + \ln \left(1-c_{ij, s}\right) + \varepsilon_{ij, s}^{\lambda }\right],\; $$

where the first term in brackets denotes the pre-migration utility, which depends on the income of individual λ ($\omega _{i}^{\lambda }$), the heterogeneous effort required to acquire college education ($e_{h}^{\lambda }$), and a scale factor that proxies the mean access to education in region i (G _i).Footnote ⁷ No effort is required if the individual does not acquire higher education (i.e. $e_{l}^{\lambda } = 0$). The individual cost to acquire higher education $e_{h}^{\lambda }$ is distributed on [0,  1] according to the following cumulative distribution function: $F_{2}( e_{h}) = e_{h}^{z + 1}$, where $z \in {\mathbb R}^{ + }$ is a parameter governing the slope of the density function and the elasticity of human capital to the expected return to higher education. If z = 0, F ₂ is the uniform distribution; the greater z, the lower the fraction of individuals with low education costs (i.e. z is a proxy for the scarcity of talent).

The second term in brackets denotes the post-migration utility, which depends on the average income at destination (w _j,s), the average level of mobility costs (c _ij,s), and a random component ($\varepsilon _{ij, s}^{\lambda }$) that captures heterogeneity between individuals in preferences, in moving costs, in the ability to value work-related skills and experience in a different location, etc. As is standard in the literature dealing with migration, we assume that the random component of utility $\varepsilon _{ij, s}^{\lambda }$ follows a Type I Extreme Value distribution with a dispersion parameter μ. As discussed below, the inverse of this parameter is the elasticity of dyadic migration to the income differential between origin and destination countries, and is a substitute for the elasticity of utility to income. In this setting, location decisions are governed by a multinomial logit expression.

Timing and expected utility

The timing of decisions reflects the availability of information about the two random components of utility. In the first stage, individuals make higher education decisions before discovering their migration type, $\varepsilon _{ij, s}^{\lambda }$, but they know its distribution. Assuming perfect expectations about w _j,s and c _ij,s, each individual decides to acquire higher education if the expected utility gain from being educated exceeds the cost. Given the distributional assumptions on $\varepsilon$, the expected level of maximum utility of type-s individuals born in department i is given by McFadden (Reference McFadden1974):

(2)$${\mathbb E}\left(U_{i, s}^{\lambda }\right) = \left[\ln \omega_{i}^{\lambda} + \ln \left(1-{e_{s}^{\lambda }\over G_{i}}\right)\right] + \ln \sum_{\,j = 1}^{{\mathbb J}}\left(w_{\,j, s}\right)^{1/\mu }( 1-c_{ij, s}) ^{1/\mu},\; $$

where the first (pre-migration) term is known by the individual when making education decisions, while the second term is the unconditional expected value of maximum utility in the post-migration period, also termed log-sum or inclusive value of the underlying multinomial logit model.

The latter term includes the utility in the home location ($i \in {\mathbb J}$). Hence, for individuals born in department i, the expected value of post-migration maximum utility (i.e. the second term in Eq. (2)) can be rewritten as:

(3)$$\Omega_{i, s} = \ln \left[( w_{i, s}) ^{1/\mu } + ( W_{i, s}) ^{1/\mu } + ( W_{i, s}^\ast ) ^{1/\mu } \right],\; $$

where $( W_{i, s}) ^{1/\mu }\equiv \sum _{j\in J \setminus \{ i\} }( w_{j, s}) ^{1/\mu }( 1-c_{ij, s}) ^{1/\mu }$ is the expected utility component related to internal mobility prospects, and $( W_{i, s}^\ast ) ^{1/\mu }\equiv \sum _{j\in J^\ast }( w_{j, s}) ^{1/\mu }( 1-c_{ij, s}) ^{1/\mu }$ is the component related to international mobility prospects for type-s individuals. In a no-migration economy (NM), these two terms are nil, implying that $\Omega _{i, s}^{NM} \equiv ( 1/\mu ) \ln w_{i, s}$.Footnote ⁸ In a context with mobility, the influence of internal and international mobility prospects is large if the levels of W _i,s and W _i,s* are high in comparison with w _i,s. This is the case when wages in alternative destinations are high, and moving costs are low. The level of Ω_i,s is a proxy for the average welfare of type-s individuals born in i, as argued below.

