Does Education Reduce Sexism? Evidence from the ESS

Transcription

1 Does Education Reduce Sexism? Evidence from the ESS - Very Preliminary - Noelia Rivera Garrido January 30, 2017 Abstract This paper exploits several compulsory schooling laws in 17 European countries to estimate the causal effect of education on the attitude toward women s place in the family. In particular, we study if education affects people s opinion about if the woman has to leave labor market to take care of the family. We exploit reforms changing the number of years of compulsory education to obtain a source of exogenous variation that can be used as an instrument for education. We found that education reforms provide a powerful instrument which significantly increases years of schooling. Results indicate that education significantly reduces the probability of agreeing with women leaving labor market to take care of the family in more than 4 percentage points. Moreover, this effect is heterogeneous according to gender such that it is stronger for men. Keywords: Education, beliefs, female labor force participation, empowerment, compulsory schooling laws, instrumental variables, ESS JEL Codes: A13, C26, J16, J21, J26 Thanks to Pedro Albarrán, Marisa Hidalgo and Iñigo Iturbe-Ormaetxe for their help and advise supervising this project Department of Economics, University of Alicante, Spain. noelia.rivera@ua.es

2 1 Introduction Sexist perceptions in society may have negative implications for the economy as a whole. If the majority opinion is that women must stay at home, the economy will lose an important part of the talent and this could be very detrimental. There is evidence showing that countries in which women participate little in the labor market grow less 1. In Table 1 we show that there is an association between the percentage of people who agree with the statement: a woman should be prepared to cut down on her paid work for sake of her family and some socio-economic indicators. In particular, countries with greater percentage of people agreeing with this statement present lower women employment rates, percentage of women responsible for supervising other employees, gender empowerment, and higher share of overeducated workers (women) and more permissive attitude toward violence. Therefore, the answer to this question seems to be related with other culture traits and, because of the strong correlation with gender empowerment (-0.82), it can be considered as a signal for empowerment. Our purpose is to identify if there exists a causal relation between education and woman s place in the family. In particular, to study if education explains people s opinion about the role woman should play in society. If this is the case, educational policies could be implemented to reduce social prejudice. Identifying if the correlation between education and others variables can be interpreted as a causal effect is difficult because of the potential endogeneity of education. Randomly assigning individuals different levels of education and seeing the differences on the variable of interest between individuals with different level of education some years later could be an ideal experiment to determine the causal effect of education in other variables. Obviously, this ideal experiment is not possible. Thus, researchers have tried to deal with the endogeneity of education by looking for a source of exogenous variation in schooling. Numerous studies have dealt with endogeneity of education by using schooling reforms to instrument education effects on different variables such as healths outcomes and 1 See D. Cuberes and M. Teignier, 2015 for further discussion on this point 1

3 wages, among others. Most of them find that changes in compulsory schooling laws are a strong instrument for education using different reforms for US (Lleras-Muney, A. (2005)), England, Scotland and Wales (Silles, M. (2009)), Czech Republic (Brunello et al. 2015), France (Albouy et al. (2009)), The Netherlands (Van Kippersluis et al. (2011)) and Austria, Belgium, Denmark, France, Germany, Greece, Hungary, Ireland, Poland, Spain, and the United Kingdom (Mocan and Pogolerova (2014)), among others. In this work, we use data from the European Social Survey Cumulative File Rounds 1-6 for 17 European countries to identify the effect of education on the attitude toward womens place in the family. To do that, we use changes in compulsory education laws to deal with the potential endogeneity of education. In particular, we use the number of years of compulsory education as an instrument for the number of completed years of full-time education. From the ESS data, we find that the percentage of people that think women have to be prepared to leave job market to sake of family is surprisingly high in a lot of countries 2. The Nordic countries are the countries with the lowest shares while the countries with the highest shares are the Eastern countries (See table 2). To give an example, in Lithuania, Ukraine and Turkey this share is higher than 70% whereas in other countries as Sweden and Denmark it is lower than 20% (See figure 1). Moreover, we observe that the percentage of people thinking women have to stay at home is higher among those with a low level of education and it decreases as the level of education increases indicating a negative correlation which is a behavior repeated across countries (See figure 3). Our results indicate that the number of years of compulsory education as instrument for education is valid and strong. The first stage results show that there exists a significant and positive effect of the number of compulsory schooling on the number of years of fulltime education. Moreover the F- test statistic is higher than 10, which indicates that the instrument is valid. We found that there exists a causal effect of education on people s opinion about whether women should be prepared to leave labor market for the sake of 2 The middle share of all countries contained in the ESS is about 45% 2

