The Importance of Cognitive Domains and the Returns to Schooling in South Africa: Evidence from Two Labor Surveys

Highlights

• We study the returns to cognitive domains in two samples from Sub-Saharan Africa

• Each additional year of schooling increases earnings by approximately 18-20 percent

• The 2SLS estimates of the returns to schooling are above the OLS estimates

• Executive functioning skills are important drivers of earnings in the rural sample

• Higher-order cognitive skills are more important for earnings in the urban sample

Abstract

Numerous studies have considered the important role of cognition in estimating the returns to schooling. How cognitive abilities affect schooling may have important policy implications, especially in developing countries during periods of increasing educational attainment. Using two longitudinal labor surveys that collect direct proxy measures of cognitive skills, we study the importance of specific cognitive domains for the returns to schooling in two samples. We instrument for schooling levels and we find that each additional year of schooling leads to an increase in earnings by approximately 18-20 percent. The estimated effect sizes—based on the two-stage least squares estimates—are above the corresponding ordinary least squares estimates. Furthermore, we estimate and demonstrate the importance of specific cognitive domains in the classical Mincer equation. We find that executive functioning skills (i.e., memory and orientation) are important drivers of earnings in the rural sample, whereas higher-order cognitive skills (i.e., numeracy) are more important for determining earnings in the urban sample. Although numeracy is tested in both samples, it is only a statistically significant predictor of earnings in the urban sample. (JEL I21, F63, F66, N37)

Introduction

Education is widely accepted as a leading instrument for promoting better economic outcomes at an individual level (UNESCO, 2005). Following the innovative analyses by Mincer (1958, 1974), the literature examines the economic returns to differing levels of schooling at an individual level.¹ Education can positively influence non-monetary outcomes, as it can reduce crime (Lochner and Moretti, 2004), improve health (Currie and Moretti, 2003), and improve civic engagement (Milligan et al., 2004). A large body of empirical literature investigates the monetary returns to education in developed and developing countries.² In general, estimated returns to education are larger in developing countries than in industrialized countries. Previous studies that examine the returns to education in developing countries rely on observational study designs. For sub-Saharan Africa, where rates of poverty are among the highest in the world (World Bank, 2016), education is particularly important. The issue of causal estimation of the returns to schooling is central for careful policy targeting. However, previous empirical estimates that used data from sub-Saharan Africa rely on observational study designs, which are ill-equipped to detect true causal effects. Precisely measuring the returns to schooling in the context of sub-Saharan countries is a challenge plagued by econometric issues, data availability constraints (Card, 1999, 2001), and survey design issues (Serneels et al., 2017).^3,4,5 For example, the socioeconomic characteristics of households and communities are important determinants of both schooling and labor market outcomes in developing countries. The omission of individual ability measures in empirical estimations can lead to large upward bias (ability bias) of observational study estimates (Lang, 1993). In practice, the bias present in estimates based on observational study designs could be quite large.

In this paper, we examine the returns to cognitive skills and the returns to schooling using data from South Africa.⁶ We use two surveys which collected extensive household information, including various dimensions of cognition. The surveys were conducted between 2002 and 2014 in two distinct regions of South Africa: urban and rural. They provide data on how various dimensions of cognition influence schooling and labor market outcomes in rural and urban settings. In particular, the use of two distinct surveys from two contrasting areas of South Africa enables an examination of the implications of heterogeneous local labor markets on the importance of specific cognitive domains. The first survey that we use is Health and Aging in Africa: A Longitudinal Survey of an INDEPTH Community in South Africa (HAALSI). HAALSI examines a rural population aged 40 and older. The second survey, the Cape Area Panel Study (CAPS), follows a metropolitan area population aged between 14 and 22. Both surveys collect cognitive evaluations via direct measurements of specific cognitive domains. In HAALSI (2014), four cognitive domains were tested: memory, numeracy, orientation, and attention. In CAPS (2002–2009), a cognitive evaluation was administered in Wave 1 (2002) and evaluated the domains of literacy and numeracy. Using data from the five waves of the CAPS panel data and the baseline wave of the 2014 HAALSI, we examine the influence of each cognitive domain on reported earnings. Furthermore, we use principal component analysis (PCA) to compile information from specific cognitive domains into an aggregate index and examine the impact of both specific cognitive measures and the aggregate cognitive index on the estimated returns to schooling.

Theoretical research by Ben-Porath (1967) and Becker (1975, 101, n. 89), and empirical studies in sub-Saharan countries (Boissiere et al., 1985; Field et al., 2009), indicate that cognitive skills influence schooling attainment. The failure to account for measures of cognitive ability in observational study designs that estimate the Mincer equation tends to bias the return to schooling estimates. Recent empirical comparisons have shown a gap between ordinary least squares (OLS) and instrumental variables (IV) estimations of the returns to schooling (Card, 2001; Trostel et al., 2002). The magnitude of ability bias in the returns to schooling can be an important factor that contributes to this discrepancy. Without causal estimates of the returns to schooling, gauging the ability bias in observational study designs can shed some light on the true lack of accuracy of observational study estimates on the returns to schooling in low-income countries.

