Economic Cost of Gender Gaps: Africa s Missing Growth Reserve. Amarakoon Bandara 1. Abstract

Economic Cost of Gender Gaps: Africa s Missing Growth Reserve By Amarakoon Bandara 1 Abstract In this paper we apply the dynamic GMM estimator for an endogenous growth model to analyze the impact of gender gap in effective labor defined as the combined effect of the gender gaps in labor force participation and education-on economic output per worker. Our results indicate that gender gap in effective labor has greater negative effect on the economic output per worker in African countries. A one percent increase in gender gap in effective labor leads to a reduction in output per worker by 0.43-0.49 per cent in Africa, 0.29-0.50 per cent in Sub-Saharan Africa and 0.26-0.32 per cent in a wider group of countries from Africa, and Asia. Total annual economic losses due to gender gaps in effective labor could exceed $ 60 billion for Sub-Saharan Africa while such costs could be as high as $ 255 billion for the African region as a whole. Results seem to indicate that the impact of gender gaps in effective labor could be larger in North African countries. Our results confirm the notion that Africa is missing its full growth potential as a sizeable portion of its growth reserve-women-is not fully utilized. JEL Classifications: F43, J01, J16 Key words: gender gap, labor, growth Author s e-mail address: amarakoon.bandara@undp.org 1 Economics Advisor, UNDP-Tanzania. The author wishes to thank Heri Marco for his assistance in compilation of data. The views expressed in the paper are those of the author and do not necessarily reflect those of the UNDP. Any remaining errors and omissions are the responsibility of the author. 1

Introduction Africa, in particular Sub-Saharan Africa has seen a period of remarkable growth since the beginning of 2000s after a long period of dismal record of growth. Average real gross domestic product (GDP) grew at an average of 2.2 per cent during 1987-1996 to over 5 per cent since 2004. The recent growth trend is primarily driven by export of natural resources and the flow of foreign direct investment particularly into natural resource sectors although improvements in macroeconomic conditions have provided an enabling environment. But many think Africa is missing its full growth potential. One particular growth reserve that has not been fully utilized is its people, especially its women (World Bank (2000) and (2001)). If labor is considered a critical factor of production, exclusion of a sizeable portion of women from the labor force could itself drag the economy with adverse implications for per capita income growth and poverty reduction. This could be compounded if women are also excluded from education as it affects their productivity and earning potential. Investment in people, particularly in women is important not only because Africa s future growth potential depends less on natural resources and more on skilled labor, but also due to its positive effects on poverty reduction (World Bank 2000). The average gender gaps in labor force participation in Africa and Sub-Saharan Africa during 1970-2010 are 54 per cent and 35 per cent, respectively (Table 1). 2 Some improvements have been made in recent years-the gender gap in labor force falling from 53 per cent to 48 per cent in Africa and from 35 per cent to 30 per cent in Sub-Saharan Africa, respectively during the last decade. However, as reflected in Table 1, it is not just labor force participation that matters. Access to education equally matters to make raw labor more productive. Although Africa fares well in labor force participation compared to other regions such as South Asia, the gender gap in the stock of educated labor (or effective labor as defined in section 3) seems to adversely affect Africa than elsewhere. The effect of education on labor is notable in all 2 These together imply an even higher gender gap in labor force participation in North Africa. 2

other regions. The United States with a similar level of gender gap in labor force participation as in Sub- Saharan Africa appear to have more than made up in gender gap in education to reduce the gap in effective labor. Table 1. Gender gap in labor participation and effective labor-1970-2010 Male labor Female labor Gender gap in Gender gap in participation participation labor participation(%) average years of schooling (%) Gender gap in effective labor (%) Africa 0.80 0.52 54 42 130 Sub-Saharan 0.81 0.62 35 38 79 Africa South Asia 0.83 0.34 143-16 99 South East 0.82 0.56 45-10 69 Asia USA 0.74 0.56 31 1 32 What are the growth implications of gender gap in labor force participation and effective labor? Would a reduction in gender gap in labor force participation alone good enough in recovering the lost growth due to such gaps? Or should countries focus on reducing gender gaps in both labor force and education for effective utilization of untapped human capital in women for higher growth and improved welfare? These are the questions that the current paper attempts to answer. The basic premise in this paper is that while gender gap in labor force participation itself can reduce economic growth from its potential level, the combined effect of gender gaps in labor force participation and education (reducing the effective labor) could do more damage for growth. We make two fundamental contributions to the body of knowledge on the subject under investigation. First, we estimate the gender differentiated impact of both the labor with no education and with education on growth. Ours is the first to make this distinction. Second, we not only look at the gender differentiated impact of both labor and education but also the combined effect of gender differentiation in labor and education through the introduction of effective labor. This is also a first, as previous studies have focused only on gender gap in the labor force. 3

The rest of the paper is structured as follows: Section II provides a brief literature review on gender gap and its implications on economic growth. Section III describes the modeling framework used to assess the impact of gender gaps in labor and education on economic growth. Estimation procedure and data issues are briefly discussed in Section IV. A brief analysis of estimation results and policy implications is provided in Section IV. Section V concludes. Section II: Literature Review If the finding by Lagerlöf (2003) that the European growth experience is an ultimate outcome of a gradual improvement of educational equality over the past 2000 years throws any light on the public knowledge of the impact of gender equality on economic growth, then it could very well be that gender inequality has been a social issue for thousands of years. Underlying the argument of Lagerlöf (2003) is the impact that women s education and literacy had on increased instructions for girls as well as boys. Yet, gender disparities continue to this day- an inefficient outcome given potential benefits of gender equality to the society at large. These inefficiencies exist either because institutions fail to change in response to changing economic incentives or because of market failures (Folbre 1994b). Externalities and complementarities play a crucial role in gender equality according to Lagerlöf (2003). Most of the literature on gender inequality and economic growth primarily focus on economic efficiency but few others also consider it on the ground of human rights. 3 As Folbre put it, gender inequality lead to inefficient outcomes as productive capacities are not maximized. Efficiency could come from two sources. One source is factor accumulation- the use of human capital that is not fully utilized in production could have an impact on growth. An early example of this is provided in Young (1995), which provides evidence for the fundamental role of factor accumulation in the East Asian miracle-women 3 See Mikkola and Miles (2007) and Braunstein (2007) for a review of economic literature on the relationship between gender inequality and economic development. 4

