Education Capability: A Focus on Gender and Science

The focus of the paper is on the measurement of science education capability with a gender perspective and in the capability approach framework. Measuring science education capability implies going beyond the measurement of children test scores. In the capability approach, we aim at the real opportunities that children can develop later in life and therefore it is important to include some measures of non-cognitive skills. We utilize, therefore, different indicators in addition to test scores in science: enjoyment in science, interest in science, general and personal values of science, self-confidence in performing science related tasks, awareness and perception of environmental issues, and responsibility for sustainable development. We utilize the 2006 PISA survey for Italian 15 years old children because it contains a particular focus on science and we estimate a Structural Equation Model to take into account that capabilities are latent constructs of which we only observe some indicators. We also investigate the determinants of children’s science education capability in Italy taking into account household, individual and school factors. Results confirm that boys outperform girls in science education capability. Our theoretical construct for the science education capability confirms that all the indicators are relevant to measure this capability. School activities to promote sciences improve girls’ capability and interactive methods of teaching improve both girls and boys capability. The household educational resources and the household educational possession are also positively correlated with girls’ and boys’ science education capability.

In this paper, we focus on the science education capability for Italian children in a gender perspective. Following the capability literature on education and the educational literature on cognitive and non-cognitive skills (Heckman 2008;Cunha et al. 2010;Sikora and Pokropek 2012;Cornwell et al. 2013;Gutman and Schoon 2013;Heckman and Mosso 2014), we believe that the use of test scores is limited. In the spirit of the capability approach, we would like to focus on the real opportunities that children have to become knowledgeable (educated) adults and therefore include some measures of non-cognitive skills. We utilize, therefore, different indicators in addition to test scores in science: enjoyment in science, interest in science, general and personal values of science, selfefficacy (confidence in performing science related tasks), awareness and perception of environmental issues, and responsibility for sustainable development.
We apply the simplest Structural Equation Model, a Multiple Indicators Multiple Causes model (MIMIC), to OECD PISA 2006 micro-data for Italy. The MIMIC model has two important features: first, it allows the estimation of the science capability as a latent construct of which it is possible to observe only some functionings i.e. the indicators listed above (Krishnakumar and Nagar 2008). Second, it allows the presence of exogeneous cause variables that determine the latent capability.
The Italian PISA data provide a relevant example of a country with low achievements on science test scores and high level of gender inequality in the society. OECD (2007) shows that the Italian average test score in science is equal to 475 (477 for males and 474 for females), against an OECD average equal to 500. Further, according to EIGE (2013), the gender equality index in Italy is 40.9 against an EU-28 average by 54.
The major contribution of the paper is to provide a new concept of science education capability that is defined not only on test scores but on a broader set of indicators. This definition of the education capability is particularly relevant, as our results show, for analysing gender differences. Moreover, we utilize a MIMIC model which allows both to consider capabilities as latent variables of which we observe some indicators and to estimate the effect of individual, family, and institutional variables on the latent capability.
The paper is organized as follows: Sect. 2 presents the existing literature focusing on the gender gap in science, and on the family and institutional conversion factors referred to this capability. Data and descriptive statistics are provided in Sect. 3. The MIMIC model is described in Sect. 4 and the results of the estimated model are shown in Sect. 5. The last Section draws conclusions. especially from mothers to daughters, as the performance of girls-not that of boys-is better in families where the mother works outside home. The gender gap in math decreases in more gender equal societies (Guiso et al. 2008). Also OECD (2015) confirms a positive correlation between women's participation rates and girls' performance in math.
The literature shows that the gender gap in sciences does not emerge until high school. This result suggests that gender differences are due to socialization and to the educational process rather than connected to biological factors (Good et al. 2010;Bleeker and Jacobs 2004;Brownlow and Durham 1997). Sikora and Pokropek (2012) using data from PISA 2006 surveys for 50 countries analyse gender differences in science and they find that the male-female gap in science self-concept (similar to our self-efficacy in Table 1) is larger in advanced industrial countries.

