Gender and education impacts on employment and earnings in West Africa: Evidence from Guinea

Economic Development and Cultural Change

Volumn 45, Issue 4, 1997, Pages 793-823

  Glick, Peter ^a   Sahn, David E ^a  

^a Department of Obstetrics Gynecology   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EARNINGS; EMPLOYMENT; GENDER INEQUALITY; LABOUR FORCE PARTICIPATION; RETURNS TO EDUCATION; WAGE DIFFERENTIALS;

GUINEA, CONAKRY;

EID: 0031428915     PISSN: 00130079     EISSN: None     Source Type: Journal    
DOI: 10.1086/452308     Document Type: Article

References (55)

1. S. Appleton, P. Collier, and P. Horsnell, "Gender, Education, and Employment in Côte d'Ivoire," Social Dimensions of Adjustment Working Paper no. 8 (World Bank, Washington, D.C., 1990); P. Glewwe, "Schooling, Skills, and the Returns to Education: An Econometric Exploration using Data from Ghana," Living Standards Measurement Working Paper no. 76 (World Bank, Washington, D.C., 1990).
2. S. Appleton, P. Collier, and P. Horsnell, "Gender, Education, and Employment in Côte d'Ivoire," Social Dimensions of Adjustment Working Paper no. 8 (World Bank, Washington, D.C., 1990); P. Glewwe, "Schooling, Skills, and the Returns to Education: An Econometric Exploration using Data from Ghana," Living Standards Measurement Working Paper no. 76 (World Bank, Washington, D.C., 1990).
3. W. Vijverberg, "Educational Investments and Returns for Women and Men in Côte d'Ivoire," Journal of Human Resources 28 (1993): 933-74.
4. Some 32,000 workers were released from the public payroll in Guinea during 1985-89. D. Sahn and B. Mills, "Reducing the Size of the Public Sector Work Force: Institutional Constraints and Human Consequences in Guinea," Journal of Development Studies 31 (April 1995): 505-28.
5. D. McFadden, "Conditional Logit Analysis of Qualitative Choice Behavior," in Frontiers of Econometrics, ed. P. Zarembka (New York: Academic Press, 1973).
6. The multinomial logit formulation has the limitation that the ratio of the probabilities of choosing two alternatives is independent of the number of alternatives (the "independence from irrelevant alternatives" property). We attempted instead to estimate the less restrictive nested multinomial logit model, using alternative nestings of choices (e.g., self-employment and nonparticipation vs. private and public wage employment). In each case, however, the term approximating the correlation of errors fell outside the unit interval, indicating misspecification (see G. S. Maddala, Limited Dependent and Qualitative Variables in Econometrics: Econometric Society, Monograph no. 3 [Cambridge: Cambridge University Press, 1983]). Therefore, we report only the estimates from the nonnested logit models below.
7. ij is the predicted probability of observing individual i in sector j, given by the multinomial logit model. For more details, see P. Glick and D. Sahn, "Labor Force Participation, Sectoral Choice, and Earnings in Conakry, Guinea," Cornell Food and Nutrition Policy Program Working Paper no. 43 (Cornell University, Ithaca, N.Y., 1993); and L. F. Lee, "Generalized Econometric Models with Selectivity," Econometrica 51 (1983): 507-12.
8. ij is the predicted probability of observing individual i in sector j, given by the multinomial logit model. For more details, see P. Glick and D. Sahn, "Labor Force Participation, Sectoral Choice, and Earnings in Conakry, Guinea," Cornell Food and Nutrition Policy Program Working Paper no. 43 (Cornell University, Ithaca, N.Y., 1993); and L. F. Lee, "Generalized Econometric Models with Selectivity," Econometrica 51 (1983): 507-12.
9. The table excludes the small percentage of the employed (less than 5%) who report working in more than one job and whose sector assignment would therefore be ambiguous. Given this ambiguity, as well as the small number of observations involved, we also dropped these workers from the econometric analysis of sector determination in the following section.
