Culture, Gender and Math Revisited

Transcription

1 Culture, Gender and Math Revisited Brindusa Anghel Banco de España Núria Rodríguez-Planas* City University of New York (CUNY), Queens College Anna Sanz-de-Galdeano University of Alicante and IZA January 2018 Abstract Using 2003 to 2015 PISA data and across country variation, we first show that the positive correlation between gender equality and the gender gap in mathematics only holds for OECD countries or those above-average gender equality. In contrast, in non-oecd countries or below-average gender equality, the math gender gap widens in more gender-equal countries, suggesting that in traditional societies, there may initially be male backlash as social gender norms relax. Exploiting time-variation in different measures of gender equality, we then explore whether changes in the math gender gap within a country are associated with changes in gender equality. While we find no evidence of this in OECD countries or those above-average gender equality, there is evidence that improvements in female political empowerment within a country reduce the math gender gap in non-oecd countries or those below-average gender equality. Our findings are the first to underscore a differential relationship between gender equality and the math gender gap by gender-parity level. Keywords: Gender Gap in Math, Culture and institutions JEL: I21, I24, J16, Z13 * To whom correspondence should be addressed. nuria.rodriguezplanas@qc.cuny.edu 1

2 Guiso and co-authors seminal Science paper (2008) was the first to provide suggestive evidence that societal factors (as opposed to biological-based explanations) are behind the underperformance in math tests of girls relative to boys. 1 Using the 2003 PISA test and a variety of gender equality indices, these authors studied whether countries with greater gender equality have smaller math gender gaps. Focusing on the 40 countries that conducted PISA testing in 2003, they found cross-sectional evidence that this is indeed the case: i) using countrylevel data and controlling for the countries economic conditions with the GDP per capita, and ii) using student-level data and controlling for individuals characteristics and countries GDP per capita. Taking advantage of four new waves of PISA tests and over a decade of variation in different gender equality indices, we reassess the relative importance of societal factors behind the math gender gap. Our strategy is three-fold. First, we replicate Guiso et al. s (2008) cross-country analysis and additionally study whether their finding holds for subsequent years. Table 1 and Appendix Table A.1 show that the positive correlation between gender equality and the gender gap in mathematics holds for the same sample of countries as in Guiso et al. (2008) and for OECD countries, respectively. [insert Table 1] However, in non-oecd countries there is no evidence that greater gender equality is associated with a smaller math gender gap (shown in Appendix Table A.2). In fact, for non-oecd countries, the relation between gender equality and math gender gap is negative (and sometime statistically significant) when gender equality is measured with an index of cultural attitudes towards women based on the average level of disagreement to such statement as: When jobs are scarce, men should have more right to a job than women from the World Value Survey, or with the political empowerment index from the World Economic Forum. Most interestingly, we then pool both OECD and non-oecd countries and uncover a non-linear pattern between gender equality and the math gender gap (Table 2). [insert Table 2] 1 These findings were later supported by others, such as Fryer and Levitt (2010) or Pope and Sydnor (2010), also using cross-sectional data, in this case from TIMMS (the former) and the US (the latter). With TIMMS data, results only held if Muslim countries were excluded from the sample. 2

3 More specifically, we find that in those countries with gender equality indices below the mean, greater gender equality is associated with lower girls performance in math tests relative to that of boys. In contrast, in countries with above-average gender equality, the performance of girls relative to boys in math tests improves with gender equality. This pattern is robust to alternative measures of gender equality such as the World Economic Forum gender gap index, the World Value Survey cultural index towards women, the female labor force participation or the World Economic Forum political empowerment index. It suggests that while in non-traditional societies, the math gender gap is smaller and disappears in those countries with greater gender equality, in more traditional societies, the math gender gap widens in those countries with greater gender equality. This is consistent with the male backlash hypothesis by which as traditional societies social gender norms relax, men may initially feel threaten by potentially more resourceful women (via income, employment, education, or political empowerment), and they may seek to prevent that future generations (their daughters) engage in subjects traditionally more prevalent among men (such as mathematics) to counteract women s relative increase in power or to extract women s resources (Heise and Garcia-Moreno, 2002; Eswaran and Malhotra, 2011; True, 2012). Once a minimum threshold of gender equality has been reached (which appears to be around the mean), then the association between gender equality and math gender gap is reversed and becomes positive. Because three fourths of the countries used in Guiso et al. s paper were OECD countries, their finding is explained by the fact that in most of their countries the level of gender equality was above-average. Finally, we exploit both across-country and across-time variation in our measures of gender equality and ask a different question, namely whether countries with larger increases (decreases) in gender equality experience greater decreases (increases) of the math gender gap. By exploiting both across time and country variation, our identification strategy controls for time-invariant unobserved factors that are both correlated with gender equality and the math gender gap through the full set of time and country fixed effects. Most importantly, by adding country time trends, identification emerges from deviations of the gender equality indices from their long-term trend. [insert Tables 3 and 4] Interestingly, we find again a differential pattern for OECD and non-oecd countries. Among the former, we find no evidence that the math gender gap decreases in those countries with greater increases in gender equality, suggesting that earlier findings for OECD countries were picking up spurious correlation between the math gender gap and gender equality instead of a causal relationship 3

