Suffrage, Democracy and Gender Equality in Education

Transcription

1 Oxford Development Studies ISSN: (Print) (Online) Journal homepage: Suffrage, Democracy and Gender Equality in Education Arusha Cooray To cite this article: Arusha Cooray (2012) Suffrage, Democracy and Gender Equality in Education, Oxford Development Studies, 40:1, 21-47, DOI: / To link to this article: Published online: 15 Feb Submit your article to this journal Article views: 1010 View related articles Full Terms & Conditions of access and use can be found at Download by: [ ] Date: 30 January 2016, At: 07:10

2 Oxford Development Studies, Vol. 40, No. 1, 21 47, March 2012 Suffrage, Democracy and Gender Equality in Education ARUSHA COORAY ABSTRACT Examining the influence of women s suffrage and democracy on gender equality in education in a sample of 80 countries, covering Asia, Africa, the Middle East, South America and Eastern Europe, this study shows that countries with a longer duration of suffrage tend on average to perform better in terms of gender equality in education. The empirical association between democracy and gender equality in education disappears when religion is controlled for. In Asia and Africa, other factors, including income, employment in agriculture and colonialism, also help explain the under-representation of girls in education. JEL Classification: O11, O15, O43, O57 1. Introduction One of the United Nations Millennium Development Goals (MDGs) was to eliminate gender disparity in primary and secondary education, ideally by 2005, and in all levels of education by 2015 (United Nations, 2008). However, gender disparity in education at the primary, secondary and tertiary levels persists in certain regions. Several studies have shown that gender inequality in education can have a negative impact on macroeconomic variables such as economic growth (Shultz, 1994; Dollar & Gatti, 1999; Knowles et al., 2002; Klasen, 2002), child mortality (Klasen & Wink, 2002) and fertility (Basu, 2002). The education of girls has been shown to be an important foundation for creating the next generation of human capital, because mothers play an important role in determining the education and health of their children (Schultz, 2002). Fafunwa (1974), examining the history of education in Nigeria, observed that the roots of gender discrimination can be found in the colonial administration, under which women were marginalized and their participation in social and economic activity restricted. Similarly Mann (1984), in a study of Lagos in Nigeria, noted that with the creation of an elite under colonial rule, the role of women was restricted primarily to social activity. This approach was inherited by postcolonial governments, which continued to prioritize male education. Arusha Cooray, School of Economics, University of Wollongong, Northfields Avenue, NSW 2522, Australia. arusha@uow.edu.au The author wishes to thank two anonymous referees, the Editor of the journal, Frances Stewart, Arye Hillman, Niklas Potrafke, the participants of the Far East and South Asia Meeting of the Econometric Society, Tokyo, 2009 for valuable comments, and Pascale Ratovondrahona (UNESCO) for supplying the author with some of the data series. ISSN print/issn online/12/ q 2012 Oxford Department of International Development

3 22 A. Cooray Several factors have given rise to an increased focus on women s education in the recent past, including the collapse of authoritarian regimes in Eastern Europe and Latin America, the commitment to reducing fertility rates in Asia and Africa, globalization, and the increased participation of women in economic activity. Gender equality in education can therefore only be understood in the context of these changes in the political, social and economic spheres. A growing number of studies are investigating the effects of education on democracy. For example, the hypothesis that education leads to increased democracy is supported in the work of Barro (1999), Glaeser et al. (2004) and Papaioannou & Siourounis (2005). However, much less attention has been paid to the effects of democracy on education. The influence of democracy on gender equality in education was examined by Brown (2000, 2004), Beer (2009) and Cooray & Potrafke (2011). Using the ratio of the average number of years women attended school to the average number of years men attended school in 1990 to measure gender equality, and the subindicators of POLITY III to measure democracy, Brown (2004) found that only the executive-recruitment components of democracy have a positive influence on gender equality in education. 1 That is, where the process of executive recruitment is competitive, regularized and open, politicians are in a better position to implement policies that favour women, which in turn will contribute towards a narrowing of the gender gap in education. Beer (2009) measures gender equality in education by the average years of educational attainment of women and men. Democracy is measured by the level and stock of the POLITY IV democracy indices, as well as the year in which women gained the right to vote. The results show that in countries with longer-term democracy and longer duration of women s suffrage, the ratio of female-to-male life expectancy is higher, fertility rates are lower and rates of labour force participation by women are higher. However, both the stock of democracy and the year woman gained suffrage have a negative influence on gender equality in education. Beer (2009) argues that this might suggest that long-term participation in the political process by women is necessary for long-term improvements in their conditions, rather than current opportunities. Beer, however, concludes that both democracy and suffrage are significantly related to gender equality as they move in the hypothesized direction for three of the five indicators examined. Cooray & Potrafke (2011) investigate the influence of democracy on gender equality in education using the ratio of primary, secondary and tertiary enrolment rates of girls to boys to measure gender equality in education. Democracy is measured using the Democracy and Dictatorship (DD) measure of political regimes in Cheibub et al. (2010), and a subcomponent of the POLITY IV index, XCONST (constraints on chief executive). The study finds that religion plays a more important role in influencing gender equality in education than political institutions. This study represents an initial effort to examine the effects of the duration of suffrage on gender in/equality in education. 2 Many countries in the sample are former colonies and gained suffrage only after independence. Ramirez et al. (1997) noted that the acquisition of suffrage permits the fostering of better political structures and social movements for women. These movements then act as a foundation for the promotion of equal opportunities for women. It can be argued that the longer the duration of suffrage the more likely it is that institutions have been created that are conducive to promoting the interests of women. Therefore, a variable representing the duration of suffrage attempts to capture changes in political institutions over time favourable to gender equality. Similarly, democratic institutions provide greater individual freedom and policies that do not discriminate by gender. In countries with