In the second stage, after the education decision is implemented, individuals discover their migration type, $\varepsilon _{ij, s}^{\lambda }$, and decide where to emigrate, or to stay in their home country. Given heterogeneous preferences, the dyadic emigration rate from department i to destination j is defined as:

$$m_{ij, s} = {\mathbb P} \left[\ln w_{\,j, s} + \ln \left(1-c_{ij, s}\right) + \varepsilon_{ij, s} = \max_{k\in {\mathbb J}} \ln w_{k, s} + \ln \left(1-c_{ik, s}\right) + \varepsilon_{ik, s} \right].$$

Assuming regional characteristics (w _i,s and G _i) and dyadic mobility costs (c _ij,s) are exogenous, we characterize below the solution for the three variables of interest (H _i, m _ij,s, and h _i).

Human capital formation (H _i)

For individual λ investing in college education is optimal when ${\mathbb E}( U_{i, h}^{\lambda }) \geq {\mathbb E}( U_{i, l}^{\lambda })$. Given Eq. (2), this condition holds if the cost of acquiring higher education is not too large:

(4)$$e_{h}^{\lambda }\leq G_{i}\left[{\Lambda _{i}-1\over \Lambda _{i}}\right],\; $$

where Λ_i is the expected “return” to higher education accounting for wage rates in all possible destinations, weighted by their accessibility. It can be expressed as:

(5)$$\Lambda _{i} \equiv {\exp{( \Omega_{i, h}) }\over \exp{( \Omega_{i, l}) }} \equiv {( w_{i, h}) ^{1/\mu} + ( W_{i, h}) ^{1/\mu} + ( W_{i, h}^\ast ) ^{1/\mu}\over ( w_{i, l}) ^{1/\mu} + ( W_{i, l}) ^{1/\mu} + ( W_{i, l}^\ast ) ^{1/\mu}},\; $$

where the key variables W _i,s and $W_{i, s}^\ast$ $\forall s$ capture the components of Λ_i driven by internal and international mobility. In a no-migration economy, the return to higher education is determined by the ratio of local wage rates (Λ_i ≡ (w _i,h/w _i,l)^1/μ). In an economy open to emigration, the expected return to higher education increases (resp. decreases) with wage rates, and accessibility to all alternative destinations for the highly educated (resp. less educated).

Given our distributional assumptions on e ^λ, it follows that:

(6)$$H_{i} = G_{i}^{1 + z}\left[{\Lambda _{i}-1\over \Lambda _{i}}\right]^{1 + z} .$$

Hence, the model has desired properties in line with the existing literature. Migration opportunities stimulate pre-migration human capital formation, H _i, if the education premium is greater in alternative destinations than in the region of birth (i.e. W _i,h/W _i,l or ${W_{i, h}^\ast }/{W_{i, l}^\ast }$ are greater than w _i,h/w _i,l), in line with Mountford (Reference Mountford1997), Stark et al. (Reference Stark, Helmenstein and Prskawetz1997), Vidal (Reference Vidal1998), Beine et al. (Reference Beine, Docquier and Rapoport2001, Reference Beine, Docquier and Rapoport2008). In addition, the greater G _i, the more a rise in skill premium in alternative destinations stimulates H _i, in line with Stark and Wang (Reference Stark and Wang2002) and Djajić et al. (Reference Djajić, Docquier and Michael2019).

