Female Education Externality and Inclusive Growth

Abstract

Education is generally believed to be beneficial in fostering, independent of gender, higher labor productivity. Female education may, however, cause other socio-economic gains which are not captured by higher wage or better compensation package for the educated female worker in the labor market (positive externality). This paper investigates the casual effect of enhancing female education and reducing gender education inequality on various measures of sustainable development. After addressing the endogeneity issue associated with gender education inequality employing a novel instrumental variable (IV), we find mitigating gender education inequality to be associated with lower infant mortality and poverty rates and improvements in health and environmental conditions. Our IV estimation result reports that a one-standard-deviation increase in the female-to-male ratio of average years of schooling is associated with a lower poverty rate by about 0.98 percentage points. The results indicate that expanding women’s educational opportunities is an effective way to promote inclusive growth.

1. Introduction

The standard theory of human capital and economic growth posits that a substantial portion of human capital is accumulated through schooling and utilized in producing goods and services. The price of human capital is based on its contribution to the production process, which in turn determines ex-ante education investment. The socially efficient level of education investment presumes the private and social return to education to be the same. Motivated by this argument, the present paper sheds lights on the female education externality by shifting the focus on social return to female human capital from the traditional economic growth measure of gross domestic product (GDP) to alternative inclusive growth measures. Specifically, it shows that promoting female education and reducing gender education inequality has a significant impact on sustainable and inclusive growth measures not considered in private returns to education, but not on the traditional measure of economic growth expected to be highly correlated with private returns.

Despite an extensive literature on the gender education gap and discriminatory private returns to education, no previous studies, to the best of our knowledge, consider gender-specific social returns to education, except for the work done by Cooray et al. [1], who reported differential effects of gender-specific human capital on traditional economic growth, but do not consider non-traditional measures of economic growth. It follows from the premise that human capital should be genderless such that it should be treated and exploited in the same manner regardless of gender. In this case, the notion of gender-specific social returns to education no longer makes sense. It becomes manifestly clear, however, that when the scope of analysis is expanded to consider gender differences in nurturing, feeding, and educating their families (including themselves), the marginal benefit of education differs between mothers and fathers. We extend this idea to examine social returns to female rather than male human capital because a larger gap between social and private returns is more likely to be observed in the case of the former.

The new notion of inclusive growth, unlike the traditional growth measure, in embracing equity of health and nutrition, inequality and poverty, social protection, environmental quality, and food security, as well as traditional GDP growth, facilitates investigation of social returns to education not captured by private returns. We attempt to identify the casual effect of female education on inclusive economic growth by combining the health index (HI), human development index (HDI), and environmental performance index (EPI) and infant mortality rate (IMR) and poverty rate (PR) with the educational attainment data provided by Barro and Lee [2]. The collected cross-country data enable us to run both ordinary least squares (OLS) and instrumental variables (IVs) regressions and exploit the panel aspect of our sample to analyze the relationship among growth rates from 2000 to 2010.

Estimation results consistently indicate a significantly positive impact of enhancing female education and mitigating gender education inequality on inclusive growth. Although we did not find (statistically) significant evidence that GDP growth is improved, we did find enhancing female education and reducing gender education inequality to lower IMR and PR, and to improve health and environmental conditions. Furthermore, we found that the gender education inequality mainly occurs at the primary education level in both developing and developed countries. Our results suggest that expanding women’s educational opportunities especially at the primary education level is an effective way to promote inclusive growth for developed as well as developing countries. More specifically, the policy tools designed to encourage female education, such as providing female scholarship, female dormitory, and girls’ schools, as well as implementing (genderless) compulsory primary schooling, are expected to lead the economy onto a sustainable growth track.

Kim, et al. [3] analyzed the social cost of the gender education gap in terms of the human development, environmental performance, and other indexes and measures. To measure the gender education gap, Kim, et al. [3] adopted the gap between the secondary education completion rate by females and males. The present study extends their work by incorporating analysis of the poverty rate and performing IV and fixed effects regressions to support casual interpretation of our results. Specifically, we adopted as a proxy variable for gender education inequality the ratio of average years of schooling by females to those by males, and as the instrument variable, the ratio of primary education completion rates by females and those by males. We showed the IV to crucially affect the proxy for gender education inequality, but not the inclusive growth measures as well as the traditional growth measure. It enables us to focus on the causal relationship between female education and those various growth measures. In the panel approach, we showed the growth rate of the proxy variable to be negatively associated with the growth rates for the inclusive growth measures.

