The Impact of Gender Inequality in Education and Employment on Economic Growth: New Evidence for a Panel of Countries

Abstract

Using cross-country and panel regressions, we investigate to what extent gender gaps in education and employment (proxied using gender gaps in labor force participation) reduce economic growth. Using the most recent data and investigating an extended time period (1960–2000), we update the results of previous studies on education gaps on growth and extend the analysis to employment gaps using panel data. We find that gender gaps in education and employment considerably reduce economic growth. The combined “costs” of education and employment gaps in the Middle East and North Africa, and South Asia amount respectively to 0.9–1.7 and 0.1–1.6 percentage point differences in growth compared to East Asia. Gender gaps in employment appear to have an increasing effect on economic growth differences between regions, with the Middle East and North Africa, and South Asia suffering from slower growth in female employment.

Keywords:

• Economic development
• economic growth
• economics of gender
• JEL Codes: J7, J16, O4

ACKNOWLEDGMENTS

A version of this contribution was written as a background paper for the World Bank Flagship Report “Gender and Development in the Middle East and North Africa: Women in the Public Sphere” (2004). Funding from the World Bank in support of this work is gratefully acknowledged. We would like to thank: Nadereh Chamlou; Susan Razzazz; four anonymous referees and the guest editors of this special issue; participants at the MENA consultative council meeting on gender at the World Bank; Paula Lorgelly; participants at seminars in Munich, IZA in Bonn, Harvard University, and at the 2008 IZA/World Bank Conference on Employment and Development in Rabat; and a workshop dedicated to this volume in New York in 2008 for helpful comments and discussion.

Notes

¹ See Stephan Klasen and Claudia Wink (2003) and Stephan Klasen (2002, 2007) for a discussion of these issues.

² This literature can be seen as part of a larger literature on the impact of inequality on growth. See, for example, Klaus Deininger and Lyn Squire (1998) and Christiano Perugini and Gaetano Martino (2008).

³ Among the problems in their findings, Barro and Lee (1994) and Barro and Sala-i-Martin (1995) identify the absence of regional dummy variables, particularly for Latin America and East Asia. In the former, low initial gender gaps were accompanied by low growth, while in the latter, relatively high initial gender gaps were accompanied by high subsequent growth. In the absence of regional dummy variables, a causal link is made between these associations. It is quite likely, however, that the growth experiences of these regions were also influenced by other region-specific factors that are largely unrelated to gender gaps. The fact that these regional dummies are (at least jointly) statistically significant and that then the negative effect of female education reverses itself once regional (or country fixed) effects are considered supports this view. Further problems with these studies are the use of initial period education variables, the high collinearity between male and female education, and the endogeneity of these variables. For a discussion of these issues see Dollar and Gatti (1999), Paula Lorgelly and Dorian Owen (1999), Forbes (2000), and Klasen (2002).

⁴ The reported figures in Klasen (2002) are actually slightly different, as Israel, Sudan, and Turkey were all included in the Middle East region. For the report in the present study, they were allocated to other regions (Israel to the OECD, Turkey to Eastern Europe, Central Asia and Sudan to Sub-Saharan Africa) and therefore the analysis in Klasen (2002) was redone to reflect this. The figures reported above are based on that analysis.

⁵ See, for example, Abu-Ghaida and Klasen (2004), Stephan Klasen (2006), Janet Stotsky (2006), and Mark Blackden, Sudharshan Canagarajah, Stephen Klasen, and David Lawson (2007) for a review.

⁶ See Bloom and Williamson (1998) and Klasen (2002) for a full exposition of these arguments.

⁷ Klasen (2006) reviews the literature, and also notes that such strategies have now been extended with some success to countries such as Tunisia, Bangladesh, China, and Vietnam.

⁸ There is also some empirical support for the claim by Seguino (2000a, 2000b) that higher gender wage gaps were a further pre-condition of these export-oriented strategies. There is a related debate as to whether growth has reduced these gender wage gaps, which appears to be the case in many but not all countries; also, they remain large, particularly when controlling for education. For a discussion, see Seguino (2000a, 2000b), Klasen (2002), Busse and Spielmann (2006), and Stotsky (2006), among others.

