Culture, migration and educational performance: a focus on gender outcomes using Australian PISA tests

Culture, migration and educational performance: a focus on gender outcomes using Australian PISA tests A.M. Dockery, Bankwest Curtin Economics Centre, Curtin University Ian Li, School of Population and Global Health, UWA Paul Koshy, National Centre for Student Equity in higher Education, Curtin University Abstract Cultural background has been shown to be a significant factor in shaping gender differences in labour market outcomes. This paper explores whether the seeds of such cultural effects are sown in differences in the academic performance of boys and girls in high-school. Scores from the 2015 Programme for International Student Assessment tests are analysed of for Australian children from migrant and non-migrant families, conditional upon a measure of gender equity in secondary level education in their country of ancestry. Sizable and statistically significant effects are identified that show children who come from a cultural background that affords lower emphasis on the schooling of children of their gender in turn achieve lower scores on the PISA reading, mathematics and science tests. This result holds when first-generation migrant children are excluded from the sample (ie. the sample is restricted to children born in Australia), providing strong evidence that the effect is transmitted through cultural effects. A surprising finding is that such effects are markedly more pronounced for boys, rather than girls. We also find that such effects of cultural background are best modelled on the basis of the country of birth of the parent of the same gender as the student, suggesting such cultural effects are transmitted through that parent, as a role model, shaping students preferences and aspirations. Key words: migrants, culture, academic performance, PISA (Programme for International Student Assessment), gender. Corresponding author: Mike Dockery, m.dockery@curtin.edu.au

1. Introduction The introduction of standardised global testing such as the Programme for International Student Assessment (PISA) in 2000 has seen an increased focus on educational performance, both within and between countries, in the context of known socioeconomic contributors to outcomes: students cultural background; parental educational and occupational background; household and national income levels; and school and education systems types. PISA, a triennial test of the skills and knowledge of 15 year old students in mathematics, reading and science, has provided insights into how these factors contribute to outcomes around gender, and in particular the way in which broader cultural factors affect educational outcomes for children on the basis of gender. A recurring feature of findings from PISA outcomes, and those of similar types of testing, is the emergence of gender gaps in performance, with males tending to outperform females on mathematics and science tests and females outperforming males on reading tests (see for instance, OECD, 2016). Cultural factors are increasingly being identified as the key drivers of such outcomes. Citing a study by Halpern, Wai and Saw (2005) on student performance on mathematics tests, Guiso et al. note that Social conditioning and gender-biased environments can have very large effects on test performance (2008: 1164). The question addressed in this paper is whether cultural differences in the roles of women and men persist in shaping opportunity for migrant children now living in Australia. Following this line of inquiry, the purpose of the paper is twofold. First, to provide further evidence on the factors affecting school achievement of students in Australia, and particularly differences by gender. Second, to add to the empirical literature on whether and how cultural factors around gender shape outcomes. Specifically, we analyse the effect of cultural background on achievement in tests of mathematical, reading and scientific literacy completed by students in Australia as part of the PISA at age 15. Differences in secondary school enrolment rates by gender in the country corresponding to students ethnic background are used to generate a proxy measure of the familial cultural attitudes towards gender roles. We hypothesise that in families that come from cultures that place low priority on girls education and on women s labour market participation, girls will display lower achievement due to cultural influences. This may occur through less parental investment in resources and time in the education of daughters relative to sons, and reinforced through an array of cultural norms and rolemodels played out in the media and within family and ethnic networks that influence boys and girls aspirations and behaviours. The following section outlines the observed patterns in the gender gap in the PISA test and provides an overview of research into potential explanations for these outcomes. This is followed by a section outlining the data and methodological issues in the empirical research undertaken for this paper, a section outlining results in view of the key research questions