The paper proceeds as follows. Section 2 reviews the previous literature with an emphasis on female education. Section 3 and Section 4 present the descriptive and regression analyses, respectively. Section 5 provides the results from robustness exercises. Section 6 discusses limitations of the present study and offers suggestions for future research. We conclude in Section 7.

2. Literature Review

The traditional viewpoint on the contribution of education to growth is well documented [4,5,6,7,8,9,10]. On top of these studies, Schultz [11] described how educational investment may lead to economic growth and stresses the importance of policies targeted toward investment in schooling in low-income countries. Collins et al. [12] examined the East Asian growth experience and confirm that educational investment plays a key role in accounting for the rapid economic growth between 1960 and 1994 in major East Asian countries. Mercan and Sezer [13] investigated the impact of education expenses on economic growth employing Turkish data and conclude that there exists a positive correlation between education expenditure and economic growth between 1970 and 2012 in Turkey. Lin [14] studied the case of Taiwan and shows that education had a significantly positive impact on economic growth over the period of 1965–2000. Using panel data from 2000 to 2016 on Guangdong Province in China, Liao et al. [15] demonstrated that local financial investment in education results in a statistically significant positive impact on sustainable economic growth. Diebolt and Hippe [16] focused on the long-run consequence of human capital on economic development and demonstrate how the past regional human capital contributes to rationalizing the current disparities in economic development in Europe. Overall, the existing literature shows that educational investment is the key determinant of economic growth across time and countries.

Educational opportunity for women has also received a great deal of political attention in many developing countries and most existing studies report positive effects of female education on economic progress, as expected [17,18,19,20,21,22]. A recent study by Cabeza-García, et al. [23] examines the link between economic growth and gender factors, where the latter is analyzed through four dimensions of education, fertility, employment, and democracy. In particular, the results from the study reaffirm that greater access to female education has significantly positive impacts on economic growth. In contrast to the extensive literature on the role of female education in economic development, there have been relatively few scholarly attempts to study the effects of the gender education gap on economic growth. Dollar and Gatti [24], using data from more than 100 countries over three decades up to the 1990s, showed gender education inequality to negatively affect economic growth, and Chen [25] identifies three channels through which this occurs: the selection-distortion factor, the environment effect, and the demographic transition effect. Baliamoune–Lutz and McGillivray [26] demonstrated, in the case of Africa and the Middle East, that the gap from female to male primary and secondary school enrollment rates has had a negative effect on income per capita. Klasen [27] found that, using the non-Barro coefficient, reducing gender inequality in education boosts economic development. Lagerlof [28] investigated the relationship between gender inequality in education and long-run economic growth and argues that gender gap in educational investment would lead to higher fertility, thereby causing lower economic growth. Benavot [29] also made a sharp contrast between male and female educational investment. Using cross-national data from a sample of 96 countries, he finds a greater impact of increased educational opportunities among school-age girls on long-run economic growth compared with that among school-age boys. Knowles, et al. [30] constructed a neoclassical growth model incorporating female and male education as separate components. By estimating the model using a cross-section of countries, they draw long-run implications of educational investment on economic development by gender. In particular, they find that female education is positively associated with aggregate labor productivity, while the relationship between male education and labor productivity is less clear.

Although previous studies generally suggest a negative correlation between gender education gap and economic progress, there exist a small number of studies that report beneficial impacts of gender inequality in education on economic growth. For example, Barro and Lee [31] employed the same data we used in this study and present empirical evidence that gender education gap contributes to economic growth. Barro and Lee [32] also found that there exists a statistically significant negative correlation between female secondary schooling and economic growth. The negative association between female education and economic growth seems to be partially explained by the negative impact female education has on fertility, which is expected to lower the stock of human capital in the next generation [33]. Despite the inconclusive finding regarding the relationship between female education and economic growth, one common feature of the previous studies is that their focus has been limited to traditional measures of economic growth such as income and employment. In addition, the results of most existing studies on gender education gap and growth are based on data from developing countries or prior to 2000 in which the level of female education is presumed to be substantially low. Therefore, empirical studies that employ more recent data on developed countries may produce divergent results.