4 Suffrage, Democracy and Gender Equality in Education 23 legitimate governments, legal systems and well-established property rights, target-oriented economic policies yield the desired results as opposed to autocracies (e.g. De Haan & Sturm, 2003). Democratic governments can implement economic policies aimed at improving the status of women through educational development, increased employment opportunities and empowering women to gain positions of leadership. Democratic institutions are therefore conducive to gender equality, including in particular in education. Therefore, this study also investigates the influence of democracy on gender equality in education. Huntington (1991, p. 7) defines a government as democratic to the extent that its most powerful collective decision makers are selected through fair, honest and periodic elections in which candidates freely compete for votes and in which virtually all the adult population is eligible to vote. In short, democracy can be said to incorporate the components of competition, participation and civil liberties as well as voting. Suffrage can be said to be an element of democracy that falls under the participation component. 3 Paxton (2000) argues that none of the definitions of democracy put forward by scholars explicitly incorporates women s suffrage, and that incorporating female suffrage into measures of democracy changes the dates of democratic transitions, in some cases by 50 years. According to her,... to operationalize democracy as currently defined, we must incorporate the extension of female suffrage (Paxton, 2000, p. 106). Hence, incorporating female suffrage in the analysis is particularly relevant. In contrast to the Brown (2004), Beer (2009) and Cooray & Potrafke (2011) studies, this study also disaggregates the sample by region in order to account for sample heterogeneity (see Quah, 1996). Drawing on a sample of 80 countries covering Asia, Africa, the Middle East, South America and Eastern Europe over the period , the study finds a positive relationship between democracy and greater gender equality in education for some regions only. Moreover, when religion is controlled for the relationship between democracy and gender equality in education weakens. A preliminary investigation of the data shows that girls are under-represented at both the primary and secondary and the tertiary levels in Asia and Africa. To attempt to explain this, the study controls for a number of other factors, including income, employment in agriculture, culture and colonialism. These factors are found to influence gender inequality in education in Asia and Africa. In order to correct for any potential endogeneity bias in the model, instrumental variable (IV) estimation is also employed. The rest of this paper is structured as follows: Section 2 states the hypothesis, Section 3 explains the data, Section 4 discusses the empirical results and Section 5 concludes. 2. The Hypothesis In many countries, women did not have the right to vote until the 20th century. It can be argued that greater democracy improves the status of women by increasing educational development and employment opportunities, empowering women and enabling them to secure positions of leadership. Women in positions of responsibility and leadership may promote improved conditions for girls and women at grass-roots levels. Accordingly, with the objective of enhancing the understanding of gender inequality in education, this study investigates two hypotheses: (1) The duration of women s suffrage influences gender equality in education. (2) Democracy influences gender equality in education.

5 24 A. Cooray 2.1 Does the Duration of Female Suffrage Matter for Greater Gender Equality in Education? The acquisition of female suffrage fosters better political structures and social movements for women (see Ramirez et al., 1997). These movements then act as a basis for the promotion of equal opportunities for women. Examining the relationship between democracy and gender equality, Beer (2009) noted that long-term democracy together with female suffrage permit women to promote their interests through elections. She showed that countries with long-standing democracy and women s suffrage have higher female-to-male life expectancy ratios, lower fertility rates and higher labour force participation rates by women as a result of their ability to advance their interests through voting. Similarly, Brown (2004) observed that the duration of suffrage could be important in leading to pressure on politicians to extend educational opportunities for women. Figure 1 illustrates the duration of female suffrage until 2005 (the latest year in the present study) for the sample of countries under study. The duration of suffrage is longest for females in Eastern European and Central Asia. By 1930, only women in the high-income OECD nations, the USSR, Mongolia and Ecuador had acquired suffrage (Ramirez et al., 1997). Figure 2 plots the relationship between the duration of suffrage and the primary and secondary female-to-male enrolment ratio. The figure indicates a positive relationship between the duration of suffrage and a higher primary and secondary female/male enrolment ratio. Women in many African, Asian and Middle Eastern countries gained suffrage only after independence. Although suffrage following independence permitted women to participate in the political process, progress has still to be made in the area of education in these regions. 2.2 Does Democracy Lead to Greater Gender Equality in Education? Democracy promotes gender equality through the mobilization of women and electoral accountability (see Beer, 2009). According to Beer (2009), women are better able to organize to express their views and interests, obtain and disseminate information, and lobby to improve their status through education. Democracy also increases women s bargaining power within the household (Klasen & Wink, 2003), enabling mothers to invest more in the health and education of their children. Greater democracy leads to the design of policies aimed at improving the status of women through educational development and increased employment opportunities, empowering women and enabling them to secure positions of leadership. Figures 3 and 4 plot the relationship between the Polity IV and Freedom House Democracy indices and the female/male primary and secondary enrolment ratio for The figures show a positive relationship between democracy and gender equality in education. Despite the strong association between democracy and gender equality, it is worth noting that this relationship does not hold in many African countries, particularly Malawi, which shows relatively high gender inequality, notwithstanding the relatively high levels of democracy there. This raises the important question of why there is an under-representation of girls in education in Africa. This can be attributed to several factors that have been found to affect gender educational inequality in previous studies, including: the colonial heritage (Brown, 2000); culture (Dollar & Gatti, 1999; Inglehart & Baker, 2000); trade openness (Baliamoune-Lutz