Family Involvement
The role of early parenting and older siblings on children's cognitive and non-cognitive abilities at different stages of children's life has been brought into attention by the literature (amongst others Heckman 2008; Cunha et al. 2010;Dai and Heckman 2013;Heckman and Mosso 2014;Del Boca et al. 2014). Haveman and Wolfe (1995), Peraita and Sánchez (1998) and Yeung et al. (2002), stress the impact on children's attainment of family background variables. Duncan et al. (1994) by using the Infant Health and Development Program longitudinal data detect a negative effect of poverty on child cognitive and behavioural functionings at age 5 and a positive effect of mother's education. By disentangling Bourdieu's cultural capital and habitus concepts, through a critical survey of the literature on their effects and by considering the joint impact of both in the reproduction of educational inequality, Edgerton and Roberts (2014) stress their relevant joint explanatory potential. Sullivan (2001) operationalizes the concept of cultural capital and provides measures of both parents' and children's cultural capital finding a positive effect of cultural capital on children's performance in the results of the general certificate of secondary education though the inclusion of cultural capital effects does not cancel out the significance of the effect of social class on children's educational attainment.
With specific reference to sciences, parents' perception of child ability, parental involvement in homework, sex-stereotypes in parent's evaluation of children abilities have been found to affect achievements in science and children's self-perception (amongst others: Jacobs 1991; Jacobs and Bleeker 2004;Jacobs and Eccles 1992;Bhanot and Jovanovic 2009;Twenge and Campbell 2001). Though gender gaps are decreasing over time, boys have better access to science-related resources than girls .
Mothers' encouragement in science homework has a positive effect on girls' selfassessment of science ability, and a negative effect on boys' self-assessment. Mothers' science discussions have a similar effect on boys and girls beliefs about science (Bhanot and Jovanovic 2009). Moreover, the effect of parental involvement on children's out-ofschool activities related to sciences and math are related to children's interest in science also later on in their life .

Teaching Science, Pre-schooling and Schooling
There is an extensive literature showing the impact of schooling and preschool education programs on the development of cognitive and non-cognitive skills. In particular in the last decade, economic research has been increasingly focusing on non-cognitive skills. Heckman andKautz (2012, 2014) suggest that not only cognitive skills matter and that non cognitive skills, such as ''personality traits'', ''motivation'', ''sociability'', ''self-esteem'', are important predictors of success in life and in the labour market. The authors call for a better understanding of non-cognitive skills and the promotion of policies enhancing their development. Moreover, Heckman and Mosso (2014) show that also test results depend not only on cognitive skills, but also on non-cognitive skills, such as effort and motivation. This stresses the importance of formative experiences in preschool education programs and in family upbringing focusing not merely on cognitive aspects.
High quality preschool education programs like the Child-Parent Center Program 1 or the High Scope Perry Preschool Program 2 have been shown to positively affect later life outcomes in terms of educational attainment, cognitive and non-cognitive skills, labour market status, earnings, and reduced crime (Temple and Reynolds 2007;Heckman et al. 2010a, b;Heckman et al. 2013).
The effect of schooling in affecting test scores has been widely analysed in the literature. Hansen et al. (2004) have modelled the interaction of schooling and test scores as generated by a common unobserved latent ability detecting different effects of schooling at different ability levels. Martins and Veiga (2010) detect a significant role of socioeconomic inequalities in explaining children's mathematical achievement in EU-15 countries by using the OECD Programme for International Student Assessment (PISA) 2003 data; according to their results Italy belongs to those countries (amongst them also The Netherlands, Germany, Belgium and Austria) where school seems to play an important role in the observed difference in children's achievements in mathematics.
Focussing on the gender differences in STEM and on different teaching activities, evidence shows how different approaches to math and physics can increase gender inequality in achievements. Problem-solving and cognitive activation strategies teaching methods yield positive results on STEM test scores (Aka et al. 2010;OECD 2015). Problem-solving, class-discussions and investigative work and cognitive activation strategies have been found to improve girls' performances (Boaler 2002;Zohar and Sela 2003;OECD 2015). Boaler et al. (2011) analyse other factors affecting lower achievement in math and science, e.g. images shown in textbooks. Good et al. (2010) show that, by using counter-stereotypic images with female scientists, girls' comprehension increases.
Moreover, children's overall self-perception of abilities is more affected by negative evaluation by others during teenage-hood (Bhanot and Jovanovic 2009;Twenge and Campbell 2001).