being a significant source of it. Using a growth accounting framework to decompose sources of growth, Young finds that rising participation of women in the labor force contributed to increases in per capita growth per annum in Hong Kong (1%), Singapore (2.6%), South Korea (1.2%), and Taiwan (1.3%). The second source is productivity-productivity could increase by increased participation of women in the labor force resulting in higher growth. For example, with decreasing marginal returns to education, educating more girls (who start out with lower education than boys due to gender inequalities) will yield higher marginal returns for girls than boys (Knowles, Lorgelly and Owen 2002; Schultz 2001; World Bank 2001). Applying a framework based on Solow s approach, Knowles, Lorgelly and Owen (2002) find that GDP per worker elasticity with respect to female education to be between 0.2 and 0.45 although male education productivity elasticity is not statistically significant or slightly negative. An alternative argument by Klasen (1999) is that if male and female students have equal aptitudes, then educating more boys than girls will lower the overall quality of educated individuals and thereby productivity via selection distortion effects. Using panel growth regressions for the period 1960-1992, it finds that gender inequality in employment negatively impacts growth-south Asia and Sub-Saharan Africa suffering losses of 0.3 per cent per year compared to East Asia. Similar results are found in Hill and King (1995), Blackden and Bhanu (1999) and Benavot (1989). Hill and King (1995) find that countries whose ratio of female-to-male enrollments in primary or secondary schooling is less than 0.75, GNP is 25 per cent lower than in countries who are otherwise similar except for gender gap in education. Blackden and Bhanu (1999) find that African growth rate could be as much as 0.8 percentage points higher if women were given equal access to education and productive assets. The latter also finds that Sub-Saharan Africa would have had an annual per capita growth rate 0.5 percentage points higher, if it had improved its female-male ratio of education to the level of East Asia. Dollar and Gatti (1999) finds 5

either a positive relationship between female education and growth or an insignificant effect depending on the sample. Baliamoune-Lutz and McGillivray (2007) using a panel data set for African and Arab countries find gender inequalities in literacy to have a negative effect on growth. 4 Not all studies find a positive relationship between gender equality and growth. For example, early neoclassical studies such as Barro and Lee (1994) and Barro and Sala-i-Martin (1995) find a negative correlation between women s schooling and economic growth. But Abu-Ghaida and Klasen (2004) and World Bank (2001) criticize these outcomes largely to be a result of the multicollinearity between male and female education, as well as the influence of Latin American countries which tended to have greater gender equality in education and low growth. These studies however, raise the question of causation: does improving the status of women cause greater wealth, or is gender equality a consequence of economic development. For example, in Miles (2005 and 2007) and Dollar and Gatti (1999), the causation runs from growth to gender equality. While Miles finds strong correlation between societal wealth and increased women s employment among others, Dollar and Gatti discover that increases in income lead to less gender inequality in education. Section III: The model We start with the aggregate production function 1 0 (1) where is output (GDP) at time t. and denote stock of physical capital and stock of human capital, respectively. is the share of physical capital in national output. is total factor productivity 4 Several studies analyze the implications of gender inequality on development. While Stotsky (2006) analyzes the implications of gender inequality on macroeconomic policy, Doepke (2011) investigates how female empowerment promotes economic development. 6

(TFP) reflecting technical change and innovation. We define human capital stock in the following manner: (2) where the size of the labor force is multiplied by the average efficiency units embodied in the workers that comprise the labor force. In effect this reflects the effective labor-labor with certain level of knowledge acquired through education or training. denotes the average years of schooling attained by a worker at time t and the derivative is the return to education estimated in the Mincerian wage regression (Mincer 1974). 0=0, so that a worker with no education has his/her own raw labor while a worker with E years of schooling owns efficiency units of labor. If =0 for all E then (1) and (2) reflect a standard production function with undifferentiated labor. We assume to be a log linear function defined as: (3) where and are the ratios of female and male labor with no education in female and male labor force, respectively. and denote the stock of female and male education, respectively. From (2) and (3) we can rewrite as:.... 1 and 0, 0 (4) Where and are the stock of female and male labor, respectively. It should be noted that. = and. = are the female and male stock of labor, respectively with no education, thus providing only raw labor.. = and. = are, respectively, the stock of female and male effective labor. 7

Substituting equation (4) in (1) and dividing both sides by we have (5) where lower case letters denote quantities per unit of labor. Following Mankiw, Romer, and Weil (1992), the labor force is assumed to be determined by (6) where n is the growth rate of the labor force. Thus the accumulation of physical capital can be given by (7) where is the fraction of real output invested in physical capital and is the rate of depreciation, which is assumed to be common across countries and time. Assuming the existence of a steady state, equation (7) gives us the steady state physical capital per effective unit of labor, (8) Substituting equations (6) and (8) into (5) and taking natural logarithms and rearranging gives ln (9) Following Nelson and Phelps (1966), we assume that the rate of change of TFP in a country is positively related to the size of the gap between its actual TFP at a point in time and its potential TFP at that time. (10) 8