and a summary section which positions this research in the broader context of the literature and policy issues. 2. Test scores, gender and culture There is an extensive literature examining inter- and intra-country differences in standardised test scores, such as PISA, focusing upon socio-economic gradients and differences in schooling systems and resources, among other factors (for PISA, see: OECD, 2016; for a critical commentary, see also: Tienken, 2007). Considerable gender gaps in performance have been observed. In the most recent round of PISA testing in 2015, such gender gaps were still apparent among OECD countries (the OECD 34 ) amounting to around 8 points for mathematics in favour of boys, 27 points in reading in favour of girls, and 4 points in science in favour of boys (Table 1). In an earlier analysis, Marks (2007), examines such gender gaps in the OECD, and attributes them in large part to policies designed to promote educational outcomes among girls (p. 106). However, there is also considerable variation in these gender gaps across countries, suggesting value in crosscountry analyses to explain differences in gender performance. For example, the highest gender gap in mathematics is 27 points in Austria, with Finland reporting a gender gap of -8 points (with girls outscoring boys in that country), with a similar distribution seen in country means in the reading and science tests. Table 1 Male and Female Mean Scores (Out of 1000) for 2015 PISA Test: Mathematics, Science and Reading, OECD Countries and Australia. Male Female Gender Gap Mathematics Australia 497 491-6 OECD 494 486-8 Reading Australia 487 519 32 OECD 479 506 27 Science Australia 511 509-2 OECD 495 491-4 Source: PISA International Data Explorer (2017). In Australia, the discrepancies are smaller for the mathematics and science tests and greater for the reading tests when compared to the OECD average. Australia fares well with regard to gender equality in educational opportunity, with more females than males completing secondary school and progressing into university. However, the relative exclusion of women in science, technology and mathematics (STEM) subjects and courses - and

subsequently STEM related occupations is a widely acknowledged exception (Marginson, Tytler, Freeman & Roberts 2013). Australia is also quite unique in that children of migrant families tend to perform better in education, on average, than existing Australia residents, a phenomenon attributed to Australia s skilled migration program attracting families which place a high emphasis on education (Entorf and Minoiu, 2004). Cultural explanations An immediate question in raising the issue of cultural influence on educational testing outcomes is the breadth of influences the term culture encapsulates. We adopt Guiso, Sapienza and Zingales (2006) definition of culture as: Those customary beliefs and values that ethnic, religious and social groups transmit fairly unchanged from generation to generation (p. 23). Economists and sociologists have historically emphasised various aspects of cultural influence, be it in terms of class analysis (Karl Marx), religious and cultural practice (Max Weber ) or individualism and social interaction (Adam Smith). In this context, gender has been identified as a powerful driver of expectation and practice, with the formalisation of ideas in the translation between belief and expectations and actions among individuals. Both within and outside economics, these observations about underlying cultural instincts have been classified and interpreted in terms of shaping outcomes in social and economic processes, in literature ranging from Sen s (1985) work on cultural capabilities to Hofstede s (2001) analysis of cultural dimensions in cross-cultural communications and interactions. More recently, this work has been integrated into research into the role of personal identity in economics. Akerlof and Kranton (2000) provide an analysis wherein a person s received identity, including their gender identity, impacts on their actions due to a reduced set of opportunities for action. For instance, the expectation that women in male-dominated occupations such as law will adopt male attitudes and practices in the work environment may influence women s level of engagement and practices in the law (p. 722). Such belief systems can significantly inhibit performance and reduce the incentive for higher achievement. In turn, internalised expectations are reinforced by shared cultural attitudes which reinforce attitudes towards perceived aptitude in certain academic and professional areas (see Reuben, Sapienza and Zingales, 2014, for a relevant discussion in relation to hiring practices in science). In addition to the development of theoretical literature, there is a growing body of empirical literature that investigates how culture influences outcomes at a societal and individual level. For individuals, much of the focus is on the role of culture in fostering a positive sense of identity or self-esteem for ethnic minorities as a causal mechanism linking culture to outcomes, or the enculturation hypothesis (Zimmerman et al 1994).

