2.1 Economic development, structural transformation, and demand for labor

In general, economic activities comprise the production and transaction of goods and services. When the economic activity of countries grows, this process coincides with the structural transformation of the economy, meaning the reallocation of economic activity away from agriculture to manufacturing and services (Chenery, Reference Chenery1960; Kuznets, Reference Kuznets1973). The shift away from agriculture is accompanied by the introduction of new technologies and international integration, the facilitation of technology transfer, foreign direct investment, and international trade (Goldberg and Pavcnik, Reference Goldberg and Pavcnik2007; McMillan et al., Reference McMillan, Rodrik and Verduzco-Gallo2014). The level of economic development is thus directly linked to the stage of structural transformation of the economy.

At low levels of development, agriculture is often the dominant sector with the largest employment share. However, it also tends to be a comparably low productive sector and income from agriculture remains small (Restuccia et al., Reference Restuccia, Yang and Zhu2008; Gollin et al., Reference Gollin, Lagakos and Waugh2014). Earning opportunities do not require extensive skill sets and the agricultural sector does not reward or employ highly skilled workers. Skills acquired through education are an essential determinant for the productivity of workers. In other words, different levels of education convey different skill levels and indicate worker productivity (Jones, Reference Jones2001; Spitz-Oener, Reference Spitz-Oener2006). Economies at low development levels largely require less productive, low-skilled workers; these are workers without significant educational achievements. Dominated by agriculture and non-competitive manufacturing, these economies generally provide unfavorable environments for generating income and securing economic welfare. Educated individuals, in particular, are likely to find it hard to receive adequate returns on their labor given the lack of demanding, high-skilled jobs.

With rising levels of development, the agricultural employment share decreases and the ratio of workers in the manufacturing and service sector increases (Herrendorf et al., Reference Herrendorf, Rogerson, Valentinyi, Aghion and Durlauf2014). The expansion of the manufacturing and service sector depends on technological innovations, which changes the demand for labor and its allocation across the three broad sectors (Lee and Wolpin, Reference Lee and Wolpin2006). Both manufacturing and services are more technology intensive compared to agriculture and require more skilled labor input. In addition, productive sectors and firms are more prevalent at higher levels of development. They replace less productive and unprofitable firms that are not fit to compete, especially when markets open up to international competition (Melitz, Reference Melitz2003; Helpman et al., Reference Helpman, Itskhoki and Redding2010). Overall, the output of the economy grows, technology advances, and more goods and services with higher value are produced and traded. The production of these goods and services requires highly skilled workers. Thus, the demand for labor shifts toward more skilled workers, who are on average more educated (Topel, Reference Topel, Ashenfelter and Card1999; Krueger and Lindahl, Reference Krueger and Lindahl2001).

2.2 Labor market demand, perceived risks, and welfare

Economic development affects labor market dynamics and the relative demand for skilled labor. I argue that people's economic welfare perceptions are dependent on the fit of their individual skills and the local demand for labor. When individual skill levels match the type of labor that is relatively more demanded, people should feel less at risk and satisfied with their economic situation. A relative mismatch indicates inadequate or comparably low returns to labor and adverse results in comparison to others in the local labor market. This is important as research shows that social comparison, performed consciously or unconsciously, plays a vital role for perceptions of welfare (Wolbring et al., Reference Wolbring, Keuschnigg and Negele2013; Reyes-Garcia et al., Reference Reyes-Garcia, Baibumira, Pyhälä, Wunder, Zorondo-Rodriguez and Angelsen2016). While people can compare their welfare to different benchmarks (Festinger, Reference Festinger1954), and it is not possible to clearly identify these reference point(s), we know that comparison to others plays a central role in the assessment of income (Clark and Senik, Reference Clark and Senik2010). Therefore, discrepancies between own skill level and local demand for labor should adversely affect individuals’ views of their labor market risks and welfare.