Here represents the country s potential level of TFP at time t, which we assume to follow the following rule (11) where is an index of fixed factors specific to the country, and is an index of technology which grows over time. is a vector of exogenous factors that drives TFP. These factors could include human capital, stock of Research and Development (R&D), macroeconomic stability, financial market developments, and external sector developments among others. 5 In our model we assume inflation and trade openness to be key factors in driving TFP. It should be noted that human capital is already a key factor of production in our model through effective labor and foreign direct investment is captured in the capital stock. We are constrained in using R&D as a variable due to data limitations. Inflation is an outcome of macroeconomic management and a better indicator to gauge how well the economy is managed. Trade openness could also reflect how attractive the country is for private investment, including foreign direct investments and technical change. With these assumptions and taking natural logarithm of (9) and with the addition of a disturbance term we get ln (12) where is a fixed effects term, is inflation rate to reflect macroeconomic stability, is trade openness, is a time trend and is a random error term. Substituting equation (12) in (9) we get ln (13) 5 See Akinlo (2006) 9

where /1 is a random error term. We can re-parameterize our model for the gender gap in effective labor to enter the model directly. Rearranging equation (13) gives us ln (14) Essentially this represents a traditional fixed effects model. and are the gender gaps in labor with no education and effective labor. We expect the sign of the coefficient of the gender gap in effective labor,, to be negative. As 1, this implies 0. Section IV: Estimation procedure and data The choice of the econometric technique to estimate a model depends to a large on data availability and the sample size, and recent developments in techniques that enable addressing some of the critical issues relating to regression. The data we use in this analysis is a set of an unbalanced panel data. The use of a panel substantially increases the efficiency and power of the analysis as the degrees of freedom increases in a panel of several countries (Goodhart and Hofmann (2008)). This is particularly so for Sub- Saharan African countries where availability of data is an issue. A panel approach also helps uncover common dynamic relationships which might otherwise be obscured by idiosyncratic effects at the individual country level (Gavin and Theodorou (2005)). However, the use of panel data also raises some estimation issues. One such issue is parameter heterogeneity across cross sections leading to inconsistent and misleading results of short-term effects of shocks. Questions have been raised on the appropriateness of standard techniques for estimating systems with panel data as the pooling of data from different cross sections imposes the constraint that the underlying structure is the same for each cross-sectional unit. The constraint that the time series 10

relationship of variables is the same for each cross-section is likely to be violated in practice and it is desirable to relax this restriction. The problem arises because when the regressors are serially correlated, incorrectly ignoring coefficient heterogeneity induces serial correlation in the disturbances (Pesaran and Smith 1995). This generates inconsistent estimates in models with lagged dependent variables even as. The bias of the estimator is likely to be serious when is small. They proposed using the mean group (MG) estimator, which provides consistent estimates of the mean effects by averaging the coefficients across countries. Later, Pesaran, Shin and Smith (1999) developed the Pooled Mean Group (PMG) estimator, which allows the intercepts, short-run coefficients and error variances to differ across groups, but constrains the long-run coefficients to be identical. The OLS and (and GLS-type methods such as SURE estimation) yield consistent estimates only if the country-specific error terms are uncorrelated with the explanatory variables. A widely used technique in response to this problem is GMM estimator in a dynamic panel specification (Caselli et al., 1996; and Forbes, 2000). In an alternative approach proposed by Cornwell et al. (1990) to account for the unobserved heterogeneity in the data, they allow the intercept to vary across cross sections. In the fixed effects model they used, the vector of individual specific effects contains in addition to a constant, a time trend and a squared time trend allowing individual effects to vary over time. In a study similar to ours, Knowles, Lorgelly and Owen (2002) apply OLS in cross-section regressions with time-averaged data to estimate the parameters in the long-run steady-state levels relationship. Their application has been motivated by recent developments in regression limit theory for non-stationary panel data by Pesaran and Smith (1995), and Phillips and Moon (1999, 2000). A second issue relates to causality-although we assume the variables in the right hand side of the regression equation to be endogenous, causality could run in both directions. Simultaneity is a potential source of bias in estimates from cross-country regression. If the causality runs from the dependent 11

variable to any of the right hand side variable then these regressors may be correlated with the error term. One way to address these problems is to verify the causality and exclude any regressors that are influenced by the dependent variable. To solve this problem one would usually use fixed effects instrumental variables estimation (such as Two Stage Least Squares) but the estimates could be biased if the instruments are weak. Time invariant country characteristics (fixed effects) may also be correlated with the explanatory variables. An alternative approach, proposed by Arellano and Bond (1991), differences the endogenous and predetermined variables and uses lags of their own levels as instruments to deliver consistent estimates. We use dynamic GMM estimator to estimate the models in (13) and (14) and their variations in an unbalanced five year averaged panel data set for the period 1970-2010. The GMM estimator delivers unbiased results in the presence of endogenous explanatory variables. Our primary focus is on Africa and Sub-Saharan Africa. However, comparisons are also made against a panel comprising countries from Africa, Sub-Saharan Africa, South Asia, and South East Asia. We also perform Sargan test of overidentifying restrictions and Arellano-Bond 2-Step estimator to test for autocorrelation in differenced residuals. Data Most of the data used for this estimation are from the World Development Indicators of the World Bank. 6 These include the inflation rate based on the percentage change in the consumer price index, trade openness derived from the sum of imports and exports as a percentage of GDP, labor participation rates for males and females aged 15 and over, primary enrolment rate for girls and boys and the male and female average years of schooling of the population aged 15 and over. Real GDP, real GDP per worker, gender disaggregated labor force, and physical capital stock reflecting the investment share of 6 A summary of data used in the estimations for Africa are provided in Appendix Table A8. 12