10. See also H. Alderman and V. Kozel, "Formal and Informal Sector Wage Determination in Urban Low-Income Neighborhoods in Pakistan," Living Standards Measurement Study Working Paper no. 65 (World Bank, Washington, D.C., 1989); T. H. Gindling, "Labor Market Segmentation and the Determination of Wages in the Public, Private-Formal, and Informal Sectors in San Jose, Costa Rica," Economic Development and Cultural Change 39 (1991): 585-605; and S. Khandker, "Labor Market Participation of Married Women in Bangladesh," Review of Economics and Statistics 69 (1987): 536-41.
11. See also H. Alderman and V. Kozel, "Formal and Informal Sector Wage Determination in Urban Low-Income Neighborhoods in Pakistan," Living Standards Measurement Study Working Paper no. 65 (World Bank, Washington, D.C., 1989); T. H. Gindling, "Labor Market Segmentation and the Determination of Wages in the Public, Private-Formal, and Informal Sectors in San Jose, Costa Rica," Economic Development and Cultural Change 39 (1991): 585-605; and S. Khandker, "Labor Market Participation of Married Women in Bangladesh," Review of Economics and Statistics 69 (1987): 536-41.
12. See also H. Alderman and V. Kozel, "Formal and Informal Sector Wage Determination in Urban Low-Income Neighborhoods in Pakistan," Living Standards Measurement Study Working Paper no. 65 (World Bank, Washington, D.C., 1989); T. H. Gindling, "Labor Market Segmentation and the Determination of Wages in the Public, Private-Formal, and Informal Sectors in San Jose, Costa Rica," Economic Development and Cultural Change 39 (1991): 585-605; and S. Khandker, "Labor Market Participation of Married Women in Bangladesh," Review of Economics and Statistics 69 (1987): 536-41.
13. The model for men does not include a dummy for age 21-30, so the excluded category is age 15-30. The very small number of men age 21-30 in public sector employment precluded obtaining a reasonable estimate for this covariate.
14. A sizable anthropological literature for Africa, and West Africa specifically, indicates that men and women within households do not pool income or make expenditure decisions jointly. See, e.g., E. Fapohunda, "The Nonpooling Household: A Challenge to Theory," and M. Munachonga, "Income Allocation and Marriage Options in Urban Zambia," in A Home Divided: Women and Income in the Third World, ed. D. Dwyer and J. Bruce (Stanford, Calif.: Stanford University Press, 1988); J. Guyer and P. Peters, "Introduction," Development and Change 18, no. 2 (1987): 197-214, and references therein.
15. A sizable anthropological literature for Africa, and West Africa specifically, indicates that men and women within households do not pool income or make expenditure decisions jointly. See, e.g., E. Fapohunda, "The Nonpooling Household: A Challenge to Theory," and M. Munachonga, "Income Allocation and Marriage Options in Urban Zambia," in A Home Divided: Women and Income in the Third World, ed. D. Dwyer and J. Bruce (Stanford, Calif.: Stanford University Press, 1988); J. Guyer and P. Peters, "Introduction," Development and Change 18, no. 2 (1987): 197-214, and references therein.
16. A sizable anthropological literature for Africa, and West Africa specifically, indicates that men and women within households do not pool income or make expenditure decisions jointly. See, e.g., E. Fapohunda, "The Nonpooling Household: A Challenge to Theory," and M. Munachonga, "Income Allocation and Marriage Options in Urban Zambia," in A Home Divided: Women and Income in the Third World, ed. D. Dwyer and J. Bruce (Stanford, Calif.: Stanford University Press, 1988); J. Guyer and P. Peters, "Introduction," Development and Change 18, no. 2 (1987): 197-214, and references therein.
17. Since some of the demographic variables (marriage, children, number of adults) may be simultaneously determined with an individual's participation, it would be preferable to use instrumental variable methods to control for endogeneity. However, our data set (like most data sets) lacks adequate instruments for these variables. An alternative is to drop potentially endogenous covariates, although this will yield the correct reduced form only if the eliminated variables are in fact endogenous; if they are not, there is a risk of omitted variable bias from the correlation of omitted exogenous variables with the included variables. In alternative specifications of our logit models excluding all demographic variables, the magnitudes of the coefficients on a number of the remaining variables do change substantially, but for schooling (the main policy variable of interest) the basic pattern across sectors and by sex was unaltered.