Another strand of research related to our study, focused on external effects of improving female education, identifies substantial social benefits that exceed private returns. Using data from developing countries, Hill and King [34] emphasized that whereas the expense of female education expense is personally incurred, benefits from investment in female education accrue to the public. They further report that a high gender gap in education has a detrimental effect on a number of indexes including life expectancy and mortality rate. Subbarao and Raney [35] found improvement in female secondary education to be negatively associated with lower fertility and infant mortality and positively associated with children’s educational attainment. Schultz [36] claimed that governments should invest in women’s education because mother’s education is critical to the children’ health and schooling. Whaley [37], studying the relationship between gender education inequality and crime rates, found the gender education gap to be positively related to changes in rape statistics, and Karoui and Feki [38] described a process whereby female education lowers the fertility rate and improves educational attainment for the next generation. Kim, et al. [3], investigating the link between gender education inequality and inclusive growth, employing an approach and data similar to those used in this paper, found that a high gender gap, although not significantly related to GDP, negatively influences sustainable economic development. The present study contributes to the existing literature by empirically examining the causality between female education and various measures of inclusive growth.

3. Data and Descriptive Analysis

The primary dataset utilized in the present study, the Education Attainment for Population Aged 25 and Over from Barro-Lee Educational Attainment Data, provides educational attainment estimates for 146 countries from 1950 to 2010 at five-year intervals. It includes the distribution of the highest educational attainment at seven levels of schooling and average years of schooling at all levels, disaggregated by gender. The seven levels of schooling include no formal education, primary school total, primary school completed, secondary school total, secondary school completed, tertiary school total, and tertiary school completed. Within this dataset, years of schooling and primary education completion rates for 144 countries from 2000 to 2010 were the focus of the analysis. The proportion of primary school education, constructed as the sum of no schooling and incomplete primary school (primary school total–primary school completed), subtracted from 100 and disaggregated by sex, was expressed as primary education completion rates in the ratio of females to males.

We investigated the impact of the gender education inequality on economic growth by considering a series of the dataset as a dependent variable. We used as the traditional measure GDP per capita in current US dollars calculated by World Bank national accounts data and OECD National Accounts data files, and as alternative measures the HI and HDI created by the United Nations (UN) Development Programme, EPI 2018 scored by Yale University and Columbia University in collaboration with the World Economic Forum, IMR per 1000 lives estimated by the UN Inter-agency Group for Child Mortality Estimation, and a poverty headcount ratio at $1.9 a day calculated by the World Bank. We considered as well gross capital formation and gross fixed capital formation in constant 2010 US dollars obtained from the World Bank and OECD National Accounts data files. Because cultural influence based on religious belief could affect economic performance, we also included religion variables in the regression model. We used data from The Global Religions Landscape (2012) of the Pew Research Center, in which individuals are classified into the eight major religious groups of Christian, Muslim, religiously unaffiliated, Hindu, Buddhist, folk religion, other religion, and Jewish. Each religion variable reports the share of the religious group in the national population in each country based on self-identification.

Table 1. Descriptive statistics.

Variables	(1) Obs.	(2) Mean	(3) Std. Dev.	(4) Minimum	(5) Maximum
Years of schooling, male (2000)	144	7.56	2.85	1.38	13.17
Years of schooling, male (2010)	144	8.45	2.87	1.45	13.89
Years of schooling, female (2000)	144	6.59	3.28	0.42	12.89
Years of schooling, female (2010)	144	7.74	3.4	0.79	13.44
Years of schooling, female/male (2000)	144	0.82	0.22	0.21	1.39
Years of schooling, female/male (2010)	144	0.88	0.2	0.23	1.34
Primary education completion, female/male (2000)	144	0.83	0.24	0.17	1.81
Primary education completion, female/male (2010)	144	0.9	0.22	0.16	2.19
GDP per capita	143	14,723.2	19,650.4	231.19	104,965
Health index (HI)	143	0.78	0.14	0.38	0.97
Human development index (HDI)	141	0.7	0.16	0.32	0.94
Environmental performance index (EPI)	139	57.63	13.53	27.43	87.42
Infant mortality rate (IMR)	142	25.66	24.29	2	106.7
Poverty rate (PR)	99	9.62	17.47	0	71.4
Gross capital formation per capita	137	3362.04	4476.09	63.69	22,306.3
Gross fixed capital formation per capita	135	3053.62	3985.62	63.69	18,482.4
Christian	144	54.41	37.9	0	99.5
Muslim	144	24.71	37.2	0	99.9
Hindu	144	2.68	11.11	0	80.7
Buddhist	144	4.74	16.31	0	96.9
Folk religion	144	2.57	7.74	0	58.9
Jewish	144	0.6	6.3	0	75.6
Unaffiliated	144	9.86	14.07	0	76.4
Other religion	144	0.35	0.97	0	9.7