6 Suffrage, Democracy and Gender Equality in Education 25 Ukraine Belarus Lithuania Latvia Kyrgyzstan Estonia Azerbaijan Armenia Turkmenistan Tajikistan Mongolia Kazakstan Ecuador South Africa Uruguay Thailand Maldives Brazil Sri Lanka Bolivia Dominican Cuba Indonesia Guatemala Trinidad and Slovenia Panama Croatia Vietnam Venezuela Philippines China Argentina Syria India Costa Rica Chile ElSalvador Nepal Jamaica Bhutan Colombia Peru Nicaragua Ghana Ethiopia Sudan Senegal Pakistan Niger Mauritania Lao Congo Comoros Chad Cameroon Cambodia Benin Malaysia Algeria Albania Tunisia Morocco Mauritius Nigeria Zimbabwe Sierra Leone Paraguay Burundi Uganda Libya Kenya Iran Malawi Botswana Lesotho Yemen Bangladesh Bahrain Jordan Figure 1. Duration of suffrage to Note: In South Africa white women acquired suffrage in 1930 but black women did not gain suffrage until Source: Dates of suffrage acquisition from Ramirez et al. (1997). Duration of suffrage calculated as the difference between 2005 and the year of suffrage.

7 26 A. Cooray Mauritania Panama Uganda Vietnam Nepal Burundi Comoros Pakistan Uruguay 0.6 Yemen Morocco Chad 0.4 Malawi Time Ratio Female/Male primary and secondary enrolment Fitted values Figure 2. Ratio girls/boys in primary and secondary education and duration of suffrage. Note: The regression represented by the fitted line reports a coefficient of (robust SE ¼ 0.001), N ¼ 78, R 2 ¼ 0.07 from a regression of ratio of girls/boys in primary and secondary education on the duration of suffrage. & McGillivray, 2007); high fertility (Klasen, 2002); and employment in agriculture. Hence, these variables are considered as additional control variables in the empirical estimation that follows Libya Belarus China Uganda Sa Syria Vietnam Bhutan udi Arabia Zimbabwe Sudan Nepal Lao Congo Cameroon Pakistan Morocco Yemen Chad CambodiaNigeria Armenia Comoros Burundi Ethiopia Niger Indonesia Mauritius 0.4 Malawi Polity IV Measure of Democracy Ratio Female/Male Primary and secondary education Fitted values Figure 3. Ratio girls/boys in primary and secondary education and democracy Note: The regression represented by the fitted line reports a coefficient of 0.72 (robust SE ¼ 0.18), N ¼ 80, R 2 ¼ 0.15 from a regression of ratio of girls/boys in primary and secondary education on the Freedom House Political Rights Index.

8 Suffrage, Democracy and Gender Equality in Education Libya Belarus China Saudi Syria Vietnam Zimbabwe Arabia Sudan Lao Bhutan Cambodia Cameroon Pakistan Chad Armenia Uganda Nepal Burundi Congo Ethiopia Morocco Yemen Comoros Nigeria Indonesia Niger Mauritius 0.4 Malawi Freedom House Political Rights Index 3. Data Ratio Female/Male Primary and Secondary Education Fitted values Figure 4. Ratio girls/boys in primary and secondary education and democracy Note: The regression represented by the fitted line reports a coefficient of 0.68 (robust SE ¼ 0.14), N ¼ 80, R 2 ¼ 0.12 from a regression of ratio of girls/boys in primary and secondary education on the Freedom House Measure of Democracy. The dependent variable in the present study is the female-to-male enrolment ratio at the primary and secondary, and tertiary levels. The primary and secondary enrolment ratios are combined. These outcome variables are used specifically because they are associated with the third MDG of education for all. The sample comprises 80 countries (see the Appendix), which constitutes a representative cross section of the regions covering Eastern Europe and Central Asia, the Middle East, South America, South and East Asia and Africa. The high-income OECD countries are excluded from the sample as gender parity in education exists in these nations. The data sources are provided in the Appendix. Estimation is carried out using both cross-sectional and panel data methods. All data were collected for the years 1990, 1995, 2000 and The observations are taken every fifth year for the panel data estimation year rather than averaging the data for every 5 years to minimize problems associated with serial correlation (see Acemoglu et al., 2008). For the cross-sectional estimation, the observations for the independent variables are averaged. A preliminary examination of the data presented in Table 1 shows that there are regional differences in the ratio of girls-to-boys in primary and secondary, and tertiary education. An enrolment ratio of unity indicates equality between females and males and deviations below (above) unity can be interpreted as a degree of male (female) advantage on the enrolment measure. South Asia and Africa show the greatest persistence in underrepresentation of girls. Regional differences are particularly striking at the tertiary level, where there are more female enrolments in all regions overall, except Asia and Africa. Summary statistics for the variables used in the study are presented in Table 2. The main independent variables in the study are a variable Time, created to measure the duration of suffrage, and democracy. Democracy in the present study is measured by