As pointed out, the economic opportunities in less developed environments are limited and labor market demand is skewed toward low-skilled workers. The wage premium for educated workers is negligible and well-educated people, who could in principle take on more demanding jobs, only have limited economic opportunities in these environments.Footnote ¹ In contrast, labor markets in more-developed regions favor well-educated workers with higher productivity. In economically thriving areas, the wage premium of educated workers increases disproportionately to those of the less educated.

The diverging demand for labor in differently developed environments entails a clear mismatch of relative demand for skilled labor and individuals with high-educational achievements in less-developed areas. This misfit results in a lack of skill-adequate income opportunities. While it does not preclude highly educated workers from gaining employment, their income opportunities remain limited and it is unlikely that they will earn significantly more than those with low-educational attainments in their community. Concerns about appropriate wages and employment opportunities should therefore prevail among the highly educated. In contrast, educated workers in well-developed environments benefit from the higher demand for skilled labor and related income opportunities. Meeting the skill requirements of the labor market they are more sought after and able to secure employment and appropriate income, translating into less concerns about their economic situation. More favorable employment opportunities and higher income should generally improve the welfare perceptions of this group.

People with low-educational attainment, on the other hand, are favored by the demand for labor in less-developed environments. Income might be difficult to sustain overall, but these less-developed labor markets predominantly offer employment opportunities that fit the skill level of less-educated people. In turn, welfare insecurity for this group should be lower in less-developed areas than in booming environments. Here, the relative demand for labor does not comply with skill level. The disproportionate demand for educated labor results in a bigger wage gap between the high- and low-skilled. Living in these more-developed environments might not directly deteriorate the economic situation of low-skilled labor, however, the increasing wage gap changes the benchmark for assessing their economic welfare. Despite living in thriving environments, the low-skilled are not able to benefit to the same extent than more educated people. Welfare perceptions of the less educated should therefore be less favorable the higher the economic development level of their environment, reflecting a sense of being left behind. Overall, this suggests that the level of development and corresponding labor market demand should affect perceived economic welfare differently depending on people's educational achievements.

3.1 Local economic development

Even though perceptions of welfare are not independent of aggregate economic development levels, I argue that we need to focus on the economic conditions and labor markets in more restricted areas around people's place of residence. First, economic conditions do not change uniformly but vary substantially within countries (Kanbur and Venables, Reference Kanbur and Venables2005; Kim, Reference Kim2008). Therefore, some people still live in environments dominated by small-scale agriculture, while others live in industrialized, highly developed areas. Second, people usually live at a specific location and while they can commute to engage in economic activities, their daily mobility is limited (Marchetti, Reference Marchetti1994; Kung et al., Reference Kung, Greco, Sobolevsky and Ratti2014). Thus, the local economic development to which people are exposed and the labor markets they participate in are limited by the time and corresponding distance they are willing and able to commute.

Figure 1 visualizes the concept of local economic development. First, local development levels are dependent on the site of residence of the individual and encompass the area people are able to commute to on a daily basis. Second, these local environments are not restricted by subnational boundaries, as crossing provincial borders is not costly. People can reside in an economically weak region but might be able to commute to more prosperous areas to earn money. Thus, subnational administrative borders are unsuitable to capture the specific economic development levels people are exposed to. Third, crossing national borders often imposes costs or is sometimes even impossible. Therefore, local environments are confined within national boundaries. The conceptualization of local economic environments takes persistent and substantial within-country variation in economic conditions into account. Thereby, it captures the “economy” that directly affects people's livelihoods.

Figure 1. Local economic development.

While the economic development and structural transformation of advanced industrialized economies also vary within and over time, shifts in the importance of sectors and related demand for labor are far less pronounced than in developing and emerging countries. Therefore, my empirical analysis focuses on 36 African countries characterized by high levels of within-country variation in economic development. Africa is the continent with the largest share of developing, least developed, and low-income countries. While the level of economic development still varies widely between and within the analyzed countries, the overwhelming majority pursues policies to enhance their economic development (Fosu and Ogunleye, Reference Fosu, Ogunleye, Monga and Lin2015). However, a comprehensive analysis of the subjective, individual-level effects of different stages of development is still missing (Diao et al., Reference Diao, Harttgen and McMillan2017).