Several recent studies address cross-country comparisons of standardised test scores. Guiso et al (2008) examine data from 40 countries whose students sat the 2003 PISA tests for reading and mathematics, a sample of 276,165 15-year-old students in total. They find that on average boys outscored girls in the mathematics test (girls had a 10.5 point lower mean score, a gap of 2%) while girls outscored boys in reading (with a mean 6.6% higher than boys). However, there is considerable dispersion in the cross-country gaps in these scores, notably the mathematics scores, with the raw gap for Turkey equal to 22.6 points compared with Iceland where girls had a positive gap of 14.5 points. They estimate a model of test scores including: a measure of real GDP per capita, country dummy variables, various measures of gender inequality such as the World Economic Forum s Gender Gap Index (GGI) and women s workforce participation for each country, and a genetic distance measure to control for biological differences. With these controls in place, they find that the gender measures are highly significant in explaining differences in student performance across countries, with the mathematics gap disappearing in more gender-equal societies and the reading gap tending to widen. In a partial critique of Guiso et al (2008), Fryer and Levitt (2010) confirm the existence of similar cultural effects on the maths gap in the PISA test data set, but note that this effect disappears in a similar analysis using the broader data set from the TIMMS test, with 41 countries who were also subject to the World Economic Forum s gender survey. The reason for the elimination of the cultural gender effect is the inclusion of a number of Middle East countries with relatively lower scores on gender equality, yet smaller gender gaps in mathematics testing (p. 229). One reason they proffer for this outcome is the existence of predominantly single-sex classrooms in Middle East high schools. This qualification is recognised more broadly in the literature. For instance, Guiso, Sapienza and Zingales (2006), point to a general limitation in the literature linking outcomes to culture in that cultural interpretations are often applied ex poste to empirical findings or presented with vague and untestable hypotheses. There is a general assumption that differences in culture are reflected in differences in values or preferences which in turn affect outcomes, but often those intermediate linkages are left untested and findings open to endogeneity or omitted variable bias. Giso et al (2006) note studies by Fenandez, Fogli and Oivetti (2004) and Ferna ndez and Fogli (2005) as examples of more robust research designs. In those studies the labour force participation rates and fertility rates of women of second generation migrant women in America are modelled, using the female participation rates and fertility rates in the women s ancestral home country as instruments to capture culture. This establishes a clear link between the cultural measure as one relating to differences in values and preferences, as well as eliminating the possibility of endogeneity. Migration studies

The interaction between migration and gender influences complicates analysis of the role of culture in explaining migrant outcomes in host countries. Central to this issue is the role of assimilation that process by which new migrant families and groups are integrated into a country s host culture, a notion that was primarily developed by Warner and Srole (1945) in their study of the US and which emphasised the replacement of immigrant cultural traits with those present in the host society. More recent work, such as that of Portes and Zhou (1993) has noted the increased cultural and economic diversity of immigrant groups and have observed that the traditional view on assimilation, fashioned from an analysis of immigration patterns to the middle of the twentieth century, only holds in cases where migrant groups have a cultural heritage that is sufficiently amenable to integration with that of their host country. Alternatively, it is also possible that migrant groups do not readily assimilate and therefore suffer a diminution in opportunity, both academic and economic, or have cultural and economic practices which ensure success in their host countries. This has led to a subsequent connection to a more general literature on social capital, with a focus on both positive and negative associations flowing from resulting cultural and economic outcomes (for a discussion in sociology, see: Portes, 1998). A number of studies look exclusively at the impact of cultural norms and social capital on academic performance in the context of immigrant communities. These studies follow an epidemiological approach, where studies attempt to identify the effect of culture through variation in economic outcomes of individuals who share the same economic and institutional environment, but whose social beliefs are potentially different (Ferna ndez 2011). Dronkers and Kornder (2015) use data from the 2009 PISA test to examine gender differences in the educational performance of first and second generation migrant children in 17 OECD countries, originating from 45 countries. They find that migrant girls tend to perform slightly better than migrant boys relative to domestic averages with a greater gap in the reading tests favouring girls and lower gaps in the mathematics test (p. 631) but that in general, children from countries with lower levels of measured gender equality tended to do less well in their new countries. In relation to migrant academic performance, Figlio, Giuliano, Özek and Sapienza (2016) in a study of administrative educational records in Florida, find evidence that students from cultures with long-term orientation have higher levels of individual attainment than those from countries which do not emphasise long-term goals, with the effect being amplified where more students from similar backgrounds are present in a given school.