Data and Descriptive Statistics
We utilize the 2006 survey of the Programme for International Student Assessment (PISA) conducted by the OECD. Our sample consists of 8582 boys and 8369 girls. PISA tests are collected on 15 years old children.
The 2006 PISA survey contains a particular focus on sciences, which is useful to extend the estimation of the education capability in science beyond test results. In particular, in Listen to radio programmes about advances in broad science Read broad science magazines or science articles in newspapers addition to the test scores, we use data on interest in science, enjoyment in science, science self-efficacy (confidence in performing science related tasks), general and personal values of science, science activities, perception of environmental issues, responsibility for sustainable development and awareness of environmental issues as indicators of the cognitive capability in science. Each of these additional dimensions is a synthetic index of a set of items and is provided by OECD. 3 Table 1 shows the lists of indicator for each index. PISA provides normalized indicators: test scores' mean is equal to 500 and their standard deviation equal to 100 at the whole OECD level, while Table 1 indices have zero mean and standard deviation equal to one. As our primary purpose is not to carry out an international comparison, but to compare Italian children by gender, in presenting these statistics, we re-normalize all these indices within our sample (i.e. the variables have been Table 1 continued

Perception of environmental issues
Student's perception about the serious concern for him/her or others of the following problems Air pollution Consequences of clearing forests 3 OECD inverts the items if appropriate, such that the higher the index, the better the performance of the child in a particular dimension. Each index is computed by OECD by running a Confirmatory Factor Analysis using Weighted Likelihood Estimator on the corresponding set of items. The (latent) index scale is then set with mean equal to zero and standard deviation equal to 1 at the OECD level. A positive or a negative index score of a child or of a group of children has therefore no meaning per se: it is interpretable only in comparison with the scores of another peer or of a group of peers (see OECD 2009 for details).
Education Capability: A Focus on Gender and Science 799 standardized to have zero mean and variance equal to one in the whole sample of boys and girls). Table 2 shows boys' and girls' achievements in terms of the indicators of the latent science education capability. Table 2 also shows the gender gap (male-female mean) and the result of the mean-comparison test. As expected, we find a significant gender gap in the test scores in favour of boys equal to 0.13 standard deviations: boys' scores are 0.06 standard deviations above the overall mean and girls' scores are 0.06 below the overall mean. The gender gaps are statistically significant for all indicators, but for interest in science. Gender gaps are higher for performing science activities (?0.22) and for the general and personal value of science (?0.22 and ?0.17 respectively). Boys' higher perception of personal value of science can also be correlated to different expectations in terms of future career. The percentage of male students in STEM's subjects in tertiary education is higher than that of girls (OECD 2012). In addition, STEM's related jobs are mainly males (OECD 2006).
Boys show also higher level of enjoyment in science (?0.10). Boys are better off than girls also in self-efficacy in science (?0.17), suggesting that boys are more confident (see Bhanot and Jovanovic 2009 for a survey). Boys show better achievements also in term of awareness of environmental problems (?0.17).
On the other hand, girls outperform boys in the perception of the gravity of environmental issues (by 0.17) and in the responsibility for carrying out activities towards sustainable development (by 0.08). Table 3 shows the description of the above-mentioned indicators and of exogenous individual, household and school variables.
At individual level, we consider gender and immigration status, also interacted. At the household level, we consider the following variables: an index on the household's possession of cultural goods; an index of educational resources in the household; mother and father education; mother and father occupational status, also in terms of socioeconomic occupational status (see Ganzeboom et al. 1992).
School's variables include hours of science at school, a variable for participative teaching methods, a factor describing the development of activities for the promotion of science at school, the shortage of science teachers. Table 4 shows descriptive statistics on the conversion factors described above. We have two types of variables: binary variables and continuous variables provided and scaled by PISA in a way similar to the one explained for the functioning indicators. Also in this table, we re-normalize continuous variables within our sample (i.e. the variables have been standardized to have zero mean and variance equal to one in the whole sample of boys and girls). Table 4 shows that about 3.5 % of children in our sample are immigrant. Mothers and fathers show a very similar educational level but they are very different in terms of occupation: one-third of the mothers do not work in the labour market, this is against about 2 % of the fathers; fathers are also better off in terms of socio-economic occupational status. As for the school factors, boys are more likely than girls to enjoy interactive teaching methods, science promotion activities, more hours of science at school (about 30 % of boys have more than 4 h weekly, against 26 % of girls). 4