9 28 A. Cooray Table 1. The ratio of girls-to-boys in primary and secondary and tertiary education Country Ratio of girls/boys in primary and secondary education Ratio of girls/boys in tertiary education Middle East Algeria Bahrain Iran Jordan Libya Morocco Saudi Arabia Syria 0.94 n/a Tunisia UAE Yemen Eastern Europe and Central Asia Albania Armenia Azerbaijan Belarus China Croatia Estonia 1.00 n/a Kazakhstan Kyrgyzstan Latvia Lithuania Mongolia Slovenia Tajikistan Ukraine South and South East Asia Bangladesh Bhutan Cambodia India Indonesia Lao Nepal Malaysia Pakistan Philippines Sri Lanka 1.00 n/a Thailand Vietnam South America Argentina Bolivia 0.98 n/a Brazil Chile Colombia Costa Rica Cuba

10 Suffrage, Democracy and Gender Equality in Education 29 Table 1. Continued Country Ratio of girls/boys in primary and secondary education Ratio of girls/boys in tertiary education Ecuador El Salvador Guatemala Jamaica Nicaragua Panama Paraguay Peru Trinidad and Tobago Uruguay Venezuela Africa Benin Botswana Burundi Cameroon Chad Comoros Congo Ethiopia Ghana Kenya Lesotho Malawi Mauritania Mauritius Niger Nigeria Senegal 0.91 n/a Sierra Leone n/a 0.40 Sudan South Africa Uganda Zimbabwe Notes: Data sources provided in the Appendix; n/a: not available. two indices. 4 They are: the Polity IV Democracy Index, which uses a scale of zero to 10, with zero representing no democracy (full autocracy) and 10 representing full democracy; and the Freedom House Political Rights Index, which assigns countries a numerical rating from one to seven, with one indicating the highest degree of freedom and seven the lowest degree of freedom. To maintain consistency with the Polity IV Index, the Freedom House Political Rights Index is reversed so that seven represents the highest level of democracy and one the lowest level (see Brown, 2004). Both democracy indices are normalized to one. For the empirical estimation, the democracy variables for 1990, 1995, 2000 and 2005 are taken for the panel data estimation and the data points for these years are averaged for the cross-sectional estimation. Figures 3 and 4 plot the democracy variables against gender equality in education for 2005, the most recent year

11 30 A. Cooray Table 2. Summary statistics Variable Obs. Mean Standard deviation Minimum Maximum GDP per capita 1990 ($) 80 3, , ,830 Girls/boys in primary and secondary education (ratio) Girls/boys in tertiary education (ratio) Secondary enrolment ratio girls (in per cent) Secondary enrolment ratio boys (in per cent) Democracy Index Polity IV Freedom House Political Rights Index Government expenditure per primary student (as per cent of GDP per capita) Government expenditure per secondary student (as per cent of GDP per capita) Government expenditure per tertiary student (as per cent of GDP per capita) Employment in agriculture (as per cent of total employment) Fertility rate (births per woman) Current account balance (as per cent of GDP) Time (number of years) Notes: Data sources provided in the Appendix. The Polity IV Democracy Index ranges from zero to 10 and the Freedom House Political Rights Index from on to seven. Summary statistics reported for cross-sectional data. in the sample. Democracy appears to explain gender differentials in education in most countries except some African and Asian countries that fall below the regressions represented by the fitted lines in Figures 3 and 4. The other variable of interest is the duration of suffrage. It is possible to argue that the longer the time period since women acquired the right to vote, the greater the opportunity for women to avail themselves of educational expansion and increased employment opportunities. A variable Time is created to capture the length of time since women were granted the right to vote. This is calculated by subtracting from the latest year (2005 in this study) the year in which women gained the right to vote in each country. To control for the effect of other factors, the study includes a number of variables based on findings in the previous literature. As a high level of per capita income is associated with greater gender equality (Dollar & Gatti, 1999; Klasen, 2002), per capita income is used to measure a country s overall level of economic development. Government expenditure per student at the primary, secondary and tertiary levels is included to capture government investment in education. The initial (1990) secondary and tertiary enrolment ratios for girls and boys are also considered as they can have explanatory power for gender equality in education. Reduced fertility has been shown to improve gender parity in education and to lower the dependency burden Therefore, the fertility rate is used to estimate the effects of fertility on gender equality in education (see Dollar & Gatti, 1999; Klasen, 2002). As girls are most under-represented in education in Asia and Africa, and these countries are predominantly agricultural, the percentage of the population engaged