Nollenberger et al. (2016) use PISA test data (2003 to 2012) for second generation migrant children to examine the impact of gender attitudes on mathematics scores in PISA. They utilise the Gender Gap Index to measure attitudes towards gender in both the host country and country of ancestry for second generation students, as well as a host of control variables, including per capita GDP. They find that a one standard deviation increase in the level of gender equality in the country of ancestry, as proxied by the GGI, results in a 29% reduction in the gender gap on the PISA mathematics test. In terms of the host country, they find that a one standard deviation increase in gender equality results in a 42% reduction in the maths gap, from which they calculate a lower bound on the impact of the transmission of cultural beliefs on the gender gap at around 67% (29/42) of the measured gender gap that is due to gender-related cultural factors. 3. Methodology and Data We follow the lead of Ferna ndez, Fogli and Olivetti (2004), Ferna ndez and Fogli (2005) and Nollenberger et al. (2016) in testing cultural influences on differential educational outcomes of migrant adolescents in Australia by gender using measures relating to the ancestral country of origin to instrument cultural influences. The instrument used is based on gender differences in secondary school enrolments by country, and thus is designed to relate directly to cultural attitudes towards education for girls and boys. This paper also builds on previous work by extending analysis of the gender gap to all tests in PISA, namely those for mathematics, reading and science. Data This study draws on data from the 2015 PISA test in Australia. The Australian PISA dataset constitutes a rich source of data relating to the performance and personal characteristics of 15 year old students: gender, Indigenous status, migrant history, area of residence, parental occupation and education levels, household socioeconomic status, amount of educational resources at home and country of birth. In addition, data relating to educational characteristics of the student is available, including school sector (i.e. private, Catholic or public school) and educational attainment in mathematics, reading and science, as measured by the PISA tests. In addition to these variables, we include a measure designed to capture the cultural background of the children s families, and aligned specifically with cultural norms relating to gender equality in education. This variable is derived from the school enrolment Gender Parity Index (GPI) generated by the United Nations Educational Scientific and Cultural Organization (UNESCO) Institute for Statistics, and sourced from the World Bank (2017). The GPI is constructed using data on gross secondary school enrolment ratios, with the numerator being the secondary enrolment rate for females, and the denominator the

secondary enrolment rates for males. The GPI has a theoretical range from zero (if female secondary enrolment rates were zero and males rates positive) to infinity (if female enrolment rates are positive and male rates approach zero), with 1 denoting gender equality in secondary enrolment outcomes. For 2015, the GPI varied from 0.46 for the east African country of Somalia, to 1.28 for Tuvalu. GPI values for countries around the world were matched to the 2015 Australian PISA data, based on the students report of their parents country of birth. Initially we use the GPI of the father s country of birth, and the country of birth of the mother where data for the father is missing. Respondents with missing data relating to the variables used in the analysis were removed from the sample. After these exclusions, 12,440 observations remained in the sample. Construction of the cultural variable Using the GPI data allows us to construct a variable, Equity, which proxies the cultural background of the student s family with regard to equality in educational opportunity. The PISA student questionnaire asks students to indicate their country of birth, and the country of birth of their mother and father. This is defined for females and males as follows: Gender GPI Range Value of Equity 01 ln Female = 1 1 0 Male = 1 01 0 1 ln 1 The log transformation means that Equity has a value of 0 for both males and females when there is parity in secondary school enrolment rates (GPI=1) in their ancestral country. We assign the same null value for boys or girls where the enrolment rates are in their favour in the country of ancestry. The hypothesis implicit in this specification is that cultural discrimination against a gender (typically girls) may result in them achieving lower results in their education in Australia, but a cultural norm of discrimination in favour of a gender (typically boys) will not advantage them in their Australian schooling. So the impact of gender discrimination is modelled as asymmetric in this instance. Figure 1 shows the relationship between the constructed Equity variable and the underlying Gender Parity Index for the 2015 UNESCO data. Where the GPI (shown in the top panel) is less than one, the values of Equity are negative for females, but increasing towards zero as the index approaches parity (GPI=1). Moving from left to right as the GPI increases above 1, reflecting lower opportunity for males relative to females in the country of origin, the value of equity becomes increasingly negative for males. The GPI for Australia for 2012 was 0.946, giving values of the Equity variable of -0.056 for Australian girls, and zero for Australian boys.