PPP converted GDP per capita are from Penn World Tables/Table Statistics. is assumed to be 5%. We use unbalanced panel data for 53 African countries, including 44 countries from Sub-Saharan Africa. The wider group of countries also includes 9 countries each from South Asia and South East Asia. Data are for five year averages for the period 1970-2010. Section V: Estimation results and analysis Dynamic panel GMM estimation results for Africa are provided in Appendix Tables A1 and A2. Column (i) reports the estimation results for equation (13). We observe that openness of the economy is positively but weakly impacts on output per worker while female effective labor has a positive and significant effect. Labor with no education has a negative effect on output regardless of gender considerations. The effect of male effective labor is negative but weak. In column (ii) we report estimation results of equation (13), but excluding inflation as we reject the null hypothesis of economic output does not Granger cause inflation. 7 We note that earlier results become stronger. In columns (iii) and (iv), respectively, are the estimates with gender disaggregated labor and education variables entering the model separately. Note however, that in the specification in column (iii) we reject the null hypothesis of no autocorrelation in differenced residuals as per Arellano-Bond test. The results may be biased and inconsistent. The model in column (v) is similar to that in (iv) except that education variable is now disaggregated by the level of education (with and without education). Both male and female labor has significant positive effects on output. Appendix Table A2 provides estimation results of equation (14) and its variations, where the gender gap enters the model separately. Column (i) presents results for the model represented in equation (14). We observe that openness positively impacts output while inflation has a negative but insignificant effect. 7 The Sargan test results for all estimated models indicate that the instrument sets used in them are valid 13

Female labor with no education has a negative effect on output. The gender gap in labor with no education (labor that provides raw labor) negatively affects output. Notable is the significant and large negative impact of the gender gap in effective labor (labor with education) on economic output. A higher gender gap in effective labor tends to reduce output. 8 Column (ii) repeats the estimation in column (i) but excluding the inflation as before. The results become stronger and more significant. The gender gap in effective labor has a greater impact while the gender gap in labor with no education becomes insignificant. Knowles, Lorgelly and Owen (2002), which estimates a model that includes female and male education as separate explanatory variables, contend that the interpretation of the coefficient of the gender gap in education depends crucially on what education variables appear in the equation. Noting their concerns, we remove the other effective labor variables in estimating the model in column (iii). This also allows us to interpret the coefficients of the gender gaps as in Hill and King (1995)- a highly significant coefficient on the gap term to imply a greater effect on the independent variable, output per worker. Column (iv) provides similar estimates with inflation also excluded. Both estimates indicate somewhat lesser impact of the gender gap in effective labor on output per worker in comparison to that in Column (ii) but still larger than that in Column (i). Estimates in column (iv) indicate the gender gap in education to have a significant negative effect on output but not in the gender gap in the labor force. The results indicate that a one percent increase in the gender gap in effective labor reduces output per worker by 0.43-0.49 per cent in African countries. The models are also applied to Sub-Saharan African countries, results of which are reported in Appendix Tables A3 and A4. The results could also be treated as a robustness test. Estimation results of model in equation (13) are given in column (i) in Table A3. In Sub Saharan African countries, female labor with no 8 Knowles, Lorgelly and Owen (2002) interprets the coefficient on the gap differently-it reflects the output elasticity with respect to male education. 14

education has a negative impact on output while female effective labor has a positive effect. Male effective labor has a negative but weakly insignificant effect on output. When inflation is excluded (Column (ii)), as we reject the null hypothesis of economic output does not Granger cause this variable (although weakly), most regressors become stronger and significant except labor variables. Both openness and physical capital variables become significant having positive effects. The model in column (iii) fails to reject autocorrelation between regressors and the error term. When gender disaggregated labor and education enter the model as separate explanatory variables (columns (iv) and (v)), education or labor alone does not seem to have an impact on output, except in the model in column (iv) where male education has a positive effect on output. Models in Appendix Table A4 introduce the gender gap. Results in column (i) indicate female labor to have significant negative effects regardless of their level of education as does the gender gap in effective labor. Openness and physical capital do not have any significant effect on output. However, exclusion of inflation from the model due to its possible autocorrelation with the error term improves the impact of the latter two variables greatly (column (ii)). Gender gap in effective labor remains significant with a slightly lower degree of impact on output. Exclusion of other effective labor variables from the model produces similar results but gender gap in effective labor having a lesser impact (Column iv). As for the African region, Sub-Saharan African countries also seem to suffer from gender gap. On average a one percent deterioration in gender gap in effective labor leads to a reduction of output by 0.29-0.50 per cent. We also apply the model in equation (13) to a wider group of countries representing Africa and Asia. We disregard the results in Appendix Table A5 except those in column (ii) and (v) as we reject the null hypothesis of no autocorrelation between regressors and the error term. Estimates in column (ii) indicate that female raw labor has a negative effect on growth while female effective labor has a 15