Internal and international mobility

Once the education decision has been made, individuals discover their migration type, $\varepsilon _{ij, s}^{\lambda }$, and make the choice between emigrating to another location or remaining in their home country. Assuming a Type I Extreme Value distribution, the likelihood that a type-s individual born in region i relocates to destination j is determined by the multinomial logit expression:

$$m_{ij, s} = {( w_{\,j, s}) ^{1/\mu }( 1-c_{ij, s}) ^{1/\mu }\over ( w_{i, s}) ^{1/\mu} + ( W_{i, s}) ^{1/\mu} + ( W_{i, s}^\ast ) ^{1/\mu}},\;$$

which implies that internal and international emigration rates for type-s individuals from department i are given by

$$\eqalign{m_{i, s} & = \sum_{\,j \in J \setminus \{ i\} }m_{ij, s} = {( W_{i, s}) ^{1/\mu}\over ( w_{i, s}) ^{1/\mu} + ( W_{i, s}) ^{1/\mu} + ( W_{i, s}^\ast ) ^{1/\mu}} \cr m_{i, s}^\ast & = \sum_{\,j \in J^\ast }m_{ij, s} = {( W_{i, s}^\ast ) ^{1/\mu}\over ( w_{i, s}) ^{1/\mu} + ( W_{i, s}) ^{1/\mu} + ( W_{i, s}^\ast ) ^{1/\mu}}.}$$

The ratios of these emigration rates determine the level of positive selection observed in internal mobility (ρ _i ≡ m _i,h/m _i,l, which increases with W _i,h/W _i,l) and international migration ($\rho _{i}^\ast \equiv {m_{i, h}^\ast }/{m_{i, l}^\ast }$, which increases with ${W_{i, h}^\ast }/{W_{i, l}^\ast }$).

Once again, the model exhibits desired properties consistent with empirical findings. The greater W _i,s or $W_{i, s}^\ast$, the greater skill-specific emigration rates, m _i,s or $m_{i, s}^\ast$. The greater W _i,h/W _i,l or ${W_{i, h}^\ast }/{W_{i, l}^\ast }$, the greater the selection ratio, ρ _i or $\rho _{i}^\ast$, in line with [Grogger and Hanson (Reference Grogger and Hanson2011); Belot and Hatton (Reference Belot and Hatton2012); Kerr et al. (Reference Kerr, Kerr, Özden and Parsons2016)]. Furthermore, in accordance with Cha'ngom et al. (Reference Cha'ngom, Deuster, Docquier and Machado2023), mobility-driven expected utility shocks (ΔW _i,s or $\Delta W_{i, s}^\ast$) induce a positive correlation between human capital formation (H _i) and the ratio of emigration rates (ρ _i or $\rho _{i}^\ast$). For example, shocks that increase the expected utility of college graduates in an alternative destination (W _i,h or $W_{i, h}^\ast$) have a positive effect on human capital formation (H _i) and on the positive selection of internal or international migrants (as reflected by the ratio of high-skilled to low-skilled emigration rates, ρ _i or $\rho _{i}^\ast$; see e.g. [Shrestha (Reference Shrestha2017); Theoharides (Reference Theoharides2018); Abarcar and Theoharides (Reference Abarcar and Theoharides2021); Khanna and Morales (Reference Khanna and Morales2021)]). Shocks that increase the expected utility of the less-educated in a different region or abroad (W _i,l or $W_{i, l}^\ast$) have a negative effect on both variables [(e.g., McKenzie and Rapoport, Reference McKenzie and Rapoport2011; de Brauw and Giles, Reference de Brauw and Giles2017; Pan, Reference Pan2017; Kosack, Reference Kosack2021)]. This establishes the micro-foundations for the link between mobility prospects and pre-migration human capital formation in a dyadic framework encompassing both internal and international locations. In contrast, local expected utility shocks (Δw _i,s) induce a negative correlation between H _i and ρ _i or $\rho _{i}^\ast$.

Lastly, it is important to highlight that under the assumption of a Type I extreme-value distribution for $\varepsilon$, the conditional distribution (conditional on the selected alternative) and the unconditional distribution of maximum post-migration utility are equivalent, regardless of the chosen alternative [de Palma and Kilani (Reference de Palma and Kilani2007)]. This implies that:

(7)$$\Omega_{i, s} = {\mathbb E}\left[\ln w_{\,j, s} + \ln \left(1-c_{ij, s}\right) + \varepsilon_{ij, s} \vert U_{i, j, s} = \max_{k \in {\mathbb J}} U_{i, k, s} \right]\quad \forall j,\; $$

motivating the choice of Ω_i,s as a proxy for welfare.