Figure 1: Relationship between the equity variable and the Gender Parity Index. Estimating strategy The model used in the estimation of the effect on Australian PISA test takers of the gender culture of their family background can be written as: (1) where denotes the academic performance of student i, Z represents a vector of characteristics hypothesised to correlate with student academic performance with associated vector α 1 to be estimated. Equity i is the log-transformed continuous variable capturing cultural norms towards gender equity in student i s ancestral country, with coefficient β to be estimated. In this case, an estimated value of β that is positive and significantly different to zero provides confirmation of the hypothesis that family cultural background affect children s academic performance in Australia. We commenced by estimating this basic model across the sample of all students and with Equity based on the GPI of the country of birth of the student s father, or mother where

information on the father is missing. In testing the sensitivity of the results to different specifications of this variable, we also estimate the model using the GPI of the country of birth of the student s mother. In comparing results by gender of the student, we discovered that in fact it the father s cultural background has the strongest influence on boy s results, and the mother s cultural background the strongest influence on girl s results. Hence the Equity variable is defined for boys on the basis of the country of birth of the father, and on the country of birth of the mother for girls; and on the country of birth of the other parent where the data for the parent of same-sex was missing. Sensitivity tests are also conducted by restricting the sample to second generation migrant students, to implement a more stringent test of the existence of cultural influences. Academic outcomes and plausible values Academic test scores for the students were available in three education categories: mathematics, reading and science. The PISA measurement of test scores utilises a methodology of plausible values as opposed to discrete test scores. The plausible values methodology mathematically computes distributions around reported test score values and assigns each student observation a set of random values drawn from the distributions. In other words, plausible values represent the range of abilities a student might have, and instead of a direct estimation of a student s ability, a probability distribution for a student s ability is estimated (OECD 2009, p. 98). Hence, in the case of each test dimension (mathematics, reading and science), ten plausible values were recorded for a student. In accordance with the PISA data analysis manual, Equation (1) was estimated for each plausible value across each of the three dimensions. The estimated population coefficients and standard errors were then averaged and reported in the results section below. 1 Covariates The PISA data allow for the inclusion in Z i of a range of control variables relating to the student, their family and school attended. These include the child s age, Indigenous status, whether born overseas and whether the main language spoken as home is a language other than English ( English as a second language ); parental education; school sector, educational resources at home, economic, social status and state of residence. The 2012 PISA data included additional variables relating to family structure and parents labour force status. While these are unavailable in the 2015 data, a comparable analysis using the earlier data showed the inclusion of these variables to have minimal impact of the estimates for the Equity variable. 1 In effect, equation (1) was estimated ten times each for the categories of mathematics, reading and science test scores. The estimation results for each category were then averaged and presented in this paper.