4 (columns 4 and 5 from Table 3). 2 Our findings underscore that greater gender equality does not lead to better relative test performance in mathematics of girls relative to boys in OECD countries. In contrast, we find that for non-oecd countries, girl s relative performance in math-test scores relative to that of boys increases in countries with greater increases in female political empowerment at a decreasing rate (Table 4). This finding is statistically significant at the 5 percent level. The effect is also positive (albeit not statistically significant) for the gender gap and the WVS index. We find that Guiso et al. s (2008) findings only hold for OECD countries or those above a certain gender-equality threshold, but that in these countries, withincountry variation in gender equality does not affect the math gender gap. In contrast, we find that the association between gender equality and math gender gap is negative for non-oecd countries or below a certain gender-equality threshold. In such cases, however, we find some evidence that improvements in female political empowerment seems to be associated with improvements in girls performance in mathematics relative to boys. Further research ought to aim at better understanding the role of different institutions and social gender norms in explaining the math gender gap. 2 These findings were later supported by others, such as Fryer and Levitt (2010) or Pope and Sydnor (2010), also using cross-sectional data, in this case from TIMMS (the former) and the US (the latter). 4

5 References Akerlof, George A., and Rachel E. Kranton Economics and Identity. The Quarterly Journal of Economics CXV(3): Albanesi, Stefania, and Claudia Olivetti Home Production, Market Production and the Gender Wage Gap: Incentives and Expectations. Review of Economic Dynamics 12(1): China Daily "Policies Help Education of Girls." January 15, Eswaran, Mukesh and Nisha Malhotra "Domestic violence and women's autonomy in developing countries: theory and evidence", Canadian Journal of Economics, Vol.44, Issue 4, Fernández, Raquel Women, Work and Culture. Journal of the European Economic Association 24(4): Does Culture Matter? In Handbook of Social Economics, Elsevier Ltd, Fernández, Raquel, and Alessandra Fogli Culture: An Empirical Investigation of Beliefs, Work, and Fertility. American Economic Journal: Macroeconomics 1(1): Fryer, Ronald, and Steven Levitt An Empirical Analysis of the Gender Gap in Mathematics. American Economic Journal: Applied Economics 2(2): Guiso, Luigi, Ferdinando Monte, Paola Sapienza, and Luigi Zingales Culture, Gender, and Math. Science (New York, N.Y.) 320: Hausmann, Ricardo, LDA Tyson, and Saadia Zahidi The Global Gender Gap Report World Economic Forum. Heise L. and Garcia-Moreno C "Violence by intimate partners", in: World report on violence and health, edited by Etienne G. Krug, Linda L. Dahlberg, James A. Mercy, Anthony B. Zwi and Rafael Lozano. Geneva, Switzerland, World Health Organization [WHO]: Marmot, M. G. et al Epidemiologic Studies Of Coronary Heart Disease And Stroke In Japanese Men Living In Japan, Hawaii And California: Prevalence Of Coronary And Hypertensive Heart Disease And Associated Risk Factors. American Journal of Epidemiologic. 102(6): Pope, Devin G. and Justin R. Sydnor Geographic Variation in the Gender Differences in Test Scores, Journal of Economic Perspectives, Vol. 24, No. 2, (pp ) True Jacqui The Political economy of Violence against Women, Oxford University Press 5