Human capital accumulation (h _i)

The post-migration share of college graduates in the regional labor force can be expressed as the ratio of college-educated non-migrants to total non-migrant populations, adjusted for the number of immigrants (I _i,s):

(8)$$h_{i} \equiv {( 1-m_{i, h}) H_{i}N_{i} + I_{i, h}\over ( 1-m_{i, h}) H_{i}N_{i} + I_{i, h} + ( 1-m_{i, l}) ( 1-H_{i}) N_{i} + I_{i, l}},\; $$

which increases with the proportion of remaining college graduates, (1 − m _i,h)H _i, and decreases with the proportion of remaining low-skilled workers, (1 − m _i,l)(1 − H _i). For a given stock of immigrants ($I_{i, s}\equiv \sum _{j\neq i} m_{ji, s}N_{j, s}$), mobility-driven expected utility shocks affecting region i (ΔW _i,s or $\Delta W_{i, s}^\ast$) induce ambiguous effects on post-migration human capital accumulation as for a given H _i, h _i decreases with positive selection, in line with Beine et al. (Reference Beine, Docquier and Rapoport2001, Reference Beine, Docquier and Rapoport2008, Reference Beine, Docquier and Rapoport2010) and Cha'ngom et al. (Reference Cha'ngom, Deuster, Docquier and Machado2023).

Parameterization

We parameterize our model to precisely align with the data on skill-specific wage rates, dyadic emigration stocks, and the size of the labor force by education level observed in the 45 départements of Senegal in 2013. To estimate skill-specific wages, we use labor income data from the Senegalese Labor Force Survey. We calculate the average yearly earnings of college graduates and less-educated workers in each département, adjusting for workers’ participation rates. We then convert these wage rates into USD PPP, using the exchange rate from the WDI. In Senegal, college graduates earn, on average, approximately three times more than less-educated individuals. For the OECD destination option, we compute a population-weighted average GDP per worker for each skill group on the sample of countries that host immigrants from Senegal.

With regard to migration, we employ data on dyadic emigration stocks (M _ij,s), encompassing the population of non-movers (M _ii,s), categorized by education level as described in Section 2. In our set of destinations, we distinguish between the 45 departments of Senegal, a unified foreign entity comprising all OECD member states, and another entity aggregating the rest of the world, primarily encompassing contiguous African countries. Using equation (??), the ratio of movers to non-movers is determined by M _ij,s/M _ii,s = (w _j,s/w _i,s)^1/μ(1 − c _ij,s)^1/μ. We observe wage ratios across all pairs of destinations and assume an elasticity of bilateral migration to the wage ratio equal to 1/μ = 1/0.7 (in line with Bertoli and Moraga (Reference Bertoli and Moraga2013)). Consequently, dyadic migration costs (c _ij,s) are calculated as residuals to precisely match the ratio of movers to non-movers from the data.

As a validation exercise, Table 1 demonstrates that the calibrated levels of internal migration costs are positively correlated with the geodesic distance between départements, and this effect remains relatively consistent across skill groups. Moving costs for college graduates decrease with cultural proximity and the mean distance to roads within the départements. While only weakly significant, moving costs for less-educated individuals show a positive correlation with access to cities, which may imply that improved mobility within départements for low-skilled workers, including rural-to-urban migration, reduces the necessity to relocate to different locations. Mobility costs exhibit a negative correlation with the average income per worker at the department level, with this effect being more pronounced for college graduates. This suggests that migration costs are more manageable when local economic conditions at the origin are relatively favorable.

Table 1. Validation of calibrated internal migration cost

Notes: OLS regressions. Standard errors are robust to heteroskedasticity and clustered at the department level.

Figure 3 provides a visual representation of the calibrated levels of international and internal migration costs. For the sake of clarity, internal migration costs are specifically depicted for migration to Dakar. In many respects, these costs closely align with disparities in emigration rates, as illustrated in Figure 2.

Figure 3. Calibrated migration costs (m _ij,s) by département, (a) International – College graduates, (b) International – Less educated, (c) Internal – College graduates, (d) Internal – Less educated.