4. Results Selected descriptive statistics of the sample are presented in Table 2. Aside from the sample characteristics for the full sample, descriptive statistics for first- and second-generation migrants from the sample are also presented. Migrant children are defined within the PISA data are those whose mother or father were born in a country other than Australia. Within the sample of migrant students, we further define first-generation migrants as students who were born overseas, and second generation migrants as students who were born in Australia. A large proportion of the sample were migrants, with ten percent being second generation migrants and ten percent first generation migrants. Values of the equity variable are lower for the migrant samples, compared to the non-migrant sample. This indicates that Australia has a more equitable gender ratio in secondary school enrolments, on average, than the countries of origin of migrant students. There were also differences in test scores across the samples. Second-generation migrant Australians consistently achieved the highest mean scores in all three educational categories, while non-migrant Australians had the lowest test scores. Small proportions of the migrant samples identified as being Indigenous, while a larger proportion reported English as being a second language. Larger proportions of migrant Australians lived in metropolitan areas, compared to the non-migrant sample. In terms of parental education, migrant parents appeared to have higher levels of post-school qualifications compared to non-migrants Australians. Parental employment status was comparable between migrant and non-migrant samples, although marginally larger proportions of migrant parents were in professional occupations. The results from the model of student academic performance are presented in Table 3. Columns 2, 3 and 4 of Table 3 present results for the educational categories of mathematics, reading and science, respectively. Attention is first drawn to the main variable of interest, the gender culture variable, Equity. The estimated coefficients for this variable are highly statistically significant (p<0.01) for all three measures of academic performance: mathematics (β=68.72), reading (β=57.97) and science (β=73.93) scores. To interpret the magnitude of these results, consider that the mean value of the GPI score for migrant students across the countries in the sample is 0.980, with a standard deviation of 0.133, indicting slightly lower educational access for girls. The predicted penalty in scores between a migrant female student from a country with GPI one standard deviation below that mean, compared to a student from an average country, is 10.0 marks on the mathematics score; 8.4 marks for the reading score and 10.7 marks for the science score.

Table 2: Selected Sample Descriptive Statistics, by Migrant Status Variable Full Non-Migrants Second-gen First-gen Equity match parent s gender -0.031-0.028-0.037-0.042 (0.047) (0.030) (0.073) (0.094) GPI match parent s gender 0.965 0.955 0.995 1.017 (0.060) (0.033) (0.096) (0.119) Mathematics score 492.239 487.826 513.917 505.846 92.153 91.156 92.273 95.479 Reading score 502.242 498.386 525.956 509.437 101.275 100.631 98.375 105.564 Science score 509.099 506.202 525.448 515.949 101.434 101.033 100.258 103.928 First-gen migrant 0.099 - - - Second-gen migrant 0.101 - - - Female 0.505 0.507 0.501 0.493 Age 15.780 15.780 15.776 15.784 (0.288) (0.288) (0.286) (0.289) Aboriginal or TSI 0.177 0.217 0.023 0.011 English as Second Language 0.097 0.029 0.272 0.465 Reside in metropolitan area 0.691 0.641 0.929 0.849 Index of Home Educational Resources 0.018-0.031 0.255 0.174 (1.059) (1.072) (0.941) (1.016) Index of Economic, Social and cultural status 0.217 0.207 0.208 0.302 (0.804) (0.792) (0.817) (0.874) Mother has post-school education 0.483 0.466 0.502 0.600 Father has post-school education 0.390 0.356 0.475 0.577 Catholic school 0.233 0.241 0.259 0.152 Private school 0.191 0.188 0.190 0.213 Public school 0.576 0.571 0.551 0.635 Victoria 0.239 0.227 0.357 0.221 New South Wales 0.156 0.151 0.203 0.148 Queensland 0.201 0.213 0.127 0.176 South Australia 0.119 0.125 0.086 0.108 Western Australia 0.130 0.116 0.142 0.233 Tasmania 0.068 0.081 0.016 0.014 Northern Territory 0.035 0.037 0.019 0.033 Australian Capital Territory 0.052 0.050 0.051 0.066 Observations 12,440 9,948 1,237 1,255