positive effect. Male effective labor has a negative impact but insignificant. Estimates in column (v) indicate that while female education has a positive impact on growth. Estimates of model in equation (14) and its variations for this group of countries are presented in Appendix Table A6. Results in column (i) indicate a significant and negative impact of gender gap in effective labor on output per worker. Estimation results excluding inflation and gender gap in raw labor variables are given in column (ii). The results indicate the stronger impact of physical capital. The impact of gender gap in effective labor is weak but statistically more significant. Female labor has a negative impact. Estimation of the model excluding other effective labor variables (column (iii)) results in a stronger effect of physical capital. Gender gap in effective labor continues to have a significant effect. Gender gap in labor with no education now turns positive. On the other hand, exclusion of above variables along with those that are auto-correlated with the error term (column (iv)) reduces the impact of gender gap in effective labor significantly. Estimates in column (v), where the gender gaps in education and labor participation enters the model separately, indicate that gender gaps in education and labor have significant negative effects on output. Physical capital continues to have a positive impact. The gender gap in effective labor has a significant negative effect on output per worker in this group of countries. A one percent increase in gender gap in effective labor reduces output per worker by around 0.26-0.32 per cent. Relatively lower gaps in effective labor in South East Asia seem to have positively influenced the results. Our results are robust to changes in the number of cross-country units across different regions. While there are some variations in point estimates, the sign of the coefficient of the gender gap in effective labor is the same and significant. While variations in impact of most variables on output per worker do exist among the samples we have examined, the direction is more or less the same. Physical capital and openness have positive effects on output per worker. Gender gap in effective labor is higher in Africa than in Sub-Saharan Africa or the 16

larger sample including South Asia and South East Asia, influenced by the wider gender gaps in labor force and education in North African countries. All experiments indicate the large economic losses due to gender gaps in effective labor. The results indicate annual economic losses due to gender gaps in effective labor exceeding $ 60 billion for Sub-Saharan Africa. 9 Economic costs of gender gaps in effective labor could be as high as $ 255 billion for the Africa region as whole as gaps in effective labor in most North African countries are much higher than in Sub-Saharan Africa. What matters most appear to be labor with certain level of knowledge that could be utilized for increased productivity than raw labor. While gender gaps in labor participation and education among others are nothing new to policy makers around the world including Africa, the striking gains from narrowing these gaps could provide a boost for concerted action, particularly at the national level. Economic and social gains by promoting gender equality in poverty stricken countries could be substantial as women tend to be the ones who suffer the most in such circumstances. Section VI: Conclusion Gender gap in labor participation is high and common in African countries. So is in education. While most studies focus on the impact of the gender gap in education or the labor force separately on growth, we deviate from this approach to consider the combined effect of the gender gaps in labor force participation and education-what we call effective labor. The impact could be huge. A classic example is the USA where the gender gap in labor participation is no different from Sub-Saharan Africa. But their gender gap in effective labor is only a half of Sub-Saharan African countries due to a very narrow gender gap in the stock of education. In this line of thinking, the basic premise of the paper has been that while 9 The economic costs for regions are estimated by multiplying real GDP by the gender gap in effective labor and the estimated coefficient from the model in equation (14). Regional real GDP for the regions are the sum of respective country real GDPs. A similar approach is used for computing the effective labor The formula for calculating the cost is given by: / 1 17

gender gap in labor participation itself can reduce growth from its potential level, the gender gap in effective labor (the combined effects of gender gaps in labor participation and education) could be more damaging for growth. We apply the dynamic GMM estimator for an endogenous growth model to analyze the impact of gender gap in effective labor on economic output per worker. Arellano-Bond step-2 test is applied to verify autocorrelation between regressors and the error term. We experiment with variations of the model for three regional groupings, Africa, Sub-Saharan Africa and a larger group of countries from Africa and Asia. We find that the gender gap in effective labor has a significant negative effect on output per worker. A significant negative effect of the gender gap in education when enters the model as separate variables could indicate that the gap in effective labor arises mainly from the gender gap in education than in labor participation. Our results indicate that gender gap in effective labor has greater negative effect on the economic output per worker in African countries than elsewhere. A one percent increase in gender gap in effective labor leads to a reduction in output per worker by 0.43-0.49 per cent in Africa, 0.29-0.50 per cent in Sub-Saharan Africa and 0.26-0.32 per cent in the wider group of countries from Africa, and Asia. A comparison of African and Sub-Saharan African countries indicates that the impact of gender gaps in effective labor could be larger in North African countries. Our results confirm the notion that Africa is missing its full growth potential as a sizeable part of its growth reserves-womenis not fully utilized. The results indicate annual economic losses exceeding $ 159 billion for Africa and $ 73 billion for Sub-Saharan Africa due to gender gaps in effective labor. 18

References Abu-Ghadia, D. and Klasen, S. (2004), The Costs of Missing Millennium Development Goals on Gender Equity, World Development, 32(7): 1075-1107. Arellano, M. and Bond, S. (1991), Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations, The Review of Economic Studies, 58, 277-97. Baliamoune-Lutz, M. and McGillivray, M. (2007), Gender Inequality and Growth: Evidence from Sub- Saharan Africa and Arab Countries, available at http://www.uneca.org/aec/documents/mina%20baliamoune-lutz.pdf Barro, R.J. and Lee, J. (2001), International Data on Educational Attainment Updates and Implications, Oxford Economics Papers, 3: 541-563. Barro, R.J. and Sala-i-Martin, X. (1995), Economic Growth, New York, McGraw-Hill. Bils, M. and Klenow, P. (2000), Does Schooling Cause Growth?, American Economic Review, 90(5):1160-1183. Blackden, C.M. and Bhanu, C. (1999), Gender, Growth and Poverty Reduction. Special Programme of Assistance for Africa, 1998 Status Report on Poverty in Sub-Saharan Africa. World Bank Technical Paper No 428. Braunstein, E. (2007), The Efficiency of Gender Equity in Economic Growth: Neoclassical and Feminist Approaches, The International Working Group on Gender, Macroeconomics and International Economics Working paper No 07-4. Caselli, F., Esquivel, G., and Lefort, F. (1996), Reopening the convergence debate: a new look at crosscountry growth empirics, Journal of Economic Growth, 1, 363-89. 19