We report three major findings. First, we find that the OLS estimate for the return to each additional year of schooling acquired is 14 percent (based on Wave 5 of the CAPS). Based on HAALSI, the OLS estimate for each additional year of schooling acquired is approximately 10 percent. These estimates are similar to estimates based on survey data from sub-Saharan Africa from Psacharopoulos and Patrinos (2004). Second, when we instrument for individual schooling levels using schooling fees, we find evidence consistent across all two-stage least squares (2SLS) specifications of higher effect size. Using the schooling fees as an instrument, we find that each additional year of schooling leads to an increase in earnings by approximately 18-20 percent (significant at the 5-percent level). When we use the quarter of birth as an instrument, the estimated return to each additional year of schooling increases up to 28 percent, but these estimates are imprecisely estimated.⁷ Our IV estimates are generally higher than the OLS estimates, a difference likely due to the 2SLS estimation being based on individuals with high marginal returns to schooling. In the third estimation, we detect evidence that suggests that specific cognitive domains play a far more important role in determining a person's earnings than general cognitive skills.⁸ Virtually no prior research examines the importance of cognitive skills for earnings in sub-Saharan countries. This is largely due to a lack of longitudinal data on specific cognitive skills and earnings. Based on the two samples, the overarching pattern that we observe is that, in the rural sample, executive functions (e.g., memory or orientation skills) are more important for earnings. Conversely, in the urban sample, higher-order cognitive skills are more important. Although numeracy skills are evaluated in both samples, it is a statistically significant predictor of earnings only in the urban CAPS sample.

This paper makes four important contributions to the labor economics literature on the returns to schooling in developing countries. First, we contribute to understanding the effects of cognitive skills on the returns to education. We estimate and demonstrate the importance of specific cognitive domain proxies in the classical Mincer equation. Our results suggest that cognitive ability scores explain a sizable, positive effect on earnings and slightly diminish the effect of schooling on earnings. Related to this issue, we show that the earnings gap between blacks/coloured and whites in South Africa is considerably reduced when we account for the ability measures. Neal and Johnson (1996) find a similar result in the context of the U.S. racial wage gap. In contrast, cognition measures do not influence the gender earnings gap. These findings imply that differences in premarket skills do account for a significant portion of the earnings gap across population groups, but do not seem to play a large role in the gender earnings gap in South Africa. Second, we contribute to the existing empirical literature by estimating the returns to certain dimensions of cognition that proxy for innate ability. Furthermore, cognitive skills measures implicitly provide information about the fundamental role of schooling quality and non-school factors in the production of human capital, which have been overlooked in the previous standard estimations of returns to schooling (Hanushek and Woessmann, 2008). Our results suggest that certain domains of cognition may be more important for earnings than other domains are. When we split the ability proxy into specific cognitive domain components, we find that only the numeracy domain in the CAPS sample and the memory and orientation domains in the HAALSI sample are statistically significant determinants of earnings. Third, although we rely on a limited sample size and an arguably imperfect instrument, we provide suggestive evidence of the causal estimates of the returns to schooling in South Africa by exploiting an IV estimation approach. We also provide suggestive evidence on the magnitude of the ability bias in the context of a developing country. Although much has been written about ability bias (Lang, 1993), and numerous empirical papers have estimated arguably causal returns to schooling in high-income countries (Card, 2001), little is known about the true casual returns in the context of low-income countries, particularly in sub-Saharan Africa. Although the measures of cognitive skills that we use are an important proxy for ability, they are not a direct measure of ability. Nevertheless, we examine how the inclusion of proxies for ability enhances understanding of the effects of cognitive skills on the returns to education. Finally, we provide additional evidence on the returns to schooling based on two survey sources drawn from two distinct demographic groups. HAALSI (conducted in 2014) surveyed an aging generation of rural South Africans, aged 40 and older. In contrast, the CAPS, collected between 2002 and 2009 in a metropolitan area, gathered information on young adults, aged between 14 and 22. By using two distinct samples—one based in an urban setting and another based in a rural setting—with rich data on various cognitive domains, we gain a better understanding of the relationship between the returns to schooling and specific cognitive domains in two distinct settings in the context of a developing country.

The remainder of the paper is structured as follows: Section II provides background information about South Africa's education system, major cognitive domains, and previous research about the role that cognitive domains play in determining a person's earnings. Section III describes the survey data, sampling strategies, and overall sample characteristics. Section IV presents the empirical strategy. Section V reports the results. Section VI describes various robustness checks. Section VII offers a discussion and concluding remarks.