Table 3: Results from the Model of Culture and Academic Outcomes Variable Math Read Science (1) (2) (3) (4) Equity 68.72*** 57.97*** 73.93*** (3.91) (3.08) (3.97) Second-gen migrant 15.98*** 19.29*** 11.81*** (5.72) (6.43) (3.87) First-gen migrant 7.74*** 5.81*** 4.34 (2.64) (1.79) (1.34) Female -6.96*** 29.57*** -4.11** (-4.23) (16.44) (-2.28) Age 17.00*** 7.29*** 15.27*** (6.64) (2.58) (5.41) Aboriginal or TSI -31.01*** -31.82*** -36.30*** (-14.55) (-13.40) (-15.28) English as Second Language -15.20*** -26.89*** -29.55*** Reside in metropolitan area Mother has postschool education Father has post-school education Index of Home Educational Resources (-4.99) (-7.90) (-8.77) -10.96*** -14.28*** -11.48*** (-6.15) (-7.18) (-5.77) 3.82** 5.69*** 2.67 (2.08) (2.82) (1.32) 14.50*** 16.43*** 13.98*** (7.88) (8.15) (6.87) 9.00*** 10.46*** 9.79*** (10.33) (10.80) (10.15) Index of Economic, Social and Cultural 16.54*** 15.46*** 18.54*** Status (14.21) (11.95) (14.37) Catholic School 7.46*** 12.15*** 8.12*** (3.97) (5.98) (3.93) Private School 24.63*** 26.53*** 26.70*** (11.90) (11.82) (11.74) Victoria -5.33-8.19** -11.76*** (-1.45) (-2.05) (-2.92) New South Wales -3.42-6.06-9.97** (-0.91) (-1.47) (-2.41) Queensland -6.58*** -3.98-8.36** (-1.77) (-0.99) (-2.06) South Australia -8.21** -6.66-12.15*** (-2.10) (-1.56) (-2.83) Western Australia 5.60-2.26 1.24 (1.45) (-0.53) (0.29) Tasmania -8.84** -10.76** -16.00*** (-1.96) (-2.19) (-3.23)

Variable Math Read Science (1) (2) (3) (4) Northern Territories -4.68-8.71-9.30 (-0.84) (-1.44) (-1.51) _cons 227.70*** 372.94*** 278.70*** (5.60) (8.35) (6.22) N 12,440 12,440 12,440 R-square 0.20 0.20 0.19 Notes: t-values in parentheses. Estimates are averages over PV1 to PV10 as suggested by PISA. ***, ** and * denote the statistical significance at one, five and ten percent levels, respectively. There were differences in academic performance by migrant background. While migrants performed better academically compared to their Australian counterparts, the results in Table 3 indicate that the type of migrant status is important. While both migrant groups performed better academically compared to non-migrant Australians, second generation migrants those born in Australia - performed the best, with the largest positive coefficients across all three test domains, in comparison to the benchmark group of nonmigrant Australians. Note, however, that this is the partial effect after controlling for English language background. Students who spoke English as a second language (and who will disproportionately be migrant students) had markedly poorer academic outcomes across the three domains, compared to students with English as a first language. Summing the coefficients on the variables for migrant status and non-english speaking background indicates that migrants with English as a second language achieve lower scores in all three domains, other things being equal. The one exception is for maths results for secondgeneration migrants, where the two effects essentially offset one another. Students who identified as Indigenous Australians also performed relatively poorly compared to non- Indigenous students. Another result which reinforces consistent findings in the education literature is that female students performed better in reading compared to their male peers, but achieved lower scores in mathematics and science. Older students performed better than their younger peers, even though the variation in age was in months, rather than years, for the large part. This is an unsurprising finding though, with a large amount of literature finding that age is a strong determinant of academic success, especially at younger ages (see for instance Bedard and Dhuey 2006; Li and Miller 2009). Parents holding post-school qualifications notably the father was positively associated with academic outcomes. Students living in non-metropolitan areas have lower test scores. Student attending private schools achieved markedly higher results compared to those in government schools, while those from Catholic schools also marginally outperform their government school peers. Having more educational resources at home was associated with positive academic outcomes, as was better economic, social and cultural status. There were also minor