6 Table 1. Gender gap in Math and Institutions. Sample of countries from Guiso et al. (2008) paper PISA 2003 PISA 2006 PISA GGI *** ** *** (24.595) (23.635) (21.489) Average WVS indicators Female labor force participation rate Political Empowerment Index Log GDP per capita (8.116) (5.599) (6.610) 0.414*** 0.242** 0.414*** (0.135) (0.117) (0.119) *** * *** (9.256) (8.419) (8.672) *** ** ** * * * * *** * *** *** (2.116) (2.982) (2.113) (2.236) (2.152) (2.549) (1.932) (2.152) (2.309) (3.237) (2.071) (2.449) Constant * * ** *** (19.637) (24.820) (20.465) (21.930) (19.696) (20.734) (18.493) (21.368) (20.022) (26.516) (19.491) (24.231) Observations R-squared

7 Table 1 cont. PISA 2012 PISA GGI *** ** (24.811) (26.065) Average WVS indicators Female labor force participation rate Political Empowerment Index Log GDP per capita (7.176) (6.185) 0.550*** 0.319* (0.154) (0.174) ** (9.119) (8.935) *** *** *** *** ** (3.024) (3.857) (2.509) (2.972) (3.229) (3.563) (2.938) (3.011) Constant ** ** ** * ** (24.422) (32.638) (23.273) (29.642) (26.184) (30.219) (27.589) (30.172) Observations R-squared Notes: The gender gap is the coefficient of the female dummy in a country-by-country regression of PISA scores on a female dummy. As in Guiso et al. (2008), for each country we use the sample of students that are above the 50th percentile of ESCS of that country. GDP per capita is GDP in PPP. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 7

8 Table 2. Gender gap in Math and Institutions. Pool of all countries from PISA sample GGI *** ( ) GGI squared *** ( ) Average WVS indicators *** (27.330) Average WVS indicators squared *** (5.160) Female labor force participation rate *** (0.301) Female labor force participation rate squared 0.013*** (0.003) Political Empowerment Index *** (12.448) Political Empowerment Index squared *** (17.788) Log GDP pc PPP *** ** * *** (1.010) (0.914) (0.955) (0.895) Year dummy *** *** *** *** (1.791) (1.973) (1.664) (1.774) Year dummy *** *** *** *** (1.559) (1.674) (1.446) (1.489) Year dummy * ** ** ** (1.641) (1.892) (1.621) (1.585) Year dummy ** * ** ** (1.682) (1.862) (1.649) (1.585) Constant *** *** *** *** (71.797) (39.006) (10.466) (8.941) Observations R-squared Notes: The gender gap is the coefficient of the female dummy in a country-by-country regression of PISA scores on a female dummy. As in Guiso et al. (2008), for each country we use the sample of students that are above the 50th percentile of ESCS of that country. GDP per capita is GDP in PPP. Here we use the pooled sample of countries from all PISA waves (PISA ). Year dummy 2015 is the reference dummy. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 8

9 Table 3. Gender gap in Math and Institutions. Panel of OECD countries from PISA sample GGI *** ** ( ) ( ) ( ) ( ) ( ) GGI squared *** *** ( ) ( ) ( ) ( ) ( ) Average WVS indicators *** *** ** * (49.157) (49.876) (96.210) ( ) ( ) Average WVS indicators squared *** *** ** (8.664) (8.875) (18.112) (19.935) (20.932) Female labor force participation rate ** (0.497) (0.498) (1.399) (1.454) (1.421) Female labor force participation 0.014*** 0.013** rate squared (0.005) (0.005) (0.015) (0.016) (0.016) Political Empowerment Index *** ** (11.315) (11.853) (15.192) (17.393) (16.336) Political Empowerment Index squared *** *** (15.557) (16.026) (30.154) (30.407) (29.823) Log GDP per capita PPP, Log GDP per capita PPP squared No Yes Yes Yes Yes Time dummies No No No Yes No Country linear time trend No No No No Yes Country FE No No Yes Yes Yes Observations 107 to to to to to 166 R-squared to to to to to Number of countries 22 to to to 34 Notes: The gender gap is the coefficient of the female dummy in a country-by-country regression of PISA scores on a female dummy. As in Guiso et al. (2008), for each country we use the sample of students that are above the 50th percentile of ESCS of that country. Each column shows the coefficients of gender equality indices and their square from separate regressions. Estimations include a constant not reported here. GDP per capita is GDP in PPP. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 9