12 Suffrage, Democracy and Gender Equality in Education 31 in agriculture is considered. The studies of Cagatay & Ozler (1995), Fontana & Wood (2000) and Baliamoune-Lutz & McGillivray (2007) also show that increased openness to trade can lead to increased gender inequality in education due to the fact that labour for export industries is supplied primarily by unskilled females. To account for this, variables for employment in agriculture and trade openness are incorporated as independent variables. Employment in agriculture is measured by those employed in the agricultural sector as a percentage of total employment and trade openness by the current account balance as a percentage of GDP. Given the regional variation in educational inequality between the genders, four dummy variables are created for Africa, the Middle East, South East and South Asia, and South America and the West Indies, with Eastern Europe and Central Asia as the benchmark group. Cooray & Potrafke (2011), Norton & Tomal (2009), Shultz (2002), Inglehart & Baker (2000) and Dollar & Gatti (1999), among others, show that cultural influences have a significant impact on education. These studies use religion as a proxy for culture. Huntington (1993, 1996) argues that the world is divided into cultural zones based on cultural diversity that has persisted for centuries. These zones have been formed by religious traditions that remain influential (Inglehart & Baker, 2000). Similarly, Brown (2004) observes that religious norms can have a significant influence on women s education. The effects of cultural heritage on gender are therefore captured by four dummy variables for religion Roman Catholicism, Islam, Buddhism and Hinduism, with other Christian religions as the base group. Several studies have explored the effect of colonial rule on schooling outcomes. Acemoglu et al. (2008), La Porta et al. (1999) and Barro (1999) highlight the importance of historical factors in influencing institutions, Brown (2000) the effect of colonialism on enrolment, Huillery (2008) education among other variables, and Cooray (2009) the literacy rate. Owing to the persistence of institutional structures, these studies have shown that institutions set up during the colonial era have had a strong influence on current institutions. Huillery (2008) provides support for the influence of colonial investments in education on current education in West Africa, and Cooray (2009) the influence of the colonial legacy on the current literacy rates of the countries that were once colonized. Studies suggest that political participation by women was undermined during the colonial era by the introduction of cash crop agriculture and the preference for recruiting women in agriculture (Adams, 2006). This perhaps explains the gender disparity in education in favour of males in Africa and Asia, despite relatively high levels of democracy. Therefore, a dummy variable is created to control for the influence of colonialism on enrolment. As many of the countries that have a greater proportion of male enrolments are either former British or former French colonies, the colonial dummy variable takes on a value of one if a country was a British or French colony, and zero otherwise. 4. Empirical Results The following model forms the basis of the cross-sectional estimation: E FMi ¼ ad i þ gs i þ x i b þ fr i þ v i þ y i ; where E FMi is the female-to-male enrolment ratio at the primary and secondary, and tertiary levels for country i. D i represents the democracy variables and S i represents the duration of women s suffrage. All other control variables mentioned in Section 3 are ð1þ

13 32 A. Cooray captured by the vector x i. R i represents a set of regional/religion dummy variables, and n i is a colonial dummy variable. y i is a random error term that captures all other variables. The study uses both cross-sectional and panel data methods to estimate the model. Several alternative regression methods are employed in this study. The initial estimation is carried out using ordinary least squares (OLS). Robustness checks are carried out using both instrumental variables estimation and panel data methods. Given that the regional coefficients are found to be significant for the full sample, estimation is also carried out by disaggregating the full sample by region. The regional estimation is carried out by using the Seemingly Unrelated Regression (SUR) estimation method. 4.1 Cross-sectional Estimation: Full Sample Table 3 presents cross section results for the full sample. Columns 1 4 in the table are estimated with the enrolment ratio of girls-to-boys at the primary and secondary level as the dependent variable and columns 5 8 are estimated with the enrolment ratio of girls-toboys at the tertiary level as the dependent variable in Both the Polity IV and Freedom House Democracy indices are used to ensure the robustness of the results to the measure of democracy. Columns 1, 3, 5 and 7 are estimated with the Polity IV Democracy Index and columns 2, 4, 6 and 8 with the Freedom House Democracy Index. The variables of interest are Time and the democracy index. The democracy index is statistically significant at the 1 and 5% levels in columns (1), (2), (5) and (6). Column 1 in Table 3 shows that a one unit increase in the democracy index is associated with an 0.08% increase in the gender parity ratio at the primary and secondary levels. Note that when religion is controlled for, the democracy indices lose statistical significance. An examination of the coefficients on the religion dummy variables shows that the Hinduism, Islam and Roman Catholicism variables are statistically significant at the 10% level. A country s religious beliefs seem to have an influence on gender dis/parity in education. While Catholicism has a positive effect on gender parity compared with the benchmark group of other Christian religions, Islam and Hinduism have a negative effect. The Time variable is positive and significant in columns 1 4, 7 and 8, providing support for the argument that the longer the duration since suffrage the higher the level of gender equality. Column 1, for example, indicates that the gender parity ratio at the primary and secondary level increases by 0.002% for each additional year of suffrage. The coefficient on the colonialism dummy variable is significant in columns 1 4, 6 and 8. The explanatory power of the models is in the range Columns (1) (2) and (5) and (6) indicate that gender parity is lower in all regions compared with the benchmark group, which is Eastern Europe and Central Asia. The coefficients on the Africa and Asia regional dummy variables are statistically significant at the 5 and 10% levels. Hence, in the regions in which per capita income is lowest, girls are least represented in primary and secondary, and tertiary education. Employment in agriculture is significant at the 5 and 10% levels in all equations and the fertility rate is significant at the 10% level in columns (3), (4) and (7). The joint statistical significance of the region coefficients cannot be rejected at the 10% level, suggesting that region matters for gender disparity in education. Estimation is therefore carried out on the sample by region in the empirical analysis that follows to account for regional heterogeneity.