Doepke, M. and Tertilt, M. (2011), Does Female Empowerment Promote Economic Development?, World Bank Policy Research Working Paper No WPS 5714. Dollar, D. and Gatti, R. (1999), Gender Inequality, Income and Growth: Are Good Times Good for Women? World Bank Policy Research Paper on Gender and Development Working Paper Series, No 1. Folbre, N. (1994a), Children as Public Goods, American Economic Review Papers and Proceedings, 84(2):86-90. ---------------------, (1994b), Who Pays for Kids?, Gender and the Structures of Constraints, London and New York, Routledge. Forbes, K.J. (2000), A reassessment of the relationship between inequality and growth, American Economic Review, 90, 869-87. Gavin, W. and Theodorou, A. (2005). A common model approach to macroeconomics: using panel data to reduce sampling error. Journal of Forecasting, 24,203-209. Goodhart, C. E. and Hofmann, B. (2008). House Prices, Money, Credit and the Macroeconomy. Oxford Review of Economic Policy, 24(1), 180-205. Hill, M. A. and King, E.M. (1995), Women Education and Economic Well-Being, Feminist Economics, 1:21-46. King, E.M. and Hill, M.A. eds, (1993), Women s Education in Developing Countries, Baltimore: John Hopkins University Press. Klasen, S. (1999), Does Gender Inequality Reduce Growth and Development?, Evidence from Cross- Country Regressions, World Bank Policy Research Report on Gender and Development, Working Paper Series, No. 7. 20

-------------------, (2002), Low Schooling for Girls, Slower Growth for All? Cross-Country Evidence on the Effects of Gender Inequality in Education on Economic Development, The World Bank Economic Review, 16(3):345-373. Knowles, S. Lorgelly, P.K. and Owen, P.D. (2002), Are educational gender gaps a break on economic development? Some cross-country empirical evidence, Oxford Economic Papers, 54:118-149. Lagerlöf, N. (2003), Gender Equality and Long-run Growth, Journal of economic Growth, 8(4):403-426. Lorgelly, P. and Owen, D. (1999), The Effects of Male and Female Schooling on Economic Growth in the Barro-Lee Model, Empirical Economics, 24:537-557. Mankiw, N.G. Romer, D. and Weil, D. (1992), A Contribution to the Empirics of Economic Growth, Quarterly Journal of Economics, 1992, 107:407-437. Mikkola, A and Miles C.A. (2007), Development and Gender Inequality: Consequences, Causes, Challenges and Cures, HECER Discussion paper No 159. Miles, C.A. (2006), Patriarchy or Gender Equality? Reading the Apostle Paul in light of ancient history and modern economics, Dialogue, June 2006 ---------------------, Because Women are Property: Issues of Gender, Food security, Propeety Ownership and quasi Development and Religion in Sub Saharan Africa, paper presented at the 2007 meeting of UNU-Wider in Accra, Ghana, May 2007, available at http://www.empowerinternational.org/empower_bu/articles/miles%20women%20as%20property%20 submission.pdf Mincer, J. (1958), Investment in Human Capital and Personal Income Distribution, Journal of Political Economy, 66(4):281-302. 21

Nelson, R. and Phelps, E. (1966), Investment in Humans, Technological Diffusion, and Economic Growth, American Economic Review, 56:69-75 Pesaran, M.H. Shin, Y. and Smith, R. (1999), Pooled Mean Group Estimation of Dynamic Heterogeneous Panels, Journal of American Statistical Association, 94, 621-634. Pesaran, M.H. and Smith, R. (1995). Estimating Long-Run Relationships from Dynamic Heterogeneous Panels. Journal of Econometrics, 68, 79-113. Phillips, P.C.B. and Moon, H.R. (1999), Linear regression limit theory for nonstationary panel data, Econometrica, 68, 79-113. --------------------. (2000), Nonstationary panel data analysis: an overview of some recent developments, Econometric Reviews, 19, 263-86 Stotsky, J.G. (2006), Gender and its Relevance to Macroeconomic Policy: A Survey, IMF Working paper No WP/06/233. World Bank, (2001), Engender Development Through Gender Equality in Rights, Resources and Voice, Washington, D.C. and New York: World Bank and Oxford University Press. -------------------, (2000), Can Africa Claim the 21 st Century?, Washington D.C.: The World Bank. Young, A. (1995), The Tyranny of Numbers: Confronting the Statistical Realities of the East Asian Growth Experience, The Quarterly Journal of Economics, 110 (3, August): 641-680. 22

Appendix Table A1: GMM Estimates of equation 13 and its variations for Africa Dependent variable: real GDP per worker Variable Coefficient (Standard Error) (i) (ii) (iii) (iv) (v) 0.741*** (0.064) 0.696*** (0.089) 0.587*** (0.041) 0.593*** (0.021) -0.012-0.007 0.208*** -0.021 (0.051) (0.019) (0.042) (0.020) 0.001 0.004-0.035*** (0.013) (0.023) (0.006) 0.103* 0.245*** -0.267*** 0.111*** (0.054) (0.037) (0.078) (0.036) 0.170 0.347*** 0.379*** -0.030 (0.115) (0.057) (0.092) (0.021) -0.066*** -0.039*** (0.014) (0.014) -0.046*** -0.020** (0.011) (0.008) (0.069) (0.089) 0.291*** 0.379*** -0.304* -0.321*** (0.165) (0.093) 1.088*** 0.593*** (0.156) (0.153) -0.625 1.640*** (0.821) (0.367) 0.035 (0.080) -0.127 (0.095) 23 0.683*** (0.148) -0.152* (0.083) -0.052** (0.025) 0.263* (0.140) -0.007 (0.116) 1.817*** (0.459) 4.755*** (1.626) 0.033 (0.110) 0.213 (0.179) -0.015 (0.027) -0.104*** (0.019) -0.037 (0.032) -0.021 (0.016) -0.002 (0.015) -0.055*** (0.019) -0.019 (0.017) J-statistic 14.629 19.309 19.821 21.555 13.439 Instrument Rank 29 28 27 32 29 Chi2(prob>chi2) a 0.74 0.44 0.40 0.48 0.70 Wald test: Chi2 b 0.39 1.42 7.47 3.62 0.71 prob 0.53 0.23 0.00 0.56 0.39 Note: ***,**,* denote 1%, 5% and 10% level of significance, respectively. a Sargan test of over-identifying restrictions b Arellano-Bond test that average autocovariance in residuals of order 2 is 0.