10 Table 4. Gender gap in Math and Institutions. Panel of Non-OECD countries from PISA sample GGI (1, ) (1, ) (1, ) (1, ) (1, ) GGI squared ( ) ( ) ( ) ( ) ( ) Average WVS indicators (64.721) (66.770) ( ) ( ) ( ) Average WVS indicators squared (13.019) (13.357) (33.614) (37.878) (36.887) Female labor force participation rate ** ** (0.339) (0.353) (2.642) (2.724) (2.734) Female labor force participation 0.009** 0.009** rate squared (0.004) (0.004) (0.024) (0.025) (0.025) Political Empowerment Index ** ** ** (40.547) (42.395) (54.810) (57.326) (57.704) Political Empowerment Index squared ** ** ** ** ( ) ( ) ( ) ( ) ( ) Log GDP per capita PPP, Log GDP per capita PPP squared No Yes Yes Yes Yes Time dummies No No No Yes No Country linear time trend No No No No Yes Country FE No No Yes Yes Yes Observations 99 to to to to to 120 R-squared 0.01 to to to to to Number of countries 31 to to to 39 Notes: The gender gap is the coefficient of the female dummy in a country-by-country regression of PISA scores on a female dummy. As in Guiso et al. (2008), for each country we use the sample of students that are above the 50th percentile of ESCS of that country. Each column shows the coefficients of gender equality indices and their square from separate regressions. Estimations include a constant, not reported here. GDP per capita is GDP in PPP. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 10

11 APPENDIX Table A1. Gender gap in Math and Institutions. OECD countries from PISA sample PISA 2003 PISA 2006 PISA GGI *** ** *** (23.297) (24.220) (24.168) Average WVS * * * indicators (10.365) (7.639) (10.201) Female labor force 0.526*** 0.385*** 0.547*** participation rate (0.132) (0.135) (0.164) Political Empowerment *** (8.370) * (8.415) ** (9.494) Index Log GDP per capita *** ** * (2.829) (4.645) (2.769) (2.881) (3.102) (4.424) (3.020) (3.086) (3.363) (5.983) (3.598) (3.620) Constant (26.548) (35.555) (26.715) (29.257) (28.068) (35.208) (28.673) (31.467) (30.269) (47.110) (33.765) (36.790) Observations R-squared

12 Table A1. cont. PISA 2012 PISA GGI *** ** (23.723) (26.983) Average WVS indicators (8.022) (8.053) Female labor force 0.430** 0.361* participation rate (0.187) (0.205) Political *** * Empowerment Index (8.177) (9.257) Log GDP per capita *** * * (3.857) (5.353) (3.927) (3.710) (4.201) (5.682) (4.217) (3.972) Constant (32.678) (47.050) (37.096) (37.836) (37.256) (51.227) (40.165) (40.733) Observations R-squared Notes: The gender gap is the coefficient of the female dummy in a country-by-country regression of PISA scores on a female dummy. As in Guiso et al. (2008), for each country we use the sample of students that are above the 50th percentile of ESCS of that country. GDP per capita is GDP in PPP. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 12

13 Table A2. Gender gap in Math and Institutions. Non-OECD countries from PISA sample PISA 2003 PISA 2006 PISA GGI ( ) (53.784) (58.457) Average WVS indicators (18.318) (6.834) (7.741) Female labor force participation rate (0.345) (0.145) (0.167) Political Empowerment (89.311) (26.300) (29.630) Index Log GDP per capita * (9.950) (7.607) (5.635) (11.240) (2.397) (2.426) (2.135) (2.290) (2.917) (2.840) (2.390) (2.763) Constant ( ) (90.372) (52.066) ( ) (46.907) (29.421) (21.369) (22.090) (56.593) (35.001) (24.034) (27.785) Observations R-squared

14 Table A2 cont. PISA 2012 PISA GGI (70.771) (50.480) Average WVS ** * indicators (7.870) (7.398) Female labor force participation rate (0.180) (0.138) Political *** Empowerment Index (30.283) (24.642) Log GDP per capita (3.253) (3.094) (2.818) (2.968) (2.749) (2.801) (2.461) (2.752) Constant (61.392) (39.134) (28.506) (30.677) (44.963) (34.280) (23.950) (27.959) Observations R-squared Notes: The gender gap is the coefficient of the female dummy in a country-by-country regression of PISA scores on a female dummy. As in Guiso et al. (2008), for each country we use the sample of students that are above the 50th percentile of ESCS of that country. GDP per capita is GDP in PPP. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 14