14 Suffrage, Democracy and Gender Equality in Education 33 Table 3. Full sample: OLS estimation Columns 1 4 dependent variable: enrolment ratio girls/boys primary and secondary 2005 Columns 5 8 dependent variable: enrolment ratio girls/boys tertiary 2005 Independent variables (1) (2) (3) (4) (5) (6) (7) (8) log per capita income (0.012)* log exp. per student at primary level (0.019) log exp. per student at secondary level (0.021) (0.011)** (0.020) (0.021) (0.014)* (0.024)* (0.015) (0.009)*** (0.025)* (0.014) (0.110)*** log exp. per student at tertiary level (0.121) log enrolment ratio secondary girls (0.072)* log enrolment ratio secondary boys (0.063)* Africa dummy (0.045)* Asia dummy (0.033)* Middle East dummy (0.081) South America dummy (0.041) (0.053)** (0.065) (0.042)** (0.034)* (0.072) (0.043) Buddhism dummy (0.025) Hinduism dummy (0.018)* Islam dummy (0.026)* Roman Catholic dummy (0.020)* (0.066)* (0.061)* (0.025)* (0.061)* (0.054)* (0.016)* (0.050)** (0.033)* (0.030) (0.020) (0.031) (0.022)* (0.030)* (0.021)* (0.141)*** (0.110)** (0.111)** (0.020) (0.015)* (0.015)* (0.044)* (0.025)* (0.037) (0.021) (0.101) (0.021) (0.021)* (0.19)* (0.018)* (0.131) (0.013)** (0.011)* (0.018) (0.011)* (0.018) (0.008)* (0.019)* (0.017)* (Continued)

15 34 A. Cooray Table 3. Continued Columns 1 4 dependent variable: enrolment ratio girls/boys primary and secondary 2005 Columns 5 8 dependent variable: enrolment ratio girls/boys tertiary 2005 Independent variables (1) (2) (3) (4) (5) (6) (7) (8) Democracy (Polity IV) (0.041)** Democracy (Freedom House) (0.010)*** log employment in agriculture (0.010)*** log fertility rate (0.025) log current account (0.001) (0.012)** (0.026) (0.002) (0.010) (0.016)* (0.017)* (0.001) (0.015) (0.017)** (0.026)* (0.001) (0.005)** (0.027)* (0.015) (0.001) (0.005)** (0.025)** (0.018) 0.01 (0.01) (0.015) (0.010)*** (0.030)* (0.001) (0.016) (0.012)*** (0.036) Time (0.001)** (0.001)* (0.001)* (0.001)** (0.001) (0.001) (0.001)* (0.001)* Colony dummy (0.031)* (0.025)** (0.020)* (0.011)** (0.014) (0.008)** (0.002) (0.001)* Joint statistical significance region dummy: p-value Joint statistical significance religion dummy: p-value Adj R Observations (0.002) Note: Standard errors reported in parentheses. ***, **, *Significant at the 1, 5 and 10% levels, respectively.

16 Suffrage, Democracy and Gender Equality in Education Panel Data Estimation Given that the cross-country results suggest region-specific effects, the model is estimated using panel data in this section to capture any country-specific effects. The advantage of using panel data is that they contain more degrees of freedom and greater sample variability than cross-sectional data, improving the efficiency of estimates. The panel data model takes the following form: E FMit ¼ ad it þ gs it þ x it b þ fr i þ v i þ y it ; ð2þ where E FMit is the female/male enrolment ratio at the primary and secondary, and tertiary levels for country i in period t. D it represents the democracy variable and S it the duration of suffrage, respectively, for country i in period t. All control variables mentioned in Section 3 are captured by the vector x it. R i represents a set of regional/religion dummy variables, and n i a colonialism dummy variable. y it is a random error term that captures all other variables. A question that arises in this regard is whether a fixed or random effects model is more appropriate. A fixed effects model involves estimating the individual country effects as parameters. This leads to a substantial loss in degrees of freedom. In particular, when the number of countries exceeds the number of time periods, as is the case in this study, a random effects model will permit more efficient use to be made of the available data. The use of a random effects model is supported by a Hausman test yielding p-values of 0.12 and 0.14, respectively, in Table 4 for columns 2 and 4. Therefore, a random effects model is used for the purpose of estimation. Estimation is also carried out using pooled OLS. A panel for the period is used with the observations taken every fifth year. The pooled OLS (ordinary least squares) and random effects GLS (generalized least squares) results are reported in Table 4. The openness variable is dropped in the regional estimation as this variable was statistically insignificant in all of the above equations. The rest of the analysis is carried out using the Polity IV Democracy Index as the measure of democracy given that both the Freedom House and Polity IV indices yield similar estimates in Table 3. The coefficient on the democracy index is significant in columns (1) and (4). The coefficients on the Time variable are significant at the 1, 5 and 10% levels. Per capita income has a positive and significant impact on gender parity in enrolment, and employment in agriculture a negative significant effect. Of the religion dummy variables, Islam and Hinduism have a significant negative effect on the gender parity ratio and Roman Catholicism a positive significant effect. The coefficients on colonialism are negative and significant in all equations. 4.3 Dividing the Sample by Region The SUR method is applied to the data to estimate the region disaggregated models. The SUR technique is a systems estimator that takes into account correlations of the residuals across equations, and therefore interdependence between the error terms. An advantage of using SUR is that it yields more efficient estimates, as opposed to estimating the equations separately by using OLS. In addition, this method does not require instruments. The base group for religion in each region (in each column) are other religions for which estimates are not reported. Table 5 presents results for estimation with the primary and secondary enrolment ratio as the dependent variable.