Note: Obs. and Std. Dev stand for number of observation and standard deviation, respectively.

presents descriptive statistics for the sample. Average schooling years for males is eight, with a discernable standard deviation and gap between minimum and maximum years in selected countries. The ratio of years of schooling between female and male is less than one suggests that, worldwide, females tend to be less educated than males. A narrower gender gap was observed in primary education completion, with a slightly higher mean. In terms of the measure of economic development, average GDP per capita is approximately fifteen thousand US dollars with a huge gap among selected countries. HI and HDI are similar, with the health dimension of HDI being assessed by life expectancy at birth, which was used to calculate HI. The average score of EPI is roughly 57, and the average number of infants who died within one year of birth per 1000 live births is 25.66. Nearly 10 percent of the sample population lives on less than $1.90 a day. In terms of religion, Christian makes up more than half the sample population, followed by Muslim.

4. Empirical Methodology and Results

4.1. Ordinary Least Squares Estimation Approach

We began our analysis by specifying the empirical model for investigating the relation between women’s education and various measures of economic and inclusive growth. Specifically, we considered the following regression equation:

$$Y_i={\alpha }_0+{\alpha }_1\left(Y{S}_{i,MALE}\right)+{\alpha }_2\left(\frac{Y{S}_{i,FEMALE}}{Y{S}_{i,MALE}}\right)+{\displaystyle {\sum }_j{\beta }_j{X}_{ij}}+{\epsilon }_i$$

(1)

where $Y_i$ is the level of economic growth of country $i$, measured by the HI, HDI, EPI, IMR, and PR, as well as the traditional growth measure of GDP per capita, $Y{S}_{i,MALE}$ denotes the average years of schooling received by the male population, years of schooling proxies for the average level of human capital possessed by the workforce, and $\frac{Y{S}_{i,FEMALE}}{Y{S}_{i,MALE}}$, the key explanatory variable, is the ratio of the average years of schooling received by the female and the male population, in country $i$. Although ideally we would control $Y{S}_{i,FEMALE}$ directly in the regression equation together with $Y{S}_{i,MALE}$ to estimate the marginal effects of female education, this approach, being likely to be vulnerable to multicollinearity due to high correlation between the two variables, might compromise the statistical precision of the estimates. Given that $Y{S}_{i,FEMALE}$ appears in the numerator and educational attainment is generally less for the average woman than for a male counterpart, the ratio of $Y{S}_{i,FEMALE}$ and $Y{S}_{i,MALE}$ is negatively associated with the level of gender education inequality. The estimated coefficient of ${\alpha }_2$ would thus indicate the impact, if any, of gender inequality in educational attainment on economic growth. A limitation of the estimation strategy shown in Equation (1) is that the impact of the ratio of average years of schooling by gender (or ${\widehat{\alpha }}_2$) would include the impact of both mitigating gender education inequality and improving women’s educational attainment, holding average years of schooling received by the male population fixed. To distinguish one from the other would require additional data on labor market participation rates by gender and educational attainment. $X_{ij}$ is a vector of explanatory variables that includes capital per person in addition to the share of each major religion in the population in country $i$. Owing to the potential for religious belief to influence individual characteristics related to economic progress [39], the proportions of religious population are included in the regression equation. In the event one of the five alternative growth measures is used, GDP per capital is also included as a control variable. Country-specific random shocks to the level of economic development are captured by ${\epsilon }_i$, which is assumed to be normally distributed with zero mean and finite variance.

Table 2. Impact of gender education inequality on growth measures (ordinary least squares (OLS) regressions).