18. Both location and receiving electricity may not be exogenous to labor market choices. However, this is likely to be less of a factor for women than for men.
19. This explanation is not incompatible with the fact that men of all other ethnic groups are actually more likely than Soussous men to enter self-employment. Men's enterprises tend to be much larger than women's, and Soussou men may lack the savings, or be unable to secure the credit, to start their own businesses.
20. P. Moock et al. also use this approach to handling multiperson enterprises in their study of informal enterprises in Peru. It is implicitly assumed, among other things, that the individual with the most schooling is the primary decision maker. P. Moock, P. Musgrove, and M. Stelcner, "Education and Earnings in Peru's Informal Enterprises," Policy, Planning and Research (PPR) Working Paper WPS 236 (World Bank, Washington, D.C., 1989).
21. To construct the selectivity terms for inclusion in the earnings equations, we first reestimated the logit models after dropping 326 employed individuals from the sample (out of a total of 2,533 employed in the sample) for whom reliable earnings data were not available. Most of these were self-employed and included individuals engaged in fishing or agriculture (whose estimated net revenues tended to be implausibly low or negative); workers in multiperson family enterprises other than the most educated worker; individuals in enterprises with negative net revenues (only 1% of enterprises excluding fishing and agriculture), for whom the semilog specification could not be used; and women engaged in home activities partly for sale and partly for household consumption, for which cost data were not collected. Alternatives to dropping these observations from the first stage estimation, such as considering them as a separate category in the multinomial logit models or counting them among the nonparticipants, yielded results for both the selection terms themselves and the other regressors that were barely distinguishable from the earnings function estimates using the chosen approach.
22. j) = 0 but does not impose equality of the variances of the disturbances for the two groups, as a Chow test (based on pooling the samples and interacting each regressor with a sector dummy) would do.
23. In the semilogarithmic specification, the percentage change in earnings due to the presence of the characteristic represented by a dummy variable is 100*{exp(c) - 1}, where c is the coefficient on the variable.
24. Schooling and capital expenditures are positively correlated for male and female enterprises, indicating that educated managers have easier access to credit or better knowledge of how to employ capital in their enterprises. Alternatively, the better educated may be able to save and invest more because their profits are higher. For both men and women, dropping the capital variable from the self-employment earnings regressions results in increases in the magnitudes and t-statistics of all the schooling coefficients, suggesting that part of the overall impact of schooling comes from its association with the level of capital inputs used in the enterprise.
25. As real earnings of public employees at all skill levels were being eroded by inflation throughout the region through the early to mid-1980s, adjustments in nominal pay structures tended to favor those at lower levels, thus narrowing the range of earnings. See D. L. Lindauer, O. A. Meesok, and P. Suebsaeng, "Government Wage Policy in Africa: Some Findings and Policy Issues," World Bank Research Observer 3 (1988): 1-25. Note that evidence of wage compression is usually gleaned from examination of official government salary scales and establishment surveys rather than through estimation on micro survey data controlling for sector selection, as done here.
26. Comparisons of expected wages of men in the public and private sectors calculated from the regression estimates (thus accounting for base earnings as well as incremental impacts of schooling and other covariates) indicate premiums to working in the private sector of about 18% and 23%, respectively, for men with secondary and university education, confirming the pattern suggested by the schooling coefficients (Glick and Sahn [n. 6 above]). This is the case despite several large civil service pay increases, primarily through increases in base salaries, granted by the government in the late 1980s at the same time as the public sector workforce was being reduced (Sahn and Mills [n. 3 above]). Since the survey was conducted, a huge increase of 145% in civil service base pay in 1991 (partly the result of a national strike) has likely substantially altered the balance between public and private sector pay.