In terms of international migration costs, they are usually substantial for college-education individuals, with the exception of the Dakar region and Kanel in the Matam region. To a lesser extent, some departments such as Saint-Louis and Louga in the Saint-Louis region, Thiés and M'bour in the Thiés region, Foundiougne in the Fatick region, as well as Ziguinchor, Sédhiou, and the eastern part of the country also experience lower migration costs for college-educated individuals. However, except for Dakar, Guédiawaye, and Pikine, individuals with lower levels of education face significantly higher international migration costs.

Shifting focus to internal migration costs, college graduates generally encounter relatively low costs within the entire country, except in departments located in the north of the country (Dagana and Podor), in the Kaffrine region, in Koumpentoum, and in border areas with The Gambia (Bounkiling, Médina Yoro Foulah). In contrast, internal migration costs for less-educated individuals are notably higher, except in departments like Thiés, Djourbel, and Fatick, as well as in Ziguinchor, Sédhiou, and Salémata.

Regarding the education technology, we determine two unknown parameters, denoted as z and G _i, to fit data on emigration stocks and human capital levels. The parameter z dictates the sensitivity of pre-migration human capital levels to the prospects of selective migration. Cha'ngom et al. (Reference Cha'ngom, Deuster, Docquier and Machado2023) calibrate this parameter to correspond with the semi-elasticities empirically estimated for four broad country income groups. We set z = 3.8, which corresponds to the long-run semi-elasticity they estimated for the lower-middle income country group, amounting to 3.2.Footnote ⁹ With z established, the scaling variable G _i is subsequently calibrated as a residual from equation (6) to match the observed pre-migration human capital levels, H _i, evident in the data.

In a validation exercise, Table 2 demonstrates that the calibrated values of G _i exhibit a positive correlation with the proportion of the population residing in urban areas, the average income per worker, and the number of schools at the department level. Despite the high collinearity among these variables, they all remain statistically significant when included jointly, and collectively explain more than 75% of the variance in G _i.

Table 2. Validation of calibrated provision of public education

Notes: OLS regressions. Standard errors are robust to heteroskedasticity and clustered at department level.

Figure 4 provides a visual representation of the calibrated level of access to education across Senegal. Except for the four departments in the Dakar region (Dakar, Guédiawaye, Pikine, and Rufisque), as well as Bambey in the Diourbel region, access to education is generally low to very low in the rest of the country. This indicates that selective emigration could potentially have little effects on human capital formation in poor regions.

Figure 4. Calibrated access to education (G _i) by département.

Effects on welfare

To estimate the welfare gains resulting from emigration opportunities, we rely on the expected value of post-migration utility denoted as Ω_i,s, as defined in Eq. (3). Remember that this variable measures both unconditional and conditional (on the chosen alternative) mean values of maximum utility for individuals born in region i. In Figure 5, we simulate the change in welfare due to migration opportunities, $\Delta \Omega _{i, s} \equiv \Omega _{i, s}-\Omega _{i, s}^{NM}$, and plot it against the no-migration counterfactuals, labeled as NM. Results for college graduates are displayed in the left panel, while those for the less educated are presented in the right panel. As Ω_i,s is the log of the expected value of post-migration income, we interpret ΔΩ_i,s as the relative change in welfare resulting from selective emigration, expressed as a percentage of the NM counterfactual level.

Figure 5. Welfare analysis $( \Omega _{i, s}-\Omega _{i, s}^{NM})$ v.s. $\Omega _{i, s}^{NM}$ by département (a) International – College graduates, (b) International – Less educated, (c) Internal – College graduates, (d) Internal – Less educated, (e) Both – College graduates, (e) Both – Less educated. Note: Authors’ computations. Each département is represented by a bubble, whose size is proportional to its population in 2013. The blue quadratic curve and its interval of confidence (dotted curves) depicts the relationship between the emigration-driven changes in welfare and the no-migration counterfactual. A negative (resp., positive) correlation means that emigration reduces (resp. increases) welfare disparities between départements.