4 family. This paper brings out the existence of a causal relation between education and sexism perceptions which can affect economic outcomes. In particular, an increase of one year of full-time education decreases the probability of agreeing with the opinion that women should leave the labor market to take care of family in more than 4 percentage points, which is a stronger effect. Moreover, this effect is greater for men. Therefore, this heterogeneity may be taken into account in educational policies that aim to reduce social prejudice. This paper is organized as follows. In Section 2 we present the data used in this analysis, section 3 discuss the empirical model, section 4 contains results, section 5 includes some robustness checks and section 6 concludes. 2 Data and Descriptive Statistics This work is based on the European Social Survey Cumulative File Rounds (ESS thereafter). The ESS is a cross-national survey that has been conducted every two years across Europe since It contains individual information about attitudes, beliefs and behavioral patterns in 32 countries 4. We restrict the sample to 17 out of the 32 countries in the ESS for which we have available information about compulsory schooling reforms. In particular, we include Austria, Belgium, Czech Republic, Denmark, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Netherlands, Poland, Portugal, Slovakia, Spain and United Kingdom. Table 3 reports information about compulsory schooling reforms used in this study. As can be seen from this table, these reforms cover the period from 1956 (Germany) to 1983 (Belgium). Depending on the reform, the number of years of compulsory education increases by 1, 2 or 3 years. And, the first cohort affected by these reforms goes from 1941 to So, while these reforms affect some cohorts of children, others are exempted. Children that differed in few years were exposed to different years of compulsory education, which in some cases had an impact 3 European Social Survey Cumulative File, ESS 1-6 (2014). Data file edition 1.0. Norwegian Social Science Data Services, Norway - Data Archive and distributor of ESS dat. 4 See for further information 3

5 on their educational attainment. Therefore, we generate treatment and control groups using the first cohort affected by the reform and a window of 7 years. That is, control group will be composed by individuals born in country z in the time interval (y z 7, y z 1) and treatment group will be composed by individuals born in country z in the time interval (y z, y z + 6), where y z is the first cohort affected by the reform in country z. Thus, we have seven cohorts in each group. For example, In Austria, the 1966 reform changed the minimum school living age from 14 to 15 which increased years of compulsory education from 8 to 9. Hence, the first cohort affected by this reform is 1952 that is composed for those individuals that were 14 years old in It means that control group is composed for those individuals that born between 1945 and 1951 and the treatment group is composed for those born between 1952 and Figure 2 shows average years of education for those individuals born before and after the reform. That is, those in control and treatment group. As we can observe from this figure, there is an increase in the average years of schooling after the reform which seems to indicate that these reforms affect years of schooling. Table 4 presents average years of education for all the sample, treatment and control group by country. As it can be observed again, average years of education is higher for those in treatment group which are those affected by the reform. Our variable of interest measures beliefs toward women s place in the family. In particular, it is a dummy that takes value 1 if an individual answers Agree or Strongly Agree to the question: A woman should be prepared to cut down on her paid work for sake family?. This question is asked to both women and men and it is available for waves 2, 4 and 5. We draw a sample of individuals aged between 29 and 65 years old, non-immigrants and non-students 5. In table 5 we present summary statistics for the sample. It provides information about the number of observations, variables and average values for all sample distinguishing between before and after the reform (control and treatment group). In figure 3, we draw the percentage of people that agree or strongly agree with the 5 We consider non-student those who report being in school as their main activity in the last 7 days 4