27. Note, however, that postprimary schooling may have an impact on women's earnings through its effect on the level of capital in the enterprise (see n. 18 above). In his study of Côte d'Ivoire, Vijverberg also finds that years of women's primary schooling but not postprimary schooling raises enterprise profits, although unlike our findings in this study, he finds no effects at all for male schooling. Few other studies of self-employment earnings in developing countries exist, and these do not always find impacts of women's (or men's) formal schooling. See S. Khandker, "Earnings, Occupational Choice, and Mobility in Segmented Labor Markets of India," World Bank Discussion Paper 154 (World Bank, Washington, D.C., 1992); S. Teilhet-Waldorf and W. Waldorf, "Earnings of Self-Employed in an Informal Sector: The Case of Bangkok," Economic Development and Cultural Change 31, no. 3 (1983): 585-607; and Moock et al. Note that in the first two studies, either sample sizes or the variation in schooling are quite small, making it less likely that significant education effects would be found.
28. Note, however, that postprimary schooling may have an impact on women's earnings through its effect on the level of capital in the enterprise (see n. 18 above). In his study of Côte d'Ivoire, Vijverberg also finds that years of women's primary schooling but not postprimary schooling raises enterprise profits, although unlike our findings in this study, he finds no effects at all for male schooling. Few other studies of self-employment earnings in developing countries exist, and these do not always find impacts of women's (or men's) formal schooling. See S. Khandker, "Earnings, Occupational Choice, and Mobility in Segmented Labor Markets of India," World Bank Discussion Paper 154 (World Bank, Washington, D.C., 1992); S. Teilhet-Waldorf and W. Waldorf, "Earnings of Self-Employed in an Informal Sector: The Case of Bangkok," Economic Development and Cultural Change 31, no. 3 (1983): 585-607; and Moock et al. Note that in the first two studies, either sample sizes or the variation in schooling are quite small, making it less likely that significant education effects would be found.
29. As noted, our multinomial logit specification includes household composition and exogenous income variables that do not enter the wage equations and can therefore be used to identify the selectivity term. However, the coefficients on these variables are not significant in the vector of parameters relating to the choice of private wage employment for women (table 3, col. 5), although for the other parameter subvectors (which also enter into the calculation of the lambda term) they are significant in a number of cases. As Maddala and others have noted, without identification by exclusion restrictions, the selectivity correction term is simply a nonlinear transformation of the variables appearing in the second-stage (wage) equation. This explains the high correlation with the education covariates.
30. Vijverberg (n. 2 above) for Côte d'Ivoire, J. Armitage and R. Sabot for Tanzania and Kenya, and Knight and Sabot for Tanzania (manufacturing sector only) all report returns to occupational or employment experience that are generally higher than ours. J. Knight and R. Sabot, "Labor Market Discrimination in a Poor Urban Economy,"
31. and J. Armitage and R. Sabot, "Discrimination in East Africa's Urban Labor Markets," in Unfair Advantage: Labor Market Discrimination in Developing Countries, ed. N. Birdsall and R. Sabot (Washington, D.C.: World Bank, 1991). Note that our occupational experience estimates would be somewhat higher if age, which is correlated with experience, were excluded. Even then, however, the experience impacts tend to be lower than in these studies.
32. It is assumed that purchases of capital goods are made to exactly replace the depreciated value of existing capital and that the rate of depreciation was the same across all firms. Note also that enterprise capital may be to an extent endogenous to profits. Instrumenting this variable is difficult, however, because it is hard to find variables that affect the amount of capital used but do not affect profits directly as well.
33. Men's average hourly self-employment profits are also markedly higher than average wages for men in the private wage and public sectors. A possible reason for this, which is discussed at length in Glick and Sahn (n. 6 above), is that enterprise managers underestimate costs. If the underestimation of costs (or overestimation of profits) is related to enterprise size, the male-female earnings differential will be overstated because women's enterprises are smaller.
34. Strictly speaking, the lack of rejection of equality of the male and female slope coefficients for the private and public wage sectors implies that the male and female samples can be pooled and a sex dummy included to capture gender effects. However, in view of the rather large differences between many of the point estimates in the male and female wage regressions (especially for the public sector), we have chosen to present the results of separate male and female equations for private and public wage sectors as well as self-employment.