The top panel of Figure 5 illustrates the welfare implications stemming from international migration opportunities. As expected, these opportunities substantially enhance the welfare of college graduates, with increases exceeding 30% in départements characterized by the highest emigration rates, such as Dakar, Guédiawaye, Pikine, Mbacké, Matam, Goudiry, Bakel, and Kanel (where welfare experiences a remarkable 85% rise). In contrast, the impact of international migration on welfare in the rest of the country is less pronounced. This is particularly evident in Salémata, a département that benefits significantly from internal migration opportunities (see below) but lacks international connectivity.

Overall, the correlation between the welfare effects of international migration and the no-migration baseline is relatively low, indicating that prospects for high-skilled emigration have a limited influence on inter-département disparities in welfare among college-educated individuals. A different pattern emerges when examining the welfare effects of international mobility on the less educated population. Except for select départements, mainly in the eastern regions bordering Mauritania and Mali, international connectivity heavily favors the wealthy départements of the Dakar region, where the welfare of the less educated increases by 5 to 13%. This increase is notably smaller compared to the welfare response to high-skilled migration. In less affluent départements, the welfare impact of low-skilled migration is low, leading to the conclusion that international migration prospects tend to exacerbate inter-département disparities in welfare within the low-skilled population.

The middle panel of Figure 5 provides insights into the welfare effects stemming from internal migration opportunities. As illustrated in Figure 2, internal connectivity is relatively high for college graduates across all regions. In contrast to international migration, the wealthy départements of the Dakar region benefit less from high-skilled internal emigration than their less affluent counterpart. While welfare gains for college graduates reach approximately 25% in the Dakar region, they approach 100% in most other regions. Salémata, in particular, stands out with a substantial internal connectivity boost of 240% in welfare. This implies that high-skilled internal migration helps to reduce inter-département disparities in welfare among highly educated individuals.

In contrast, for less educated individuals hailing from regions distant from Dakar, internal connectivity is generally low or very low, as highlighted in Figure 3, with the least internally connected départements located in the regions of Matam (Ranérou Ferlo, Kanel, Matam), Tambacounda (Goudiry, Koumpentoum, Tambacounda), Kolda (Vélingara, Médina Yoro Foulah), and Sédhiou (Bounkiling). These départements exhibit limited welfare gains of approximately 10%. Overall, internal migration prospects tend to increase inter-département disparities in welfare among the less educated, who constitute the overwhelming majority of the Senegalese population.

The combined effects of both internal and international migration, shown in the bottom panel of Figure 5, highlight that mobility slightly reduces welfare disparities within the college-educated population. It induces substantial welfare gains of approximately 100% in many départements. Conversely, mobility significantly increases welfare disparities within the less educated population, leading to substantial gains of around 60% in the wealthiest départements and smaller gains of about 15% in the least connected ones.

Effects on expected skill premium

We now turn our attention to human capital formation. Selective migration prospects affect individual incentives to invest in higher education by increasing its expected return. In the top panels of Figure 6, we focus on the expected skill premium, as measured by λ _i defined in equation (5). We plot the migration-driven relative changes in the expected skill premium, $( \lambda _{i}-\lambda _{i}^{NM}) /\lambda _{i}^{NM}$ (expressed as a percentage of the NM counterfactual level), against the no-migration counterfactual levels, $\lambda _{i}^{NM}$. These skill-premium responses are closely linked to the differential welfare effects experienced by college graduates and the less educated. Due to positive selection in international and internal migration, this differential is positive in nearly all départements. However, the magnitude of this differential varies significantly across regions.

Figure 6. Skill premium and human capital effects by département. (a) International – Expected education premium, (b) Internal – Expected education premium, (c) Both – Expected education premium, (d) International – Human capital, (e) Internal – Human capital, (f) Both – Human capital. Note: Authors’ computations. Each département is represented by a bubble, whose size is proportional to its population in 2013. Contrary to the previous figure, we do not provide quadratic trends here because their intervals of confidence are too large and average effects are driven by outliers.