6 following statement: a woman should be prepared to cut down on her paid work for sake family by level of education and country. We find that as the level education increases the percentage of people who agree or strongly agree decreases. Thus, we find a first evidence of the existence of correlation between these two variables. Nevertheless, we are interested in the causal relation. So, in order to obtain it, we use an instrumental variable approach as explained in the following section. 3 Empirical Model In order to study the causal impact of education on the attitude toward women s place in the family we adopt a two-equation model. The relation of interest between years of education and people s opinion toward women s place in the family is given by the secondstage equation: y i = β 0 + β 1 Educ i + β 2 X 1i + β 3 X 2i + v i (1) where y i denotes the outcome of interest for individual i, Educ i is the number of completed years of full-time education that individual i has attained 6, X 1i is a vector of controls that includes sex, ethnic minority status, religion, age, country, mother or father immigrants, an indicator of whether the respondent is a citizen and a dummy for whether the mother was working when respondent was 14. X 2i is a vector that contains other controls that can be determined after schooling and may affect our variable of interest. It contains indicators on whether the respondent is married or not, working, living with children, living in a city and the number of people living regularly as members of the household. Finally, v i is the error term. The problem that we face estimating this equation is that education is potentially endogenous because of unobservable individual characteristics and omitted variables related to 6 We use the number of years of education that the respondent declares correcting for some obvious mistakes in which the number of years of education and the level of education that the respondent declares does not match. 5

7 family background. As a result, estimating Equation (1) by ordinary least squares (OLS) will produce a biased estimator of the parameter of interest (β 1 ). To address this problem, we use a quasi-experimental identification strategy. We use changes in compulsory education laws in different countries as a source of exogenous variation in schooling. In particular, we use years of compulsory education as an instrument for reported years of full-time education. The impact of the number of years of compulsory education on years of education is given by the first-stage equation: Educ i = γ 0 + γ 1 Y earscomp zk + γ 2 X 1i + γ 3 X 2i + u i (2) where Y earscomp is the number of years of compulsory schooling in country z and cohort k and v i is the error term. We also study heterogeneous effect of education on our variable of interest by including an interaction between years of education and a dummy for gender. Therefore, in this case we estimate the following second-stage equation: y i = β 0 + β 1 Educ i + β 2 X 1i + β 3 X 2i + β 4 Educ i Gender + v i (3) where Gender is a dummy that takes value 1 for female and 0 for male, and v i is the error term. In this case, we have two potentially endogenous variables, education and the interaction between education and gender. Therefore, we need further instruments. Thus, we use as instruments years of compulsory education and the interaction between years of compulsory education and gender. Then, the corresponding first-stage regression in this case is estimated through the equation: Educ i = γ 0 + γ 1 Y earscomp zk + γ 2 X 1i + γ 3 X 2i + γ 4 Y earscomp zk Gender + u i (4) 6

8 4 Results The results from the first-stage equations (1) and (2) are illustrated in table 6. On the one hand, columns 1 and 3 are estimated using equation (1). The only difference between these two columns is that column 3 includes further controls. We find a statistically significant effect of the number of years of compulsory education on years of education. Moreover, the F- test is higher than 10, which indicates that years of compulsory education is a valid instrument for years of education. On the other hand, columns 2 and 4 report estimated coefficients from equation (2) finding similar results. Table 7 contains results from the second-stage equations (2) and (3). It is divided in two blocks. The first block provides the estimated coefficients from equation (1) whereas the second block the estimated coefficients from equation (3). In both cases, the first column presents the results from the Ordinary Least Square regression (OLS) using the whole sample, the second column shows estimated coefficients using Instrumental Variables (IV) and the third column estimates from OLS using the sample used in the estimation by instrumental variables. Results from instrumental variable estimation indicates that one more year of education produces an statistically significant decrease in the probability of agreeing about women leave job market to sake family by about 4.08 percentage points which is significant. We found that IV estimates are higher than OLS estimates. It could happen because returns for the subgroup who has been affected by the reform and have changed their behavior because of the reform (compliers) is higher than for the rest of people who were not affected. In the second block, we can see results from IV estimation which indicate that the effect of education in our variable of interest is higher for men although it becomes non significant when estimates by OLS. Finally, table 8 reports results from the second-stage equations (1) and (2) including the whole set of controls. As we can observe, results are in the same line. 7