Variables	(1) GDP Per Capita	(2) HI	(3) HDI	(4) EPI	(5) IMR	(6) PR
$Y{S}_{i,MALE}$	360.4 *	0.0249 ***	0.0313 ***	2.015 ***	−4.765 ***	−4.506 ***
	(211.6)	(0.00379)	(0.00249)	(0.288)	(0.582)	(0.582)
$\frac{Y{S}_{i,FEMALE}}{Y{S}_{i,MALE}}$	−2283	0.151 ***	0.207 ***	9.474 ***	−40.22 ***	−24.24 **
	(2587)	(0.0462)	(0.0309)	(3.570)	(7.095)	(9.981)
Capital per capita	4.241 ***
	(0.114)
Christian	29.69	−0.000428	−0.000653	−0.00449	0.0719	−0.0287
	(35.40)	(0.000646)	(0.000423)	(0.0506)	(0.102)	(0.0936)
Muslim	2.549	0.000263	−0.000182	−0.0104	−0.0251	−0.178 *
	(34.55)	(0.000634)	(0.000415)	(0.0494)	(0.0998)	(0.0937)
Hindu	−13.26	0.000557	0.000108	−0.162 **	−0.113	−0.177
	(49.59)	(0.000883)	(0.000578)	(0.0681)	(0.137)	(0.139)
Buddhist	−2.721	0.000718	−0.000214	−0.0265	−0.0380	−0.166
	(44.68)	(0.000815)	(0.000534)	(0.0630)	(0.127)	(0.137)
Folk religion	157.0 **	0.000395	−0.00108	−0.138	0.0877	−0.323
	(72.19)	(0.00161)	(0.00124)	(0.123)	(0.248)	(0.290)
Jewish	89.77	−0.000119	−0.000691	0.0213	0.131	0.0876
	(71.47)	(0.00130)	(0.000853)	(0.0997)	(0.202)	(0.174)
GDP per capita		2.21 × 10−6 ***	2.53 × 10−6 ***	0.000291 ***	−0.000178 **	−1.84 × 10−5
		(4.82 × 10−7)	(3.15 × 10−7)	(3.66 × 10−5)	(7.40 × 10−5)	(7.25 × 10−5)
Constant	−2402	0.411 ***	0.256 ***	29.26 ***	100.8 ***	79.71 ***
	(4310)	(0.0790)	(0.0524)	(6.174)	(12.34)	(13.14)
Observations	137	142	140	139	141	99
R-squared	0.944	0.639	0.873	0.765	0.703	0.595

Notes: Standard errors are in parentheses; *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

reports results employing the conventional OLS estimation method for Equation (1). Column (1) shows both average years of schooling of the male population and capital per capita to have a positive, statistically significant impact on per capita GDP, consistent with the idea that investment in education and capital accumulation improve labor productivity. That the ratio of $Y{S}_{i,FEMALE}$ and $Y{S}_{i,MALE}$ shows no statistically significant effect on per capita GDP suggests that elevating the level of women’s education relative to men’s does not significantly contribute to economic growth in a quantitative sense. Columns (2)–(6), however, show improved women’s education relative to men’s to have a positive influence on various alternative measures of economics growth, an increase in the ratio of $Y{S}_{i,FEMALE}$ and $Y{S}_{i,MALE}$ being associated with higher values of health, human development, EPIs and lower IMR and PR. In particular, the fact that narrowing the gender education gap is related to fewer people with income below the poverty line provides the first evidence that investment in women’s education contributes positively to inclusive growth. Overall, our results clearly show improvements in women’s education to be associated with external benefits not recouped in the labor markets, indicating a discrepancy between social and private returns to female education. Table 2 further suggests that, controlling for levels of human as well as physical capital, religious belief does not seem to affect the process of economic development.

4.2. Instrumental Variables Estimation Approach

Equation (1), although a straightforward approach to examining the effects of female education on economic growth, suffers from a couple of econometric issues. One important consideration is that a country’s level of gender education inequality is likely to be correlated with the degree of economic development through confounding (or omitted) variables, thereby invalidating the classical OLS assumption of conditional independence. It is conceivable, for instance, that in a country experiencing economic progress and maturing with respect to gender equality, access to education is expected to be the same for girls as for boys and the enrollment gap between genders to naturally disappear. In this case, the presence of an unobserved variable that might influence both gender education inequality and economic growth would result in the OLS estimates in Table 1 overstating the impact of the gender education gap on economic growth.

We addressed the endogeneity issue by employing an IV. A valid instrument in this context must be correlated with the degree of gender education inequality and independent of the level of the focal country’s economic development, conditional on the value of the gender education inequality. For this purpose, we exploited cross-country variation in the elementary school completion rate across gender. The key insight is that the gender education inequality is in part explained by the principal–agent problem that occurs when parents decide how much to invest in their children’s education [3]. To the extent that parents strategically allocate greater resources to sons’ than daughters’ education in anticipation of recouping their investment in the form of male workers’ higher wages, sons are expected to be more likely than daughters to finish elementary school. As boys and girls grow older, they are more likely to self-determine their education decision. Thus, the difference in elementary school completion rates between sons and daughters is expected to be strongly correlated with gender inequality in educational attainment, and unlikely to have a direct impact on various measures of economic growth, providing plausibly exogenous variation by which to identify the causality between gender education inequality and economic growth.