In Panel (a), we observe the most significant skill-premium responses to international migration prospects in the wealthiest départements, where human capital is already abundant, and the no-migration skill premia are low. In the Dakar region, the skill premium increases by 30 to 60%, while in Kanel (Matam region), it surges by more than 100%. Conversely, in départements with scarce human capital and larger no-migration skill premia, the skill premium rises by less than 10%. Consequently, international migration tends to enhance incentives for human capital acquisition in the wealthiest regions, signifying that the national brain gain is predominantly driven by these départements.Footnote ¹¹

Panel (b) illustrates that internal migration prospects yield qualitatively similar effects, albeit with a somewhat weaker correlation. Internal connectivity is generally more substantial than international connectivity, except in Dakar. Consequently, in the Dakar region, internal migration prospects have a limited or even negative impact on the expected skill premium. However, in most other regions, positive effects ranging from 40 to 100% are observed. The most significant responses occur in départements located in the middle of the no-migration counterfactual distribution. For instance, the skill premium nearly triples in the départements of Kanel and Ranérou Ferlo in the Matam region.

In Panel (c), when considering the combined effects of internal and international migration, we observe that the overall impact of mobility on incentives to acquire human capital is predominantly influenced by internal migration prospects. Expected skill premia experience boosts of 100 to 200% in many regions, with the most substantial effects seen in Salémata, Ranérou Ferlo, Kanel, and Oussouye. Conversely, the smallest effects are observed in the wealthiest regions such as Dakar (Pikine, Rufisque, Guédiawaye, and Dakar), Thiés (Tivaouane, M'bour, and Thiés), or Diourbel (Bambey).

Effect on human capital accumulation

We finally investigate the net effect of migration prospects on (post-migration) human capital accumulation, measured by h _i and defined in Eq. (8). In the bottom panels of Figure 6, we plot the migration-induced change in human capital, denoted as $\Delta h_{i} \equiv h_{i}-h_{i}^{NM}$ (expressed as a percentage point deviation), against the no-migration counterfactual level, $h_{i}^{NM}$. It is essential to note that human capital responses result from the reactions of natives to emigration prospects (including pre-migration incentives and post-migration composition effects) and from the migrant inflows from other départements. However, as depicted in Figure 1, the inflow of college graduates is negligible in most départements, except for Dakar and Thiés, which are joined by Diourbel when focusing on the less educated.

Moving on to Panel (d), we observe that selective international migration prospects lead to increased domestic human capital accumulation in the Dakar region. Specifically, there is about a 2-percentage-point increase in human capital in Dakar, Guédiawaye, and Pikine, and roughly a 1-percentage-point increase in Rufisque. We also see an increase in domestic human capital in Ziguinchor, Salémata, Thiés, and Saint-Louis, which are four other départements where international migration prospects enhance skill premia and access to education is not too limited. In many other regions, the human capital response is minimal and close to zero.

In Panel (e), we observe that internal migration prospects increase human capital in the départements of the Dakar region, with Guédiawaye experiencing the most significant net gain. This overall increase is primarily driven by the influx of college graduates from other regions, as internal movements from Dakar are relatively limited. Moreover, these internal movements tend to involve individuals who are negatively selected in terms of human capital, so they do not generate positive domestic incentive effects. In the poorest départements of the country, the net effect is positive but small, with the exception of Bambey and Thiés, where larger gains are observed. However, four départements with limited access to education (Saint-Louis, Oussouye, Bignona, and Sédhiou) experience net losses of human capital due to internal migration.

Finally, Panel (f) dipslays the combined effects of internal and international migration. Our analysis demonstrates that migration prospects predominantly benefit the wealthiest départements of Senegal, primarily situated in the Dakar region. These gains are driven by substantial incentive effects resulting from selective emigration opportunities to OECD countries, as well as the inflow of college graduates from other regions within the country. Additionally, the positive effect is observed in départements with good access to education and strong connections to Dakar, such as Thiés and Bambey.

Conversely, in the remaining départements, the effect is either negligible or non-existent, confirming our initial assumption that the brain gain mechanism is predominantly governed by the observed effects in a few well-connected areas. Moreover, net losses of human capital are experienced in départements that are connected, either internationally or internally, but have limited access to education, exemplified by Saint-Louis, Oussouye, Bignona, and Sédhiou. This comprehensive analysis underscores the nuanced and localized impact of migration prospects on human capital accumulation across Senegal.