9 5 Robustness Checks 5.1 Changing the Window To be sure that our results do not depend on the window that we have selected, we replicate our estimations by using different windows. In particular, we use a window of 5 years and a window of 3 years to check robustness of our results. We found that results does not change. These results are available upon request. 5.2 Measure of Years of Schooling The ESS provides two different measures of education: years of education and level of education. In our main estimation we use the years of education that the respondent declares correcting for some obvious mistakes in which number of years of education does not match with the level of education that the respondent says that he has. Therefore, to prove that our results are not affected for the measure of schooling, we repeat the analysis using the years of education without any correction. We find no change in our results. These results are available upon request. 5.3 Probit and IV Probit Model In this section, we estimate our model by using Probit and IV Probit. Results are presented in tables 9 and 10. Estimated coefficients indicates that one more year of schooling decreases the probability of being agree or strongly agree with the question survey a woman should be prepared to cut down on her paid work for sake of her family?. Moreover, we find again that this effect differs between women and men. 8

10 6 Conclusions This paper studies the causal effect of education on people s opinion about if woman has to leave labor market to take care of her family. This question is really interesting because this kind of opinion is correlated with some variables such as women employment rate, gender empowerment and domestic violence which can have a socio-economic impact. In order to study this question, we use an instrumental variable approach exploiting changes in compulsory schooling laws as a source of exogeneity for schooling. Results based on the European Social Survey and 17 countries indicate that one additional year of education reduces the probability of been agree with this kind of opinion in more than 4 percentage points and that this effect is even greater for men. Policy implications of these results are very important. The existence of heterogeneous effects makes it necessary to design different policies for women and men. Therefore, policymakers that aim to improve some measures of economic growth or reduce social prejudice should apply different policies according to the gender. 9

11 7 Figures and tables 10

12 Figure 1: Attitude toward womens place in the family Survey question: A woman should be prepared to cut down on her paid work for sake of her family? Source: European Social Survey Cumulative File Rounds 1-6 and own elaboration 11

13 Figure 2: Average Completed Years of Full-time Education Source: European Social Survey Cumulative File Rounds 1-6 and own elaboration 12

14 Figure 3: Share of Natives that answer Agree or Strongly Agree by level of education Source: European Social Survey Cumulative File Rounds 1-6 and own elaboration. Level of education defined as the higher level of education attained: 1, Less than lower secondary education (ISCED 0-1), 2, Lower secondary education completed (ISCED 2), 3, Upper secondary education completed (ISCED 3), 4, Post-secondary non-tertiary education completed (ISCED 4) and 5, Tertiary education completed (ISCED 5-6) 13

15 Table 1: Correlation coefficients between the percentage of people agree or strongly agree with the statement: women should be prepared to cut down on paid work for sake of family and other socio-economic indicators Percentage of women responsible for supervising other employees -0,4957 Women employment rate -0,6790 Share of overeducated workers (women) 0,3947 Attitude towards violence 0,3332 Gender empowerment measure -0,8157 Source: ESS, OCDE and Eurostat Gender Empowerment Measure (value). A composite index measuring gender inequality in three basic dimensions of empowermenteconomic participation and decision-making, political participation, and decision-making and power over economic resources. For details on how the index is calculated, see Technical note 1 of Human Development Report (2008) Attitude toward violence capture the percentage of women who agree that a husband/partner is justified in beating his wife/partner under certain circumstances. Overeducation refers to the proportion of workers whose educational attainment level is higher than the level required in their job (as measured based on the modal education level for all workers in the same occupation). 14

16 Table 2: Ranking of Natives Responding Agree or Strongly Agree by Country Rank Country Share percentage 1 Lithuania 81,06 2 Ukraine 72,15 3 Turkey 72,39 4 Cyprus 71,66 5 Russia 69,08 6 Italy 67,33 7 Luxembourg 60,6 8 Bosnia and Herzegovina 60,38 9 Portugal Poland 57,42 11 Estonia 49,86 12 Hungary 52,26 13 Czech Republic 51,95 14 Greece 51,66 15 Israel 51,93 16 Spain 50,9 17 Bulgaria 50,64 18 Croatia 49,08 19 Slovakia 47,77 20 Austria 46,97 21 France 44,14 22 Germany 42,36 23 United Kingdom 42,04 24 Slovenia 39,66 25 Ireland 38,47 26 Belgium 31,09 27 Netherlands 29,23 28 Iceland 27,26 29 Norway 20,93 30 Finland 21,14 31 Sweden 16,45 32 Denmark 14,74 Survey question: A woman should be prepared to cut down on her paid work for sake of her family? Source: European Social Survey Cumulative File Rounds 1-6 and own elaboration 15