To test the effect of economic development on individual welfare perceptions, I use geocoded individual survey data and combine it with information on local economic conditions. Linking geocoded individual survey data with fine-grained economic data, allows for a precise assessment of the effects of varying economic development levels on perceived individual welfare. I use survey data from the Afrobarometer between 2002 and 2014 (second to sixth waves). It provides a representative cross-section of citizens aged 18 and older for up to 36 African countries.Footnote ² The survey data were geocoded by assigning longitude and latitude to the survey clusters of respondents (BenYishay et al., Reference BenYishay, Rotberg, Wells, Lv, Goodman, Kovacevic and Runfola2017). Each survey cluster represents the smallest geographical census unit from which respondents are sampled.Footnote ³ Overall, 182,937 respondents are clustered in 13,156 survey clusters, with on average 113 clusters per country and year.

3.2 Outcome variables: perceived labor market risks and living conditions

Local development levels and perceived economic welfare are linked via the (mis)match of local labor market demand and respondents’ skill sets. I analyze two core aspects of economic welfare: perceived labor market risks and individual living conditions. The former allows to closely assess the link to labor market-specific welfare concerns, the second taps into people's individual considerations of economic welfare.

Perceived labor market risks are measured with a question asking respondents about the most pressing problem that should be addressed. Those who mention either “Wages, income, and salaries,” or “Unemployment” as their most severe concerns are coded as feeling at risk in the labor market. Roughly 26 percent of respondents report being concerned about wages or unemployment while the remainder are not predominantly worried about employment insecurity. This operationalization of employment insecurity only captures income or job-related worries. Other economic concerns such as “Poverty,” “Food shortage,” or “Social Welfare” are excluded from the employment insecurity measure. While employment insecurity is arguably closely related to other economic worries, this conservative operationalization captures people who clearly worry most about these labor market-related problems. However, also individuals who are concerned about unemployment or wages in second or third place are arguably experiencing labor market risks, even though this might be a less pressing issue for them. To test the effect of local development levels on this more encompassing perception of labor market risks, all respondents who name wages or unemployment as the first, second, or third most pressing problem are coded as worried about the labor market.Footnote ⁴

The second outcome variable measures perceived economic welfare more generally with an item on self-reported living conditions. In contrast to generic satisfaction with life or happiness questions, reported living conditions should capture the perceived economic or material welfare of respondents more specifically. Around 49 percent of respondents report “Very Bad” or “Bad” present living conditions. In total, 21% of answers fall into the undecided category of “Neither Good nor Bad,” and 30 percent report that their living conditions are “Good” or “Very Good.”

3.3 Explanatory variables: measuring local economic development and individual education

To test how subjective welfare is affected by local development, it is necessary to approximate the specific economic environments of people. The size of these local economic environments is constrained by people's (daily) mobility. Research on commuting patterns in African countries shows that mobility is limited and commuting is relatively time-consuming. While average commuting distances vary, they rarely exceed 30 km (Bryceson et al., Reference Bryceson, Mbara and Maunder2003). Therefore, I calculate a 30 km buffer zone around the location of each respondent. For robustness checks, I also use 10 and 50 km buffer zones, as well as dynamic buffers.Footnote ⁵ All local environments are clipped at national borders to account for the costs associated with crossing these boundaries.

Official national or subnational economic data are not sufficiently disaggregated and flexible to measure the development levels of these customized local environments. Therefore, I use night light emissions, illumination recorded during nighttime, as a proxy for economic development. There are two main advantages to this proxy: first, night lights have a very high resolution and can be used to measure the economic development of customized areas or environments. Second, using night lights avoids relying on national statistics with limited data quality. Particularly in developing countries, differing methodologies and motivations to collect official economic data lead to serious measurement discrepancies and unreliable data accounts (Jerven, Reference Jerven2013). Night lights are a powerful proxy for economic activity: they are highly correlated with countries’ economic output in GDP (Chen and Nordhaus, Reference Chen and Nordhaus2011; Proville et al., Reference Proville, Zavala-Araiza and Wagner2017) and capture economic activity and the development of subnational units (Henderson et al., Reference Henderson, Storeygard and Weil2011). Even at the neighborhood level, they are a reliable predictor of economic wealth (Weidmann and Schutte, Reference Weidmann and Schutte2017).