35. f are the mean values of the explanatory variables for males and females; β* is the parameter vector from a wage equation estimated on the pooled sample of men and women and is an estimate of the wage structure that would prevail in the absence of discrimination. The first term on the right hand side represents the portion of the overall wage differential accounted for by differences in male and female characteristics. The second term represents the "male advantage" from discrimination (the amount by which men with mean male characteristics are paid in excess of the nondiscriminatory wage), and the third term represents the "female disadvantage" from discrimination (the amount by which women with mean female characteristics are paid less than the nondiscriminatory wage). The sum of the last two terms equals the total differential from discrimination. D. Neumark, "Employers' Discriminatory Behavior and the Estimation of Wage Differentials," Journal of Human Resources 23 (1988): 279-95; R. Oaxaca and M. Ransom, "On Discrimination and the Estimation of Wage Differentials," Journal of Econometrics 61 (1994): 5-12.
36. f are the mean values of the explanatory variables for males and females; β* is the parameter vector from a wage equation estimated on the pooled sample of men and women and is an estimate of the wage structure that would prevail in the absence of discrimination. The first term on the right hand side represents the portion of the overall wage differential accounted for by differences in male and female characteristics. The second term represents the "male advantage" from discrimination (the amount by which men with mean male characteristics are paid in excess of the nondiscriminatory wage), and the third term represents the "female disadvantage" from discrimination (the amount by which women with mean female characteristics are paid less than the nondiscriminatory wage). The sum of the last two terms equals the total differential from discrimination. D. Neumark, "Employers' Discriminatory Behavior and the Estimation of Wage Differentials," Journal of Human Resources 23 (1988): 279-95; R. Oaxaca and M. Ransom, "On Discrimination and the Estimation of Wage Differentials," Journal of Econometrics 61 (1994): 5-12.
37. Calculations of predicted public sector wages using the male and female parameter estimates and the data means for the full sample of men and women indicate that at each level of schooling women can expect to earn less than men with equivalent backgrounds do. This occurs despite the higher incremental schooling effects observed above for female public wage employees.
38. This concentration into a small number of traditionally "female" occupations of women in the wage work force occurs throughout the Third World. See R. Anker and C. Hein, "Why Third World Employers Usually Prefer Men," International Labor Review 124, no. 1 (1985): 73-90.
39. To do this would involve estimating the probabilities of entry of men and women into each occupation and estimating earnings functions by sex for each occupation. The approach imposes potentially severe identification requirements because entry into an occupation should be treated as endogenous to earnings. See T. Schultz, "Labor Market Discrimination: Measurement and Interpretation," in Birdsall and Sabot, eds. (n. 23 above). Our initial experiments with occupation effects were more modest and involved adding occupation dummies (treated as exogenous) to the earnings functions and decompositions.
40. W. Milne and M. Neitzert, "Kenya," in Labor Markets in an Era of Adjustment, vol. 2, ed. S. Horton, R. Kanbur, and D. Mazumdar (Washington, D.C.: World Bank, 1994); Vijverberg; Glewwe (n. 1 above); Armitage and Sabot.
41. See E. Date-Bah, "Sex Segregation and Discrimination in AccraTema: Causes and Consequences," in Sex Inequalities in Urban Employment in the Third World, ed. R. Anker and C. Hein (New York: St. Martin's Press, 1986); C. Di Domenico, "Male and Female Factory Workers in Ibadan," in Female and Male in West Africa, ed. C. Oppong (London: George Allen & Unwin, 1983); M. Peil, Cities and Suburbs: Urban Life in West Africa (New York: Africana Publishing Co., 1981).
42. See E. Date-Bah, "Sex Segregation and Discrimination in AccraTema: Causes and Consequences," in Sex Inequalities in Urban Employment in the Third World, ed. R. Anker and C. Hein (New York: St. Martin's Press, 1986); C. Di Domenico, "Male and Female Factory Workers in Ibadan," in Female and Male in West Africa, ed. C. Oppong (London: George Allen & Unwin, 1983); M. Peil, Cities and Suburbs: Urban Life in West Africa (New York: Africana Publishing Co., 1981).