17 Table 3: Reforms in compulsory education 16 Country School entry age Reform date Change on Years of Compulsory Education Change on Minimum School leaving age First cohort affected by the reform Source Austria to 9 14 to Mocan and Pogorelova (2014) Belgium to to Mocan and Pogorelova (2014) Czech Republic to 9 14 to Garrouste (2010) Denmark to 9 14 to Mocan and Pogorelova (2014) France to to Mocan and Pogorelova (2014) Germany Schleswig-Holstein to 9 14 to Mocan and Pogorelova (2014) Bremen to 9 14 to Mocan and Pogorelova (2014) Niedersachsen to 9 14 to Mocan and Pogorelova (2014) Saarland to 9 14 to Mocan and Pogorelova (2014) Nordrhein-Westfalia to 9 14 to Mocan and Pogorelova (2014) Hessen to 9 14 to Mocan and Pogorelova (2014) Rheinland-Pfalz to 9 14 to Mocan and Pogorelova (2014) Baden-Wurtenberg to 9 14 to Mocan and Pogorelova (2014) Bayern to 9 14 to Mocan and Pogorelova (2014) Greece to 9 12 to Murtin and Viarengo (2011) Garrouste (2010) Hungary to to Mocan and Pogorelova (2014) Iceland to 9 15 to Birgisdttir, K.H. (2013) Ireland to 9 14 to Mocan and Pogorelova (2014) Italy to 8 11 to Mocan and Pogorelova (2014) Murtin and Viarengo (2011) Netherlands to to Borgonovi et al (2010) Brunello et al (2009) Poland to 8 14 to Mocan and Pogorelova (2014) Portugal to 6 12 to Vieira (1999) Brunello (2013) Slovakia to 9 14 to Garrouste (2010) Spain to 8 12 to Mocan and Pogorelova (2014) United Kingdom* to to Mocan and Pogorelova (2014) * According to Mocan and Pogolerova (2014), the first cohort affected by the reform is 1957 in North Ireland, England and Wales and 1958 in Scotland. The first cohort affected by the reform considered in this study are those born in 1958.

18 Table 4: Average years of education Reform Obs All Control Treatment Change (%) Austria, Belgium, Czech Republic, Denmark, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Netherlands, Poland, Portugal, Slovakia, Spain, United kingdom, Note: The Source is the European Social Survey Cumulative File Rounds 1-6. Treatment group consist of individuals born up to six years after the first cohort potentially affected by an education reform including those born in the first cohort potentially affected. Control group consist of individuals born up to seven years before the first cohort potentially affected by an education reform. 17

19 Table 5: Summary statistics 18 Variable Variable Definitions Obs Women home ==1 if respondent answer agree or strongly agree to the question A woman should be prepared to cut down on her paid work for sake of her family? All (Mean) Control (Mean) Treatment (Mean) Years of education Number of years of full-time education completed Age Age of the respondent Female ==1 if female, 0 otherwise Mother working when respondent ==1 if respondent s mother was employee was 14 or self-employee when respondent was Belong to ethnic minority ==1 if belongs to ethnic minority group Mother immigrant ==1 if mother is an immigrant, 0 otherwise Father immigrant ==1 if father is an immigrant, 0 otherwise Religion: Islamic ==1 if respondent is islamic, 0 otherwise Religion: Christian ==1 if respondent is christian, 0 otherwise Religion: Others ==1 if respondent has other religion, 0 otherwise To be very religious ==1 if respondent declares to be very religious No citizen ==1 if respondent is not citizen, 0 otherwise Big city ==1 if respondent lives in a big city, 0 otherwise Number of household Number of household members members Work ==1 if working in the last 7 days, 0 otherwise Children home ==1 if a child lives at home, 0 otherwise Married ==1 if married, 0 otherwise Affected by reform (7 year window) ==1 if the respondent is affected by the reform Years of compulsory education Number of compulsory years of education Note: The Source is the European Social Survey Cumulative File Rounds 1-6. Treatment group consist of individuals born up to six years after the first cohort potentially affected by an education reform including those born in the first cohort potentially affected. Control group consist of individuals born up to seven years before the first cohort potentially affected by an education reform.