I draw on the “stable lights” version from 1992 to 2013 by the US National Oceanic and Atmospheric Administration (Center, Reference Elvidge, Baugh, Kihn, Kroehl and Davis1997), which is adjusted for the average amount of time the illumination is detectable and excludes non-stable light sources.Footnote ⁶ The annual raster data have a resolution of 30 arc-seconds, which is approximately one square kilometer at the equator. Values for raster cells range between 0 and 63 “digital number” (DN), where 0 DN would mean that no light was detected in a raster cell and 63 DN is the maximum illumination that can be registered.

Local economic development levels are measured by extracting the average emission of night light around the respondent in the year before the survey was conducted. The results are visualized in Figure 2. The map shows the distribution of survey clusters in the countries included in the Afrobarometer, dark (light) circles show survey clusters with relatively low (high) light emission and corresponding low (high) local development levels. For the analysis all light measures are log-transformed, this corresponds to the transformation of “traditional” GDP measures, addresses their heavily left-skewed distribution and the expectation that the effect decreases with higher values.

Figure 2. Local development levels Afrobarometer survey clusters.

Overall, the average night light illumination for 30 km buffer zones ranges between 0 and 59.88 DN. The value 0 means that no nighttime illumination was detected, e.g., areas without any electrification or deserts. The highest levels of illumination are recorded for environments that encompass large cities. The average variance of lights within countries across all years is roughly 21 DN. The substantial variation of night lights within countries points to the salience of this measure for assessing the effect of spatial variation in economic development.

To test the hypothesis of a conditional effect of skill and local development on perceived welfare, individual skill is operationalized with a question asking respondents about “the highest level of education” they have completed. A number of studies suggest that an increase in years of schooling increases the skills of workers and thus their productivity and related output (Topel, Reference Topel, Ashenfelter and Card1999; Jones, Reference Jones2001; Krueger and Lindahl, Reference Krueger and Lindahl2001). Education is a very general measure of skill and cannot account for sector or task-specific skills, as well as on the job training. However, skill specificity is less pronounced in my sample of developing and emerging countries, thus education levels should still adequately measure the skill set of respondents. The education item ranges from “No formal schooling” (0) to “Post-graduate” (9). Over a third of respondents reported low levels of education (1–3). However, the share of people with intermediate education levels (4 and 5) is equally high. Another 14 percent of respondents have received some sort of tertiary education.

When levels of light emission are split into quartiles, 37 percent of respondents with no formal schooling live in environments which fall into the first quartile, while only 12 percent live in the most illuminated quartile of local environments. The distribution of low educated respondents per illumination quartile is comparable, with roughly a third of respondents in the first two illumination quartiles and approximately 17 percent in the fourth. Secondary education is most prevalent in the most illuminated environments, 33 percent of respondents with intermediary education categories reside in these areas. Their share in the first and second quartiles is roughly 40 percent and thus still quite substantial. Having received a university education makes living in highly developed areas more likely, 52 percent of university-educated respondents live in environments that are part of the most illuminated quartile. Still, 9 percent of this educational group live in the least-developed quartile of environments, while a further 12 percent reside in the second quartile.