⁹ See a related discussion in King, Klasen, and Porter (2008) about the growth and welfare effects of women as policy-makers. The “causes” of these differences in behavior may well be related to the different socialization of girls and boys, a subject that leads beyond the scope of this paper.

¹⁰ The exceptions are again the two short-term structuralist models of Blecker and Seguino (2002) where large gender gaps in pay, implicitly combined with no gender gaps in education and employment, can deliver income-enhancing effects.

¹¹ On these issues, see the discussions in Elizabeth M. King and M. Anne Hill (1993), Harold Alderman, Jere R. Behrman, Shahrukh Khan, David R. Ross, and Richard Sabot (1995), Harold Alderman, Jere R. Behrman, David R. Ross, and Richard Sabot (1996), and the World Bank (2001).

¹² Also, it is not obvious which factor is the prime cause of gender gaps to be included in a reduced-form estimation.

¹³ In the case of these papers, the focus on semi-industrialized, export-oriented countries was intended. But therefore, the findings of these papers cannot address the question of whether there is a more general relationship between pay gaps and growth in developing countries that do not belong to this small group.

¹⁴ It turns out that, in our total sample, gender gaps in education and employment are not very closely correlated, so it should be possible to separately identify the effects. This overall low correlation is largely driven by a negative correlation between gender gaps in education and employment in Sub-Saharan Africa, and to a lesser extent South Asia, while in the other regions the correlation is positive and usually large and statistically significant. This negative correlation in Sub-Saharan Africa is related to the high female employment in agriculture despite low levels of female education; in this case, low education is not a barrier to high female employment as is the case elsewhere. For Africa's formal sector, see Klasen (2006) and Blackden et al. (2007).

¹⁵ See Appendix Table 1 for a list of countries in each region for which we have data availability.

¹⁶ Also note that following the World Bank country classification system, Turkey is considered to belong to the Eastern Europe and Central Asia region, and Israel belongs to the OECD.

¹⁷ Iran is the only major oil producer included in the sample, but Egypt, Algeria, and Yemen also depend, directly or indirectly (via migration and remittances), on oil production.

¹⁸ Sub-Saharan Africa's high female labor-participation rate is largely confined to the agricultural sector, which still employs the majority of workers in most Sub-Saharan African countries. The international comparability of labor force participation data in own-account agriculture is particularly problematic. In formal-sector employment, female employment rates are much lower and the gender gap is significant; but these data are, as discussed, missing for many countries and show consistency and comparability problems.

¹⁹ The combination of rapidly shrinking gender gaps in education and large and persistent gender gaps in employment in the MENA region constitutes a major puzzle. See World Bank (2004) for a careful discussion.

²⁰ Unemployment rates for women in Latin America and MENA are several points higher than for men. Thus in these regions, the gender gap in employment is actually slightly larger than in labor force participation. But, as this gender gap in unemployment rates is rather stable over time, it would be absorbed by the country-specific effects in our panel estimation. We also tried to use sectoral employment data available for some countries since the 1980s to adjust our labor force participation data to focus on non-agricultural employment. But there were so many data gaps and measurement errors and the comparability problems were so severe that these data turned out to be unusable.

²¹ We have also undertaken some further robustness checks with more variables used in standard growth regression analysis. The results are available on request. While the use of regional dummy variables is invariably a measure of our ignorance, in many cross-country regressions they turn out to be significant, pointing to region-specific left-out variables that are hard to capture in standard cross-country regressions.

²² Knowles, Lorgelly, and Owen (2002) suggest that this is the most suitable specification for analyzing gender gaps in education. This specification was also used in Klasen (2002).

²³ Note that Equations 3 and 4 contain an additional explanatory variable with respect to Klasen (2002): openness.

²⁴ We use dummy variables for all regions, where the region left out is EAP.

²⁵ In the panel, we use the total years of schooling of the population over 25. We do so because in the panel analysis we only have a 10-year window in which human capital (and gender differences) can have an effect, and thus we want to focus our attention on the human capital of the labor force (rather than also including the 15–24-year-olds, only some of whom are in the labor force). In robustness checks, we also include the years of education of adults 15 or older to particularly capture the effects of young, educated women, who make up a large share of female employment in many developing countries.