20 Table 6: First stage results Dependent Variable: Years of education Independent Variables (1) (2) (3) (4) Compulsory education 1.014*** 0.907*** 1.001*** 0.913*** (0.0609) (0.0740) (0.0618) (0.0750) Compulsory education*female 0.178*** 0.148*** (0.0428) (0.0429) Female *** *** *** (0.0712) (0.381) (0.0704) (0.381) Mother working when respondent was *** 0.219*** 0.169** 0.173** (0.0746) (0.0745) (0.0713) (0.0712) Non citizen ** ** *** *** (0.451) (0.447) (0.451) (0.448) Father immigrant (0.185) (0.184) (0.182) (0.182) Mother immigrant 0.457*** 0.451*** 0.442** 0.437** (0.171) (0.171) (0.172) (0.171) Belong to ethnic minority (0.262) (0.260) (0.253) (0.252) Religion: Islamic * * (0.661) (0.659) (0.681) (0.676) Religion: Christian *** *** *** *** (0.0929) (0.0926) (0.0854) (0.0851) Religion: Other religion 1.220** 1.199** 1.033* 1.014* (0.573) (0.573) (0.577) (0.577) To be very religious ** * (0.0910) (0.0907) (0.0880) (0.0878) Big city 1.154*** 1.156*** (0.0964) (0.0962) Number of people living regularly as member of household (0.0396) (0.0397) Work 1.652*** 1.642*** (0.0835) (0.0826) Children home (0.0961) (0.0961) Married (0.0759) (0.0759) Observations 16,118 16,118 16,035 16,035 R-squared F-test statistic P-value Note: All regressions control for country-specific time trend and round. Robust standard errors are clustered by country and birth-cohort. The F-test refers to the joint significance of the instrument(s) in each case. ***p<0.01, **p<0.05, *p<0.1 19

21 Table 7: Effect of education on attitude toward women place in the family (1) (2) Independent Variables: OLS IV OLS(a) OLS IV OLS(a) 20 Years of education *** *** *** *** *** *** ( ) ( ) ( ) ( ) ( ) ( ) Years of education*female *** ( ) ( ) ( ) Female *** *** ** ** *** ( ) ( ) ( ) (0.0149) (0.101) (0.0266) Mother working when respondent was *** ** *** ** ( ) ( ) ( ) ( ) ( ) ( ) Non citizen (0.0393) (0.0699) (0.0682) (0.0393) (0.0707) (0.0682) Father immigrant (0.0117) (0.0220) (0.0216) (0.0117) (0.0224) (0.0216) Mother immigrant (0.0125) (0.0252) (0.0241) (0.0125) (0.0262) (0.0241) Belong to ethnic minority *** *** (0.0155) (0.0293) (0.0291) (0.0155) (0.0294) (0.0291) Religion: Islamic 0.174*** 0.163** 0.201*** 0.173*** 0.150** 0.201*** (0.0411) (0.0680) (0.0705) (0.0410) (0.0669) (0.0705) Religion: Christian *** *** *** *** *** *** ( ) (0.0111) (0.0110) ( ) (0.0112) (0.0110) Religion: Other religion (0.0275) (0.0484) (0.0434) (0.0275) (0.0485) (0.0434) To be very religious *** *** *** *** *** *** ( ) (0.0104) (0.0102) ( ) (0.0106) (0.0102) Observations 49,665 16,118 16,118 49,665 16,118 16,118 R-squared Note: Dependent variable is a dummy that takes value 1 in case respondent answer agree or strongly agree to the survey question A woman should be prepared to cut down on her paid work for sake of her family?. Regressions in columns 1 and 4 uses all sample. All regressions control for country-specific time trend and round. Robust standard errors are clustered by country and birth-cohort. ***p<0.01, **p<0.05, *p<0.1