17 36 A. Cooray Table 4. Panel data estimation Dependent variable: enrolment ratio girls/boys primary and secondary Dependent variable: enrolment ratio girls/boys tertiary (1) (2) (3) (4) Independent variables Pooled OLS Random effects Pooled OLS Random effects Per capita income (0.012)** (0.006)*** (0.040)*** (0.042)*** Exp. per student at primary and secondary level (0.010) (0.010) Exp. per student at tertiary level (0.035)*** (0.022)*** Enrolment ratio secondary girls (0.008)* (0.010)* Enrolment ratio secondary boys (0.009)* (0.008)* Enrolment ratio tertiary girls (0.007)** (0.008)* Enrolment ratio tertiary boys (0.007)** (0.008)* Buddhism dummy (0.029)* (0.004) (0.213) (0.120) Hinduism dummy (0.033)* (0.023)* (0.020)* (0.018)** Islam dummy (0.020)** (0.020)** (0.026)* (0.029)* Roman Catholic dummy (0.019)*** (0.035)* (0.020)*** (0.010)*** Democracy (0.002)*** (0.010) (0.015) (0.007)** Employment in agriculture (0.024)*** (0.016)*** (0.080)* (0.066)* Fertility rate (0.011) (0.012) (0.023) (0.053)** Time (0.001)** (0.001)** (0.001)** (0.001)** Colony dummy (0.008)* (0.008)** (0.008)** (0.010)* R Hausman test: p-value Observations Note: Robust standard errors reported in parentheses. ***, **, *Significant at the 1, 5 and 10% levels, respectively.

18 Suffrage, Democracy and Gender Equality in Education 37 Table 5. SUR estimation by region for primary and secondary level; dependent variable: enrolment ratio girls/boys in primary and secondary education Middle East Eastern Europe and Central Asia South East and South Asia Africa South America and the West Indies Independent variables (1) (2) (3) (4) (5) Per capita income (0.012)** (0.010)** (0.010)** (0.009)** (0.010)** Exp. per student at primary level (0.015)* (0.016)* (0.014)** (0.011)** (0.015)* Exp. per student at secondary level (0.019) (0.012) (0.013) (0.011) (0.011) Enrolment ratio secondary girls (0.008)** (0.009)** (0.010)* (0.009)* (0.010)* Enrolment ratio secondary boys (0.008)* (0.008)* (0.008)* (0.007)* (0.009)* Buddhism dummy (0.005) (0.003) Hinduism dummy (0.002)** Islam dummy (0.005) (0.002) (0.004)** (0.011)* Christianity dummy (0.001) (0.011) Roman Catholic dummy (0.011) (0.003) (0.002) Democracy (0.005) (0.006) (0.002)*** (0.003)** (0.001)*** Employment in agriculture (0.013)** (0.014)** (0.012)*** (0.011)*** (0.013)** Fertility rate (0.022) (0.015)* (0.018)* (0.017)* (0.020) Time (0.003) (0.002)*** (0.002) (0.007) (0.004) Colony dummy (0.011)* (0.020) (0.013)** (0.010)** (0.020) R Observations Notes: Standard errors reported in parentheses. Breusch-Pagan test of independence: x 2 (10) ¼ [0.00]. ***, **, *Significant at the 1, 5 and 10% levels, respectively.

19 38 A. Cooray The democracy index is significant at the 1% level for South East and South Asia, Africa and South America and the West Indies. An examination of the estimates in column 3, for example, indicates that a one unit increase in the democracy index leads to a 0.006% increase in the gender parity ratio. The variable Time is significant at the 5% level in Eastern Europe and Central Asia, the region in which women have had the right to vote for the longest period. Column 2 shows that an additional year of suffrage leads to a 0.005% increase in the gender parity ratio in this region. In many of the Asian, African and Middle Eastern countries, women only gained the right to vote after independence. Despite the fact that South America gained independence long before these regions, women in South America only gained the right to vote around the same time as women in these regions. Therefore, the duration of suffrage appears to matter for gender equality in education. The coefficient on the colonialism dummy variable is significant in the Africa, Asia and Middle Eastern regions, demonstrating the importance of historical factors in influencing school enrolment. Per capita income and government expenditure per student at the primary level are significant for all regions. The results presented in column 4 indicate that a 1% increase in per capita income leads to a 0.018% increase in the gender parity ratio in Africa and a 0.020% increase in the gender parity ratio in South East and South Asia. The initial level of the secondary enrolment ratio is positive and significant at the 10% level for all regions. Agricultural employment is significant at the 1% level for Africa and Asia, and at the 5% level for the rest of the regions. The fertility rate is significant at the 10% level in Eastern Europe and Central Asia, East and South Asia and Africa. The coefficients on the Islam and Hinduism dummy variables are negative and statistically significant in the East and South Asia region and the Islam dummy variable is negative and statistically significant in the Africa region. Note that the Islam dummy variable is negative but not statistically significant in the Middle East. Table 6 reports results by region with female/male enrolment at the tertiary level as the dependent variable. The results for the tertiary level are slightly different from those for the primary and secondary levels, with government expenditure per student and fertility losing significance. The coefficient on democracy is significant at the 5% level in Africa and South America and the West Indies. A one unit increase in the democracy index leads to a 0.011% increase in the gender parity ratio in Africa. The coefficient on per capita income is significantatthe5%levelforallregions.a1% increase in per capita income leads to a 0.23% increase in the gender parity ratio in the Middle East and a 0.24% increase in the gender parity ratio in Eastern Europe and Central Asia. The variable Time is significantatthe5%levelagainineasterneurope and Central Asia. As before, the colonialism dummy variable is significant in the Middle East, Asia and Africa. Religion has a significant negative impact on gender parity in education in the Middle East at the tertiary level. Similarly, the coefficient on Islam is significantly negatively related to gender parity in education in Africa, and Hinduism and Islam negatively related to education in Asia. There is a significant negative relationship between the female/male enrolment ratio and agricultural employment. The explanatory power of the models is high, with the independent variables explaining 84 89% of the variation in the dependent variable. A Breusch- Pagan test of independence of residuals from each equation yields a test statistic of x 2 (10) ¼ with a p-value of 0.00, rejecting the null hypothesis of independence of error terms.