Table A2: GMM Estimates of equation 14 and its variations for Africa Dependent variable: real GDP per worker Variables Coefficient (Standard Error) (i) (ii) (iii) (iv) (v) 0.827*** (0.079) 0.692*** (0.082) 0.727*** (0.053) 0.845*** (0.084) 0.117* 0.014 0.027 0.025 (0.061) (0.027) (0.023) (0.022) -0.003 0.009 (0.011) (0.015) 0.143** 0.235*** 0.133*** 0.426*** (0.057) (0.104) (0.034) (0.074) 0.159 0.314*** 0.351*** 0.366*** (0.117) (0.065) (0.066) (0.036) -0.115*** -0.068*** (0.020) (0.019) -0.035*** -0.009-0.002-0.009 (0.011) (0.008) (0.005) (0.006) (0.617) (0.119) -0.423*** -0.390*** -0.419*** -0.649*** -0.432*** -0.494*** (0.115) (0.125) (0.071) (0.067) 0.602*** (0.026) 0.049* (0.027) -0.030*** (0.007) 0.034 (0.041) -0.010 (0.031) -0.391*** (0.112) -0.097 (0.063) -0.023* (0.012) -0.023 (0.035) 0.042 (0.016) 0.008 (0.015) -0.003 (0.014) J-statistic 16.268 16.194 23.431 21.369 20.309 Instrument Rank 29 28 29 28 29 Chi2(prob>chi2) a 0.63 0.64 0.32 0.43 0.50 Wald test: Chi2 b 2.60 2.85 0.38 3.21 4.91 Prob 0.10 0.09 0.54 0.07 0.03 Note: ***,**,* denote 1%, 5% and 10% level of significance, respectively. a Sargan test of over-identifying restrictions b Arellano-Bond test that average autocovariance in residuals of order 2 is 0. 24

Table A3: GMM Estimates of equation 13 and its variations for Sub-Saharan Africa Dependent variable: real GDP per worker Variables Coefficient (Standard Error) (i) (ii) (iii) (iv) (v) 0.909*** (0.112) 0.693*** (0.089) 0.587*** (0.072) 0.614*** (0.016) -0.021 0.053-0.121** -0.021 (0.048) (0.031) (0.053) (0.019) -0.013 (0.010) 0.146 0.201*** 0.346*** 0.224*** (0.089) (0.021) (0.079) (0.026) 0.144 0.164** 0.039 0.311*** (0.125) (0.074) (0.039) (0.057) -0.115** -0.063 (0.048) (0.046) -0.020-0.043 (0.050) (0.029) (0.205) (0.062) 0.470** 0.064-0.490* -0.014 (0.264) (0.115) 1.858 0.429 (1.340) 5.750*** (2.018) (0.509) -0.069 (0.851) 0.080 (0.075) 0.174*** (0.066) 0.826*** (0.090) 0.001 (0.051) 0.204* (0.115) 0.077 (0.145) -0.226 (1.698) -0.343 (3.121) 0.577 (0.430) -0.839 (0.638) 0.006 (0.103) -0.085 (0.077) 0.016 (0.021) -0.023 (0.033) 0.010 (0.020) 0.030** (0.011) -0.004 (0.008) J-statistic 11.985 16.925 23.245 24.176 20.270 Instrument Rank 23 26 27 30 26 Chi2(prob>chi2) a 0.52 0.39 0.27 0.28 0.16 Wald test: Chi2 b 0.003 0.65 57.3 3.88 0.001 prob 0.95 0.41 0.00 0.48 0.96 Note: ***,**,* denote 1%, 5% and 10% level of significance, respectively. a Sargan test of over-identifying restrictions b Arellano-Bond test that average autocovariance in residuals of order 2 is 0. 25

Table A4: GMM Estimates of equation 14 and its variations for Sub-Saharan Africa Dependent variable: real GDP per worker Variables Coefficient (Standard Error) (i) (ii) (iii) (iv) 0.937*** (0.105) 0.746*** (0.072) 0.854*** (0.078) 0.121 0.048 0.051*** (0.087) (0.033) (0.016) 0.0001 (0.013) 0.137 0.176** 0.304*** (0.084) (0.086) (0.091) 0.184 0.217*** 0.243*** (0.148) (0.047) (0.058) -0.161*** -0.098*** (0.044) (0.024) -0.032-0.001-0.076*** (0.046) (0.018) (0.011) (0.275) (0.114) -0.631** -0.336*** -0.643*** -0.506*** -0.291*** (0.276) (0.123) (0.074) 0.604*** (0.013) -0.026 (0.021) 0.254*** (0.026) 0.310*** (0.061) -0.140 (0.125) -0.252** (0.105) 0.013 (0.010) 0.067 (0.052) 0.037** (0.018) 0.005 (0.015) J-statistic 12.997 18.931 18.487 21.978 Instrument Rank 24 26 25 30 Chi2(prob>chi2) a 0.52 0.33 0.42 0.52 Wald test: Chi2 b 2.62 0.17 0.108 7.19 prob 0.105 0.67 0.74 0.007 Note: ***,**,* denote 1%, 5% and 10% level of significance, respectively. a Sargan test of over-identifying restrictions b Arellano-Bond test that average autocovariance in residuals of order 2 is 0. 26