22 Table 8: Effect of education on attitude toward women place in the family (1) (2) Independent Variables IV OLS IV OLS Years of education *** *** *** *** ( ) ( ) ( ) ( ) Years of education*female *** ( ) ( ) Female *** *** *** * ( ) ( ) (0.0988) (0.0270) Mother working when respondent was *** ** *** ** ( ) ( ) ( ) ( ) Non citizen (0.0690) (0.0674) (0.0699) (0.0673) Father immigrant (0.0218) (0.0215) (0.0222) (0.0215) Mother immigrant (0.0252) (0.0241) (0.0262) (0.0241) Belong to ethnic minority (0.0291) (0.0288) (0.0292) (0.0287) Religion: Islamic 0.155** 0.190** 0.142** 0.190** (0.0707) (0.0738) (0.0697) (0.0737) Religion: Christian *** *** *** *** (0.0109) (0.0108) (0.0110) (0.0108) Religion: Other religion (0.0485) (0.0435) (0.0486) (0.0435) To be very religious *** *** *** *** (0.0106) (0.0104) (0.0108) (0.0104) Big city ** ** (0.0109) ( ) (0.0114) ( ) Number of people living regularly as member of household ** ** * ** ( ) ( ) ( ) ( ) Work * *** ** *** (0.0155) (0.0101) (0.0155) (0.0102) Children home (0.0124) (0.0121) (0.0127) (0.0120) Married *** *** *** *** ( ) ( ) ( ) ( ) Observations 16,035 16,035 16,035 16,035 R-squared Note: Dependent variable is a dummy that takes value 1 in case respondent answer agree or strongly agree to the survey question A woman should be prepared to cut down on her paid work for sake of her family?. Regressions in columns 1 and 4 uses all sample. All regressions control for country-specific time trend and round. Robust standard errors are clustered by country and birth-cohort. ***p<0.01, **p<0.05, *p<0.1 21

23 Table 9: Effect of education on attitude toward women place in the family (1) (2) Independent Variables PROBIT IVPROBIT PROBIT (a) PROBIT IVPROBIT PROBIT (a) 22 Years of education *** *** *** *** *** *** ( ) (0.0146) ( ) ( ) (0.0203) ( ) Years of education*female *** ( ) (0.0192) ( ) Female *** *** ** *** *** (0.0134) (0.0254) (0.0247) (0.0421) (0.248) (0.0751) Mother working when respondent was *** ** *** ** (0.0144) (0.0258) (0.0254) (0.0144) (0.0253) (0.0254) Non citizen (0.108) (0.189) (0.190) (0.108) (0.186) (0.190) Father immigrant (0.0327) (0.0593) (0.0604) (0.0326) (0.0588) (0.0604) Mother immigrant (0.0344) (0.0684) (0.0676) (0.0344) (0.0691) (0.0676) Belong to ethnic minority 0.105** ** (0.0414) (0.0763) (0.0781) (0.0414) (0.0749) (0.0780) Religion: Islamic 0.466*** 0.432** 0.547*** 0.463*** 0.386** 0.546*** (0.109) (0.178) (0.189) (0.109) (0.171) (0.189) Religion: Christian 0.262*** 0.199*** 0.261*** 0.262*** 0.187*** 0.261*** (0.0164) (0.0312) (0.0309) (0.0165) (0.0311) (0.0309) Religion: Other religion (0.0821) (0.139) (0.133) (0.0821) (0.136) (0.133) To be very religious 0.183*** 0.174*** 0.183*** 0.183*** 0.180*** 0.183*** (0.0161) (0.0270) (0.0273) (0.0161) (0.0271) (0.0272) Observations 49,665 16,118 16,118 49,665 16,118 16,118 Note: All regressions control for country-specific time trend and round. Robust standard errors are clustered by country and birth-cohort. The F-test refers to the joint significance of the instrument(s) in each case. ***p<0.01, **p<0.05, *p<0.1