²⁶ On the other hand, empirically, male labor force participation rates do not differ much across space and over time, so the growth effects observed are probably due to increased female employment.

²⁷ We have run the regressions for random effect but specification tests (Hausman tests) suggest that the fixed effect specification is superior.

²⁸ The previous version of the Penn World Table 5.6 (Alan Heston, Robert Summers, and Bettina Aten 1998) provided data for the following additional countries: Djibouti; Malta; Oman; Puerto Rico; Saudi Arabia; Somalia; Surinam; Iraq; Liberia; Myanmar; Reunion; Sudan; Swaziland; and Yugoslavia. For the last nine countries, Barro and Lee (2000) data on education were available. In addition, the data for Eastern European countries were not limited to the 1990s. Penn 6.1 (Heston, Summers, and Aten 2002) provides data for the entire sample set only for two Eastern European countries (Romania and Cyprus). Barro and Lee education data are suspicious for Austria and Bolivia, as they suggest stagnating or declining educational attainment despite substantial increases in enrollment. Hence, we dropped these two countries from our analysis.

²⁹ It is quite difficult to adequately measure trade openness, and the variables we use, export plus imports as a share of GDP, are not free from problems, as these ratios are systematically lower in larger economies despite identical trade policies. Other proxies have different problems. For a discussion, see Jeffrey A. Frankel and David Roemer (1999) and Dani Rodrik and Francisco Rodriguez (2000).

³⁰ The population growth regression does not pass the Reset test, suggesting that omitted variables and/or non-linearities in these regressions might be a problem. This does not affect our main results (including the size of the direct, indirect, and total effects), and could only have a possible (and likely minor) influence on the relative importance of these two indirect effects.

³¹ But here, endogeneity might be a problem, which will be partially addressed in the panel regressions.

³² While there is a large and conclusive literature that shows that female education reduces fertility (for example, see T. Paul Schultz [1997], Klasen [1999], and World Bank [2001] for a survey), the link between female education and population growth rates is weaker, as population growth is also affected by the age structure of the population. In a population with a large share of women of child-bearing age, even a low total fertility rate for each of them can generate considerable population growth compared with a population where the share of women is lower. Therefore, it is not surprising that the link here is weaker than if one used the total fertility rate as the dependent variable. When we include labor force growth in the population equation to proxy for the effect of the age structure, the effects of the initial female-male ratio of schooling and the ratio of the growth become significant, as expected.

³³ This may be related to the fact that the impact of population growth and working-age population growth materializes with some delay and may therefore not be captured well in the 10-year periods considered.

³⁴ It is even larger if we consider the reduced form estimate – that is, if we leave out the investment rate, labor force growth, and population growth. In both cases, they are larger than identical panel regressions in Klasen (2002).

³⁵ We also analyzed the sample where we dropped Sub-Saharan Africa and Latin America in the 1990s and report on the results where appropriate.

³⁶ The regression is not shown but is available on request.

³⁷ This is confirmed by regressions (not shown here) where we replaced the MACT with the total activity rate (TACT) and now find that the impact of the gender gap is larger, while the impact of the total activity rate is now negative. These regressions are available on request.

³⁸ See Klasen (2006) for further discussions on these country studies.

³⁹ Structural barriers, here, are related to the economic reconstruction, recession, and limited domestic and foreign investment.

⁴⁰ Nevertheless, the World Bank has used these household surveys to generate roughly consistent, comparable, and publicly available poverty statistics for developing countries. With these surveys, one could generate consistent and comparable statistics on labor force participation, employment, unemployment, and pay. The currently available ILO estimates are a poor substitute for such consistently generated survey-based estimates.

⁴¹ To the extent that such increased labor force participation would come in addition to non-market work, the double burden this implies for the women concerned is also not considered here but is clearly an issue that is under investigation in the literature.

⁴² Abu-Ghaida and Klasen (2004) and King, Klasen, and Porter (2008) estimate the magnitude of these effects.