Table A5: GMM Estimates of equation 13 and its variations for Africa, South Asia and Africa, South Asia and South East Asia Dependent variable: real GDP per worker Variables Coefficient (Standard Error) (i) (ii) (iii) (iv) (v) 0.687*** 0.634*** 0.570*** 0.518*** 0.630*** (0.073) -0.099** (0.045) -0.031*** (0.009) 0.154*** (0.023) 0.052 (0.049) -0.058*** (0.020) -0.041*** (0.008) 0.302*** (0.111) -0.171 (0.142) (0.062) -0.010 (0.018) 0.070** (0.032) 0.207*** (0.046) -0.045*** (0.006) -0.004 (0.006) 0.164** (0.065) -0.108 (0.066) 27 (0.036) 0.023 (0.018) 0.156*** (0.029) 0.089* (0.053) -0.419 (0.325) 0.894** (0.441) (0.040) -0.159*** (0.022) 0.067* (0.039) 0.331*** (0.049) -0.306 (0.200) 0.905*** (0.334) 0.268** (0.112) 0.202* (0.121) (0.078) -0.032 (0.042) 0.059 (0.053) 0.196*** (0.061) 0.104 (0.382) -0.207 (0.965) 0.167*** (0.050) -0.087 (0.067) -0.050*** (0.019) -0.001 (0.010) -0.029* (0.016) -0.064*** (0.012) -0.022*** (0.008) -0.024*** (0.006) 0.023** (0.010) J-statistic 20.629 21.710 30.890 24.177 21.499 Instrument Rank 30 29 35 38 33 Chi2(prob>chi2) a 0.41 0.35 0.32 0.72 0.49 Wald test: Chi2 b 7.74 1.31 68.01 27.29 1.23 prob 0.00 0.25 0.00 0.00 0.26 Note: ***,**,* denote 1%, 5% and 10% level of significance, respectively. a Sargan test of over-identifying restrictions b Arellano-Bond test that average autocovariance in residuals of order 2 is 0.

Table A6: GMM Estimates of equation 14 and its variations for Africa, South Asia and South East Asia Dependent variable: real GDP per worker Variables Coefficient (Standard Error) (i) (ii) (iii) (iv) (v) 0.775*** (0.058) 0.669*** (0.050) 0.795*** (0.043) 0.629*** (0.040) 0.676*** (0.037) -0.010 (0.075) 0.054 (0.036) 0.024 (0.032) 0.024 (0.022) 0.103*** (0.017) -0.016 (0.011) -0.012 (0.011) -0.003 (0.009) 0.107** (0.041) -0.018 (0.051) 0.133*** (0.038) 0.005 (0.040) 0.019 (0.030) 0.189*** (0.059) 0.233*** (0.049) 0.271*** (0.045) 0.133*** (0.043) 0.281*** (0.043) -0.043*** (0.016) -0.068*** (0.011) -0.013 (0.011) 0.008 (0.007) 0.008-0.367*** (0.218) -0.480** (0.190) (0.088) -0.325*** (0.073) -0.468*** (0.103) -0.259** (0.072) -0.233** (0.097) -0.139** (0.066) -0.017** (0.006) -0.055* (0.030) -0.028 (0.017) -0.050*** (0.012) -0.008 (0.007) J-statistic 23.442 22.396 26.365 18.728 28.575 Instrument Rank 30 29 30 28 31 Chi2(prob>chi2) a 0.26 0.37 0.23 0.66 0.19 Wald test: Chi2 b 0.22 3.49 0.21 0.051 5.09 prob 0.63 0.06 0.64 0.82 0.02 Note: ***,**,* denote 1%, 5% and 10% level of significance, respectively. a Sargan test of over-identifying restrictions b Arellano-Bond test that average autocovariance in residuals of order 2 is 0. 28

Table A7: Pairwise Granger Causality Tests (Sample: 1970 2010, Lags: 2) Africa Null Hypothesis: Obs F-Statistic Prob. does not Granger Cause 2.58036 0.0779 does not Granger Cause 237 3.41163 0.0347 SSA does not Granger Cause 3.22469 0.0419 does not Granger Cause 198 2.59066 0.0776 Africa-SA-SEA does not Granger Cause 3.19214 0.0426 does not Granger Cause 283 8.33070 0.0003 does not Granger Cause 1.88172 0.1567 does not Granger Cause 128 3.70495 0.0274 Table A8: Summary Statistics-Africa Variable Name a Variable symbol Median Standard (in log form) Deviation Output 8.049 0.951 Rate of growth of labor force -1.462 0.956 Inflation rate 2.114 1.183 Trade openness 4.063 0.548 Physical capital b 2.464 1.840 Female labor with no education b 1.874 1.771 Male labor with no education b 2.072 1.736 Female effective labor b 4.029 1.535 Male effective labor b 4.895 1.234 Female labor force b 3.230 1.544 Male labor force b 3.644 1.510 Female average years of schooling b 0.605 1.033 Male average years of schooling b 1.181 0.660 Ratio of females with no education in female labor force -1.333 1.007 Ratio of males with no education in male labor force -1.528 1.021 a all variables except rate of growth of labor force, inflation rate and trade openness are per unit of labor b Stock variables 29