To check the validity of the exclusion restriction regarding the use of IVs, we presented in

Table 3. Conditional completion probability by education level.

	Low-Income Countries			Middle-Income Countries			High-Income Countries
	Male	Female	F/M	Male	Female	F/M	Male	Female	F/M
Primary	0.562	0.442	0.786	0.796	0.763	0.959	0.914	0.900	0.985
Secondary	0.410	0.358	0.873	0.529	0.527	0.996	0.667	0.654	0.981
Tertiary	0.195	0.192	0.985	0.189	0.189	1.000	0.258	0.267	1.035

Source: Barro and Lee Educational Attainment Data, 2010.

the completion probabilities (conditional on enrolling) by education level and gender for each income group of countries. The country classification by income level is based on the World Bank. From Table 3, it is evident that the gender education inequality is most prominent at the primary education level across all income groups. As education level advances, the completion rate of the female population was observed to catch up with that of the male population, and the gender education gap, measured by (an inverse of) the ratio of completion rates of female and male, quickly narrows. The gender difference in school completion probabilities was further found to be most conspicuous at the primary education level in low-income countries. Note that the tertiary education completion rate by female is greater than the completion rate by males in high-income countries. Table 3 suggests that gender inequality in educational attainment is related to the principal–agent problem that occurs when parents (agents) determine the allocation of household resources to the education of their children (principals).

Table 4. The impact of gender education inequality on growth measures (instrumental variable (IV) regressions).

Variables	(1) GDP Per Capita	(2) HI	(3) HDI	(4) EPI	(5) IMR	(6) PR
$Y{S}_{i,MALE}$	−604.6	0.111 **	0.169 ***	8.929 **	−30.37 ***	−17.01 *
	(2869)	(0.0514)	(0.0346)	(3.980)	(7.935)	(10.06)
$\frac{Y{S}_{i,FEMALE}}{Y{S}_{i,MALE}}$	315.5	0.0260 ***	0.0322 ***	2.029 ***	−5.034 ***	−4.654 ***
	(207.5)	(0.00373)	(0.00245)	(0.282)	(0.575)	(0.558)
Capital per capita	4.230 ***
	(0.111)
Christian	28.20	−0.000409	−0.000633	−0.00432	0.0677	−0.0290
	(34.16)	(0.000625)	(0.000410)	(0.0488)	(0.0993)	(0.0890)
Muslim	3.226	0.000239	−0.000208	−0.0108	−0.0188	−0.169 *
	(33.33)	(0.000613)	(0.000403)	(0.0476)	(0.0969)	(0.0892)
Hindu	−11.15	0.000500	4.94 × 10−5	−0.163 **	−0.0980	−0.146
	(47.85)	(0.000854)	(0.000561)	(0.0657)	(0.133)	(0.133)
Buddhist	−5.205	0.000764	−0.000167	−0.0259	−0.0489	−0.171
	(43.14)	(0.000789)	(0.000518)	(0.0608)	(0.124)	(0.130)
Folk religion	158.7 **	0.000248	−0.00133	−0.140	0.124	−0.302
	(69.64)	(0.00156)	(0.00121)	(0.119)	(0.241)	(0.276)
Jewish	89.13	−0.000122	−0.000691	0.0212	0.132	0.0869
	(68.93)	(0.00126)	(0.000827)	(0.0961)	(0.196)	(0.166)
GDP per capita		2.24 × 10−6 ***	2.56 × 10−6 ***	0.000292 ***	−0.000187 ***	−1.34 × 10−5
		(4.66 × 10−7)	(3.06 × 10−7)	(3.53 × 10−5)	(7.19 × 10−5)	(6.90 × 10−5)
Constant	−3394	0.436 ***	0.281 ***	29.62 ***	94.53 ***	74.15 ***
	(4240)	(0.0781)	(0.0520)	(6.094)	(12.24)	(12.76)
Observations	137	142	140	139	141	99
R-squared	0.944	0.637	0.871	0.765	0.698	0.592
AP $X^2$-statistics	103.62	107.01	104.73	103.79	106.25	88.10
AP F-statistics	394.24	403.64	385.99	380.36	400.65	719.41

Notes: Standard errors are in parentheses; *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively; Angrist–Pischke $X^2$ and F-statistics report the first-stage test statistics of under-identification and weak-identification, respectively, of the endogenous regressor.