The Multiple Indicators Multiple Causes Model
We estimate the education capability utilising a Multiple Indicators Multiple Causes model (MIMIC). Multiple indicators models (Muthén 1979) link several observed outcomes to a reduced number of underlying latent variables (in this case, a single ''science education capability''). Moreover, MIMIC models also estimate the coefficients of the exogenous variables that are considered to be determinants of the latent variable.
Our justification for choosing the MIMIC specification is that the latent variable estimators delivered by this model represent the fundamental objects of interest (see Krishnakumar and Nagar 2008). We construct a system of equations which specify the relationship between an unobservable latent variable Y Ã , a set of observable endogenous ordinal indicators Y, and a set of observable exogenous variables X (causes e.g. individual, school and household variables).
The structure of the model is as follows: . . .k Y m g 0 denotes a m Â 1 parameter vector of factor loadings, with each element representing the expected change in the respective indicators following a one unit change in the latent variable. e is a m Â 1 vector of measurement errors, with H e denote the covariance matrix.
In addition, we posit that the latent variable Y Ã is linearly determined by a vector of observable exogenous variables x ¼ x 1 ; x 2 ; . . .; x s ð Þ 0 and a stochastic error 1 giving, where c is a s Â 1 vector of parameters.   (1) and (2) we may think of our model as comprised of two parts: (2) is the structural (or state) equation and (1) is the measurement equation reflecting that the observed measurements are partial indicators. The structural equation specifies the casual relationship between the observed exogenous causes and the latent variable. Since Y Ã is unobserved, it is not possible to recover direct estimates of the structural parameters c.
Combining (1) and (2) the reduced form representation is written as where p ¼ K Y c 0 is the m Â s reduced form coefficient matrix and v ¼ K Y 1 þ e is the reduced form disturbance. SEM have been widely used in the capability literature (see, among others, the work by Kuklys 2005;Di Tommaso 2007;Krishnakumar 2007;Krishnakumar and Ballon 2008;Anand et al. 2011).
In this paper, the latent variable is the ''science education capability'', the indicators are the observable functionings of Table 2 above while the exogenous variables are the set of conversion factors listed in Tables 3 and 4.
The model can be schematized as in Fig. 1.