For the analysis, I include individual controls for age, gender, employment status, type of residence (urban or rural), reported ethnic grievances, and the consumption of media in all models. I also control for incidents of lethal violence taking place within the local community in the year before the respondent is interviewed. For this purpose, I use the UCDP Georeferenced Event Dataset and overlap conflict sites with local environments of Afrobarometer respondents (Sundberg and Melander, Reference Sundberg and Melander2013; Croicu and Sundberg, Reference Croicu and Sundberg2017). In addition, I control for characteristics of the enumeration areas: Afrobarometer interviewers report available services and the quality of infrastructure of enumeration areas. From items asking about the presence of a post office, public school, police station, a clinic, and an official market I construct a services index that ranges from 0 (none of these services are present in the enumeration area) to 5 (all of the services are present in the enumeration area). An infrastructure index is constructed by combining information on whether an electricity grid, piped water system, and sewage system are present in the enumeration area. Additionally, information on the road quality in the enumeration area is included in the infrastructure index. The infrastructure variable ranges from 0 (no electricity grid, piped water, and sewage system, and non-tarred roads) to 4 (enumeration area with electricity, piped water, sewage system, and tarred roads).

3.4 Estimation method

Due to the clustered nature of the data, I estimate hierarchical models with random intercepts, with individuals i nested in enumeration areas j. This entails that a unique intercept parameter is used for each enumeration area, accounting for the diversity of enumeration areas (McElreath, Reference McElreath2016). I use a logit model for the employment insecurity variable and an ordered logit specification for the living conditions variable. The models are defined as follows:

(1)$$ \eqalign{ y^{\ast }_{ij} & = \beta_{0j} + \beta_{educ \cdot localdev} \cdot {{educ}}_{ij} \ast {{localdev}}_{\,j} \cr & \quad + \beta_{educ} \cdot {{educ}}_{ij} + \beta_{localdev} \cdot {{localdev}}_{\,j} \cr & \quad + \beta_{age} \cdot {{age}}_{ij} + \beta_{\,female} \cdot {{\,female}}_{ij} + \beta_{empl} \cdot {{empl}}_{ij} \cr & \quad + \beta_{urban} \cdot {{urban}}_{ij} + \beta_{eth} \cdot {{eth}}_{ij} + \beta_{media} \cdot {{media}}_{ij} \cr & \quad + \beta_{cyear} \cdot X_{cyear, ij} + \beta_{service} \cdot {{service}}_{\,j}\cr & \quad + \beta_{infra} \cdot {{infra}}_{\,j} + \beta_{conflict} \cdot {{conflict}}_{\,j} + \epsilon_{ij} }$$

β _0j ~ N(β _0j,  σ ²β ₀), where

labor market insecurity:

(2)$$y_{ij} = \left\{\matrix{ 0 & ( \rm{no\, employment\, insecurity}) & \rm{ if } & y^{\ast }_{ij} < 0 \cr 1 & ( \rm{employment\, insecurity}) & \rm{ if } & y^{\ast }_{ij} \geq 0 \cr }\right.$$

living conditions:

(3)$$y_{ij} = \left\{\matrix{ 1 & ( \rm{very\, bad}) & \rm{if} & -\infty < y^{\ast }_{ij} < \tau_1\cr 2 & ( \rm{\,fairly\, bad}) & \rm{if} & \tau_1 < y^{\ast }_{ij}< \tau_2 \cr 3 & ( \rm{neither\, good\, nor\, bad}) & \rm{if} & \tau_2< y^{\ast }_{ij}< \tau_3 \cr 4 & ( \rm{\,fairly\, good}) & \rm{if} & \tau_3 < y^{\ast }_{ij}< \tau_4\cr 5 & ( \rm{very\, good}) & \rm{if} & \tau_4 < y^{\ast }_{ij}< \infty }\right.$$

All models include the cross-level interaction between education and local development, as well as the above-described individual and context-level explanatory factors. The variable matrix Xcyear is an array of dummies for each country–year. All models are estimated in STAN with the No-U-Turn Sampler (NUTS) (Hoffman and Gelman, Reference Hoffman and Gelman2014) via the brms package for R (Bürkner, Reference Bürkner2017).Footnote ⁷ Convergence of the four chains is assessed with $\hat {R}$ diagnostics, which should be close to 1 and never exceed 1.03 in all models across all parameters (Gelman and Rubin, Reference Gelman and Rubin1992). The results presented below are robust to non-hierarchical, linear specifications with country–year and spatial fixed effects for raster cells and survey cluster fixed effects (see Table 18 in the Appendix).