20 Suffrage, Democracy and Gender Equality in Education 39 Table 6. SUR estimation by region at the tertiary level; dependent variable: enrolment ratio girls/boys in tertiary education Middle East Eastern Europe and Central Asia South East and South Asia Africa South America and the West Indies Independent variables (1) (2) (3) (4) (5) Per capita income (0.120)** (0.114)** (0.119)** (0.117)** (0.117)** Exp. per student at (0.026) (0.120) (0.027) (0.120) (0.120) secondary level Exp. per student at (0.106) (0.039) (0.085) (0.079) (0.079) tertiary level Enrolment ratio tertiary girls (0.014)* (0.014)* (0.015)* (0.013)* (0.015)* Enrolment ratio tertiary boys (0.013)* (0.014)* (0.012)* (0.013)* (0.012)* Buddhism dummy (0.005) (0.009) Hinduism dummy (0.004)* Islam dummy (0.002)** (0.004) (0.009)** (0.040)** Christianity dummy (0.004) (0.012) Roman Catholic dummy (0.005) (0.004) (0.004) Democracy (0.012) (0.013) (0.011) (0.005)** (0.010)** Employment in agriculture (0.011)** (0.015)** (0.010)** (0.010)** (0.013)** Fertility rate (0.020) (0.021) (0.021) (0.021) (0.021) Time (0.006) (0.002)** (0.004) (0.006) (0.006) Colony dummy (0.016)* (0.019) (0.011)** (0.014)** (0.010) R Observations Note: Standard errors reported in parentheses. Breusch-Pagan test of independence: x 2 (10) ¼ [0.00]. **, *Significant at the 5 and 10% levels, respectively.

21 40 A. Cooray 4.4 Robustness Tests Some tests have been performed to ensure the robustness of the results. Two alternative measures of democracy, the Polity IV Democracy Index of Marshall & Jaggers (2006) and the Freedom House Political Rights Index (2009), are used to ensure the robustness of the results to the measure of democracy. The results are consistent when adopting the different measures of democracy. A preliminary analysis of the data in Table 1 indicated that there were regional discrepancies in gender equality in education. In order to capture these differences, four regional dummy variables were defined for: (1) Africa, (2) the Middle East, (3) South East and South Asia and (4) South America and the West Indies, with Eastern Europe and Central Asia as the benchmark group. Given the significance of the Asia and Africa dummy variables in Table 3, estimation was also carried out by dividing the full sample by region (see Tables 5 and 6). The results indicate regional differences as suggested by the initial analysis carried out in the Data section (see Table 1). Therefore, estimation is carried out using panel data methods and the IV technique to test the robustness of the results to the estimation method. A Hausman test supports the use of a random effects model. The results of the panel data estimation are similar to those of the cross-sectional analysis, suggesting that the results are robust to the estimation technique. Given, however, that the democracy indices vary only slowly over time, greater emphasis is placed on the cross-sectional results. 5 Simple OLS estimation of the equations can lead to biased and inconsistent parameter estimates due to regressor endogeneity. If the duration of suffrage were also a function of gender parity in education, the OLS estimates would be biased and inconsistent. This can be overcome by using the IV technique. Thus, in order to correct for any endogeneity bias that may be present in the models, the equations in Table 3 are re-estimated using the IV technique. The variable Time (duration of suffrage) is instrumented with the year of suffrage. 6 The results for the IV estimation are presented in Table 7. The democracy index is significant in columns (1), (2), (5) and (6) in Table 7. It loses statistical significance when religion is controlled for. The Time variable is positive and statistically significant in all equations except columns (2), (5) and (8). Per capita income and employment in agriculture are statistically significant. The Hinduism and Islam dummy variables are negative and statistically significant. The results are consistent with those obtained under the OLS method in Table 3 and also with the results obtained under panel data estimation in Table 6. Two diagnostic tests were carried out on the IV estimates. A Hausman (1978) test for any statistically significant difference between the OLS and IV estimates suggests no evidence of any significant difference between the OLS and IV estimates. A Sargan test (1964) suggests that the instruments are valid and that the model is correctly specified. 4.5 Limitations of the Study Democracy, suffrage and culture are broad concepts that are difficult to measure, particularly at the macroeconomic level. By considering the duration of suffrage (Time) rather than the year of suffrage, the study attempts to capture the change in political institutions over time. Of course, it should be noted that this variable can account for a large number of sources of variation in the outcome variable. 7 If Time captures other omitted