presents the two-stage least squares estimation results using the ratio of primary education completion rates of the female and male populations as the IV. The Angrist–Pischke statistics reported at the bottom of Table 4 confirm that the IV strongly passes the first-stage tests for under-identification and weak identification, thereby ensuring the validity of the exclusion restriction. Overall, the results show use of the IVs to reduce the magnitudes of the estimated coefficients compared to the OLS results in Table 2. The main results were observed to hold qualitatively, enhancing women’s relative to men’s education being shown to improve all the alternative measures of economic growth, while having no discernable impact on the traditional growth measure. In particular, a one-standard-deviation increase in $\frac{Y{S}_{i,FEMALE}}{Y{S}_{i,MALE}}$ was predicted to lower PR by about 0.98 percentage points (= −4.654 × 0.211), suggesting a strong positive impact of women’s education on inclusive growth. The results in Table 2 and Table 4, in attesting to the causal impact of gender education inequality on inclusive growth measures, demonstrate that improving female education confers socio-economic gains that exceed private returns to female education.

5. Robustness

An additional concern regarding the validity of the main results reported in Table 2 and Table 4 is that identification relies solely on cross-country variation due to the structure of the collected dataset. This can be problematic in the event that countries with more balanced educational opportunities across gender tend to be advanced economies with better health and nutrition and environmental conditions and lower PRs and IMRs, in which case the estimation exercise based on a single cross-section of data may produce spurious correlations between gender education inequality and growth measures. To check whether the main results are robust to the presence of unobserved time-invariant country effects, we considered the following regression equation:

$$Y_{it}={\alpha }_0+{\alpha }_1\left(Y{S}_{it,MALE}\right)+{\alpha }_2\left(\frac{Y{S}_{it,FEMALE}}{Y{S}_{it,MALE}}\right)+{\displaystyle {\sum }_j{\beta }_j{X}_{ijt}}+{\mu }_i+{\epsilon }_{it}$$

(2)

where ${\mu }_i$ represents country fixed effects and $t$ indexes the time period. Identification of the coefficients in (2) requiring within-country variation over time, we collected, as available, data on the variables from 2010, and estimated the parameters of the model by first differencing. We were unable to gather data from the EPI and PR for 2010. Further, there being insufficient variation in the population shares of major religions over the 2000s, religion variables were also excluded from the sensitivity analysis.

To test whether our results were sensitive to alternative definitions of physical capital when the dependent variable was GDP per capita, we reported the regression results using fixed capital as well.

Table 5. Sensitivity analysis (fixed effects, OLS regressions).

Variables	(1) ∆(GDP Per Capita)	(2) ∆(GDP Per Capita)	(3) ∆HI	(4) ∆HDI	(5) ∆IMR
∆$\left(Y{S}_{i,MALE}\right)$	−0.113	−0.106	0.000204	0.00530 *	−0.0135
	(0.150)	(0.146)	(0.00455)	(0.00282)	(0.0155)
∆$\left(\frac{Y{S}_{i,FEMALE}}{Y{S}_{i,MALE}}\right)$	0.116	0.0801	0.117 ***	0.114 ***	−0.319 **
	(1.372)	(1.331)	(0.0402)	(0.0247)	(0.135)
∆(Capital per capita)	0.00021 **
	(9.13 × 10−5)
∆(Fixed capital per capita)		0.00026 ***
		(9.84 × 10−5)
∆(GDP per capita)			0.00104	0.00924 ***	−0.0469 ***
			(0.00272)	(0.00167)	(0.00916)
Constant	1.511 ***	1.463 ***	0.0391 ***	0.0309 ***	−0.198 ***
	(0.192)	(0.187)	(0.00696)	(0.00429)	(0.0234)
Observations	113	115	140	138	139
R-squared	0.055	0.065	0.060	0.288	0.193

Notes: Columns (1) and (2) differ in the definition of physical capital. Column (1) employs total capital while column (2) uses fixed capital only. Standard errors are in parentheses; *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

reports the OLS regression results from the sensitivity analysis described above. Columns (1) and (2) in Table 5 show the main results to be robust to the use of alternative definitions of physical capital, in that gender education inequality is shown to be unrelated with per capita GDP, regardless of the definition of physical capital. Regarding the estimates of the change in $\frac{Y{S}_{i,FEMALE}}{Y{S}_{i,MALE}}$ in Columns (3)–(5), we found that all the coefficients were statistically significant with the addition of country fixed effects at the 1% level, indicating that the main results hold in the presence of unobserved country effects.