Estimation Results
In this Section, we report the results of the estimation of the MIMIC model presented above. In the estimation results, we show both the standardized and unstandardized solutions. Both are meaningful. The unstandardized solution is achieved by setting a lambda parameter equal to 1 and it also reports the standard errors and significance level of the variable coefficient. The disadvantage of unstandardized solutions is that they are not easily interpretable, as they refer to changes in variables that have no clear and homogeneous measurement unit. The standardized solution overcomes this problem. Standardization is achieved by setting the variance of the latent variable equal to 1, therefore  Fig. 1 Path diagram of the MIMIC model standardized coefficients can be read as the standard deviation change in the dependent variable that follows one standard deviation change in the independent variable. Table 5 reports the results of the measurement equation.
All the included functionings are relevant and positively related to the development of the science capability. 5 This is true both for the overall model (first column) and for the model estimated separately by gender (second and third column). Moreover, with very little differences by gender, the standardized coefficients are the highest for enjoyment of science, personal value of science and interest in science, suggesting that these variables are the most sensitive to changes in the science capability. Test scores, on the contrary, show the second lowest standardized coefficient. This result supports our choice of introducing other indicators of the latent science education capability and is consistent with the literature that highlights the limits related to the use of test scores (Cunha et al. 2010;Sikora and Pokropek 2012;Cornwell et al. 2013;Gutman and Schoon 2013).
Looking at the structural model, the results (Table 6) show that the science capability is more developed for boys: looking at column one, the coefficient for the female dummy is   negative and statistically significant. This result leads us to estimate the same model on the two subsamples of boys and girls in order to analyse the differences in the effect of the conversion factors (second and third column).
Looking at the standardized coefficients, the most relevant variables are number of hours dedicated to science, interactive methods of teaching, household's cultural possession and educational resources.
The number of hours dedicated to science and interactive methods of teaching are positively correlated with both girls' and boys' science capability. Promotion activities in science have a lower positive parameter and it is significant for girls only. Being in a school characterized by a science teachers shortage negatively affects only boys' science capability. This can be determined by a higher need of teacher's supervision shown by boys in their education process.
Cultural possession at home and educational resources have a positive and significant effect on the development of science capability for both girls and boys, and they are both higher for boys. The relevance of family cultural possessions is in line with the literature showing the relevance of cultural capital and habitus (Bourdieu 1973;Edgerton and Roberts 2014) and of a stimulating learning environment at home (De Graaf et al. 2000). Being a migrant does not significantly affect science capability. This is not consistent with other results where racial disparities continue to increase throughout high school (Bacharach et al. 2003). It could be in part related to the inclusion in the model of family background variables-to the extent that immigrant households are characterized by a lower attainment in terms of socioeconomic background- (Azzolini et al. 2012;Fullin and Reyneri 2011;Shapira 2012) but even more to the inclusion of school characteristics as long as immigrant children are not randomly distributed across different types of schools (Azzolini et al. 2012;Shapira 2012).
Parents' level of education plays a higher positive effect on girls'. In particular, the parameters for mothers' educational levels are only statistically significant for girls while fathers' education affects both girls and boys but with a higher effect on girls. These results are consistent with Kleinjans' (2010) analysis on the effect of parental education and income on the educational expectations of children at age 19 in Denmark. For the USA, Bhanot and Jovanovic (2009) find a positive effect of mothers' encouragement only on girls' self-assessments of science ability at the end of the year. The relevance of parents' education on children's educational attainment is also in line with other literature results Mothers' employment status below the mean or jobless status negatively affects only boys' science capability whereas fathers' employment status below average negatively affects only girls' achievements in science. This, given the higher father's household income share, can be related to a positive correlation between household's income and girls' achievement in science. Evidence on the positive impact of household's income on children's achievement in science have been found amongst others by Beaumont-Walters and Soyibo (2001). Other studies (in particular, for developing countries) have also found that the elasticity of education demand to changes in parents' income is higher for girls than for boys (Mason and King 2001;Tansel 2002).
Differently from what has been found in other contexts, for instance in UK by Boaler et al. (2011), interactive methods of teaching do not play a positive role only for girls: in our estimates, they affect both boys' and girls' achievement in science. This result, though specific to the science education capability, is in line with the literature on the positive effect of participatory methods to the development of education capabilities (Hart 2014;Biggeri 2014).
On the other hand, promotion of science activities positively affects only girls' achievements in science. To reduce the gender gap in science, these types of activities should be promoted.
Finally, the fit of the model is satisfying as shown by the Standardized Root Mean Square Residual (SRMR) below 0.08.

Conclusions
This paper analyses gender differences in science education capability in Italy, a country characterized by a lower than OECD-average achievement of test scores in science (according to PISA 2006 survey) and by a gender gap to the disadvantage of girls. We utilize a Structural Equation Model, a technique increasingly used in the measurement of capabilities. We measure the education capability as a latent variable of which we observe ten indicators. All the chosen indicators significantly contribute to the latent science capability, showing that it is relevant not to limit the analysis to test scores.
The structural part of the estimated model reveals different impact of institutional and family variables by gender. In particular, we find that activities to promote sciences have a greater effect on girls' capability. Therefore, policies oriented to improve these activities could reduce the gender gap. Among the policies that can reduce the gender disparities in STEM subjects, OECD (2012) suggests addressing gender stereotypes in textbooks and developing teaching strategies and learning materials to encourage girls' involvement. Interactive methods of schooling and hours of science at school have a positive effect both on girls and on boys and, together with cultural and educational resources at household's level, have the highest impact on boys and girls science capability. Given Italian children low achievements in science (respect to OECD average), these results call for higher investment in education and cultural family possessions and for the adoption of more interactive and less traditional teaching methods.