43. See E. Date-Bah, "Sex Segregation and Discrimination in AccraTema: Causes and Consequences," in Sex Inequalities in Urban Employment in the Third World, ed. R. Anker and C. Hein (New York: St. Martin's Press, 1986); C. Di Domenico, "Male and Female Factory Workers in Ibadan," in Female and Male in West Africa, ed. C. Oppong (London: George Allen & Unwin, 1983); M. Peil, Cities and Suburbs: Urban Life in West Africa (New York: Africana Publishing Co., 1981).
44. Y. Cogne, "La participation des jeunes filles dans l'enseignement professionnel en Guinée" (The participation of young girls in professional education in Guinea) (Conakry, Guinea: République de Guinée, Ministère de l'Enseignement Pré-Universitaire et de la Formation Professionnelle, 1994).
45. This general conclusion emerged from a range of country studies, including several in West Africa, surveyed in Anker and Hein (n. 29 above). Conditions in Guinea are likely to be similar, but it is also necessary to gain more direct information on the Guinean case.
46. Guinea's Code de travaille prohibits discrimination in hiring on the basis of sex, but its effectiveness in practice is another matter.
47. B. Mills and D. Sahn, "Is There Life after Public Service? The Fate of Retrenched Workers in Conakry, Guinea," Cornell Food and Nutrition Policy Program Working Paper no. 42 (Cornell University, Ithaca, N.Y., 1993). This study was based on a survey of retrenched public sector workers, conducted in conjunction with the household survey used for the present article.
48. In 1993 girls represented only 25% of college (lower secondary) students, 20% of lycee (upper secondary) students, and just 6% of university students. World Bank, Developing Girls ' Education in Guinea: Issues and Policies, Report no. 14488-GUI (Washington, D.C.: World Bank, 1995).
49. Girls perform substantial amounts of household work that may interfere with their attendance and schoolwork, and they also often face a hostile classroom atmosphere because of the attitudes of teachers and male students; consequently, girls drop out and repeat grades more frequently than boys do and (if they remain in school) they have higher failure rates than boys do on exams to qualify for the next level (World Bank [n. 37 above]). In the labor market, a perception of lower quality of girls' schooling may be a reason for employers' reluctance to hire women. For earlier cohorts of students, policies under the First Republic (1958-84) that encouraged girls' education by, among other things, lowering achievement requirements for girls to enter secondary school and setting a quota for female students in the university may also have created perceptions among employers of lower quality of female graduates.
50. Female teachers have been recruited and sent to rural areas, grade repetition and student pregnancy policies have been relaxed, and greater emphasis has been given to school construction in rural areas. The last of these will improve rural enrollments overall but particularly for girls, since their enrollment is more strongly discouraged by distance to schools (World Bank).
51. Guinea's vocational education system is described in Cogne. For a study of Ghana that also emphasizes the limited training opportunities for girls, see C. Robertson, "Formal or Informal Education: Women's Education in Accra, Ghana," Comparative Education Review 28, no. 4 (1984): 639-58.
52. Note that although the training system may be used to improve gender equity in access to jobs, our earnings equations estimates indicate that the direct productivity benefits to such training are limited. Obviously, the system also needs to be improved to better impart needed skills.
53. Summing the predicted probabilities of employment in table 4 in each sector, the probability that a woman works is .23 if she has less than a primary education, .19 if she completed primary school, .23 if she completed secondary school, and .46 if she completed university. For a comparative study of female schooling and employment, see A. Smock, "Sex Differences in Educational Opportunities and Labor Force Participation Patterns in Six Countries," in Comparative Education, ed. P. Altbach, R. Arnove, and G. Kelly (New York: Macmillan, 1982).
54. A. Sow, "Enquète sur la scolarisation des filles en milieu rural: Rapport de synthèse" (Survey of girls' schooling in rural areas: Synthesis report) (Conakry: République de Guinée, Ministère de l'Enseignement Pré-Universitaire et de la Formation Professionnelle, 1994).
55. Not only are the potential benefits lower; households also face high opportunity costs of educating girls. This is because girls' contributions to household work (which are very important in rural areas) must be reduced if they are to attend school. Parents in Guinea cited this factor as a major reason for their reluctance to send girls to school, in addition to factors such as long distances to schools and coeducation of boys and girls (World Bank).