Table 6. Sensitivity analysis (fixed effects, IV regressions).

Variables	(1) ∆(GDP Per Capita)	(2) ∆(GDP Per Capita)	(3) ∆HI	(4) ∆HDI	(5) ∆IMR
∆$\left(Y{S}_{i,MALE}\right)$	0.178	0.0667	0.158 ***	0.153 ***	−0.227
	(1.454)	(1.410)	(0.0446)	(0.0274)	(0.149)
∆$\left(\frac{Y{S}_{i,FEMALE}}{Y{S}_{i,MALE}}\right)$	−0.113	−0.106	0.000420	0.00549 *	−0.013
	(0.147)	(0.143)	(0.00450)	(0.00281)	(0.0153)
∆(Capital per capita)	0.000214 **
	(8.97 × 10−5)
∆(Fixed capital per capita)		0.000264 ***
		(9.67 × 10−5)
∆(GDP Per Capita)			0.00103	0.00922 ***	−0.0469 ***
			(0.00269)	(0.00166)	(0.00904)
Constant	1.507 ***	1.463 ***	0.0365 ***	0.0284 ***	−0.204 ***
	(0.192)	(0.186)	(0.00701)	(0.00434)	(0.0235)
Observations	113	115	140	138	139
R-squared	0.055	0.065	0.053	0.275	0.190
AP $X^2$-statistics	96.98	98.90	111.54	110.35	110.82
AP F-statistics	660.04	682.12	533.01	535.00	530.99

Notes: Columns (1) and (2) differ in the definition of physical capital. Column (1) employs total capital while column (2) uses fixed capital only. Standard errors are in parentheses; *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

presents the IV regression results, and columns (1) and (2) show the results are still invariant to the use of alternative definitions of physical capital. Turning to the estimates of the change in the female-to-male ratio of average years of schooling $\frac{Y{S}_{i,FEMALE}}{Y{S}_{i,MALE}}$, the results in columns (3)–(5) demonstrate that allowing for country fixed effects generally results in the magnitudes of estimates becoming much smaller, and coefficients, in some cases, losing statistical significance. All coefficients retain the expected sign, however, indicating that the results remain qualitatively similar independent of incorporating unobserved country effects. The results of the sensitivity analysis, in confirming the positive impact of women’s education on promoting inclusive growth and yielding benefits society wide, emphasize the importance of understanding gender-specific impacts of educational investment.

6. Limitations of the Present Study and Suggestions for Future Research

Unlike the traditional viewpoint that human capital should be genderless, the present research highlights the gender-specific returns to education gauged by the inclusive growth measures. One shortcoming of this research is that without labor market participation rates by gender and education level for each country, we cannot tell whether and to what extent our conclusion regarding the impact of gender education gap derives from the gender-specific nature, or gender-specific utilization rate, of human capital. In particular, the fact that a more educated woman is more likely than a less educated counterpart to be employed may amplify the (pure) effect of the gender education gap on both traditional and inclusive growth measures. Our use of cross-country data precludes decomposing the effect of the gender-specific nature of human capital from the other effect, because it does not provide the labor market participation rate by gender and by education level. We look to future studies to remedy this deficiency using micro-level panel data.

7. Conclusions

The notion of “inclusive growth” takes into account equity of health and nutrition, inequality and poverty, social protection, environmental quality, and food security, as well as traditional GDP growth. Employing the HI, HDI, EPI and PR and IMR as measures of inclusive growth, we showed that reducing the gender education inequality has a significant impact on these new inclusive growth measures, although not on the traditional GDP growth measure. This can be (at least in part) interpreted as a consequence of the gender-specific nature of human capital accumulated through education. Overall, the empirical results found in this study are in line with the previous literature that documents the evidence of positive externalities associated with female education. By employing IVs estimation strategy and allowing for unobserved country-specific effects, this study enables us to provide a causal interpretation of the estimated coefficients related with the impacts of investment in female education on various measures of inclusive growth.

We also found that the gender education inequality takes place at the primary education level, at which educational investment is determined not by each student but by her or his parents. Given that women leave their biological families and/or change their family names, female students may get less human capital investment by their parents compared to other male students. Our results suggest that expanding women’s educational opportunities especially at the primary education level is an effective way to promote inclusive growth for developed as well as developing countries. Thus, the policy tools designed to encourage female education, such as providing female scholarship, female dormitory, and girls’ schools, as well as implementing (genderless) compulsory primary schooling, may effectively promote not only female education but also sustainable economic development.