24 Table 10: Effect of education on attitude toward women place in the family (1) (2) Independent Variables PROBIT IVPROBIT PROBIT IVPROBIT Years of education *** *** *** *** (0.0146) ( ) (0.0200) ( ) Years of education*female *** (0.0192) ( ) Female *** *** *** * (0.0265) (0.0261) (0.247) (0.0764) Mother working when respondent was *** ** *** ** (0.0256) (0.0254) (0.0252) (0.0254) Non citizen (0.187) (0.188) (0.184) (0.188) Father immigrant (0.0591) (0.0603) (0.0587) (0.0604) Mother immigrant (0.0687) (0.0682) (0.0695) (0.0682) Belong to ethnic minority (0.0765) (0.0779) (0.0750) (0.0778) Religion: Islamic 0.403** 0.510*** 0.359** 0.510*** (0.184) (0.197) (0.177) (0.196) Religion: Christian 0.189*** 0.247*** 0.176*** 0.247*** (0.0309) (0.0308) (0.0307) (0.0308) Religion: Other religion (0.141) (0.135) (0.138) (0.135) To be very religious 0.176*** 0.178*** 0.180*** 0.179*** (0.0276) (0.0278) (0.0276) (0.0278) Big city ** ** (0.0289) (0.0267) (0.0291) (0.0267) Number of people living regularly as member of household * ** * ** (0.0135) (0.0135) (0.0135) (0.0135) Work * *** ** *** (0.0416) (0.0272) (0.0409) (0.0274) Children home (0.0333) (0.0335) (0.0332) (0.0334) Married *** *** *** *** (0.0267) (0.0264) (0.0266) (0.0264) Observations 16,035 16,035 16,035 16,035 Note: All regressions control for country-specific time trend and round. Robust standard errors are clustered by country and birth-cohort. The F-test refers to the joint significance of the instrument(s) in each case. ***p<0.01, **p<0.05, *p<0.1 23

25 References [1] Lleras-Muney A. The relationship between education and adult mortality in the united states. Review of Economic Studies, 72: , [2] V. Albouy and L. Lequien. Does compulsory education lower mortality? Journal of Health Economics, 28(1): , [3] Garrouste C. 100 years of educational reforms in europe: A contextual database. European Commission Joint Research Center, Luxembourg: Publications, Office of the European Union., [4] Brunello G. Fabbri D. and Fort M. The causal effect of education on body mass: Evidence from europe. Journal of Labor Economics, 31(1): , [5] Borgonovi F. d Hombres B. and Hoskins B. Voter turnout, information acquisition and education: Evidence from 15 european countries. The B.E. Journal of Economic Analysis and Policy, 10(1 (Contributions)):Article 90, [6] Murtin Fabrice and Viarengo Martina. The expansion and convergence of compulsory schooling in western europe: Economica, 78(no. 311): , [7] Birgisdttir K. H. Education and health: Effects of school reforms on birth outcomes in icelan [8] Arendt J. Does education cause better health? a panel data analysis using school reform for identification. Economics of Education Review, 24: , [9] Vieira J. Returns to education in portugal. Journal of Health Economics, 6(4): ,

26 [10] Brunello G. Fort M. and Weber G. Changes in compulsory schooling, education and the distribution of wages in europe. The Economic Journal, 119: , [11] Cuberes D. & Teignier M. Aggregate effects of gender gaps in the labor market: A quantitative estimate. Journal of Human Capital, 10(1):1 32, [12] Silles M. The causal effect of education on health: Evidence from the united kingdom. Economics of Education Review, 28: , [13] Mocan N. and Pogorelova L. Compulsory schooling laws and formation of beliefs: Education, religion and superstition. National Bureau of Economic Research., (No. w20557), [14] Van Kippersluis H. ODonnell O. and Van Doorslaer E. Long run returns to education: Does schooling lead to an extended old age? Journal of Human Resources, 46(4): , [15] Brunello G. Fort M. Scheneeweiss and Winter-Ebmer R. The causal effect of education on health: what is the role of health behaviours? Health Economics, forthcoming. [16] European Social Survey. Ess 1-6, european social survey cumulative file, study description. Bergen: Norwegian Social Science Data Services.,