differences in academic performance by state of residence. In particular, students from Tasmania appeared to perform least well, with negative estimates across all three academic areas, and also having the largest estimated effect of a lower 16 points for the area of science compared to the control group (Australian Capital Territory). Effects of culture sensitivity analysis The initial results reported in Table 3 are consistent with the hypothesis that Australian students coming from a cultural background in which their gender is afforded lower opportunity in education in turn achieve lower scores in standardised mathematics, reading and science tests. A number of additional models were estimated to further explore this finding and its robustness. The results relating to the key variable of interest, Equity, from those additional models are reported in Table 4. The first row replicates results from the initial model, which is based on the full 2015 sample. As a more stringent test of the effect of cultural background, the models were re-estimated with the 1,255 first generation migrant students removed from the sample. Potentially, students who were born overseas and moved to Australia with their families may have already been subject to relative exclusion in educational opportunity due to the institutional arrangements in their country of origin. By limiting the sample to people born in Australia Australians and second generation migrants, we are assured that all students have had access to the Australian education system, which acts as a control for this effect. The Equity variable is thus a proxy purely of family cultural background with respect to gender norms, and how this is passed on to children. The coefficients remain positive, but reduced in magnitude for mathematics (59.75, compared to 68.72 for the full sample), and a slight increase for reading (60.01 compared to 57.97), both and are observed at lower levels of statistical significance. The results on the science see the equity variable parameter marginally reduced from 73.93 to 72.21, but retain its high level of statistical significance. In contrast to these results, the reintroduction of first generation and exclusion of second generation migrant students sees a substantial increase in the measured effect of the equity variable, for instance, in the case of the mathematics test, rising from 68.72 in the full sample to 82.79, with the estimate being significant at p<0.01 for all three tests. Hence, the influence of gender cultural attitudes on students country of origin observed on academic performance in mathematics, reading and science in the base models are stronger for first generation migrants than second generation migrants. While this remains consistent with an influence of family cultural background, it is not possible for first generation migrants to further distinguish between the relative contributions of family culture and the students actual childhood experiences in their birth country. Our prior expectation in undertaking this analysis was that cultural influences would be most pronounced with respect to inequality of educational opportunity for girls, given the

results from previous studies which showed the importance of the gender equality variable (e.g. Nollenberger et al., 2016) in relation to maths outcomes. To test this idea, we reexamined the base model for each PISA test component using male and female subsamples. Surprisingly, we see that the estimated effect of the equity variable, while still significant for both sub-samples, is greater for males by a factor of around three! Our prior expectations in undertaking this analysis - that cultural effects would primarily be seen in terms of how culturally constructed gender roles impacted upon the performance of girls appears to have been thoroughly misplaced. Table 4: Summary of results for Equity variable under different specifications Variable Math Read Science (1) (2) (3) (4) Full Sample Equity (Parent cob) 68.72*** 57.97*** 73.93*** Obs=12,440 (3.91) (3.08) (3.97) Australian and 2 nd generation migrants (1 st generation excluded) Equity (Parent cob) 59.75** 60.01** 72.21*** Obs=11,185 (2.44) (2.22) (2.67) Australian and 1st generation migrants (2nd generation excluded) Equity (Parent cob) 82.79*** 73.36*** 90.46*** Obs=11,203 (3.83) (3.20) (4.01) Female students only Equity (Mother cob) 55.13*** 45.22** 59.07*** Obs=6,285 (2.81) (2.16) (2.88) Male students only Equity (Father cob) 148.51*** 125.00** 147.38*** Obs=6,155 (3.25) (2.51) (2.88) 5. Conclusions Educational testing, including that undertaken for PISA, shows that there are observable and repeated patterns in the outcomes on the basis of gender, with boys outperforming girls on mathematics and science components and girls seeing better outcomes on reading tests. This paper has examined some of the likely factors affecting such outcomes and the particular role that cultural background plays in shaping the gender gap. It follows studies such as Ferna ndez, Fogli and Olivetti (2004), Ferna ndez and Fogli (2005) and Nollenberger et al. (2016) in using such country measures as a proxy for attitudes to gender equality, in this case, a variable based on UNESCO s GPI index for the country of the test-taker s parents (the Equity variable).

An innovation of the paper is in examining the extent to which these attitudes are transmitted via migration and the extent to which they are retained during the process of acclimatisation in host countries, in this instance via a comparison of first- and secondgeneration migrant outcomes in the PISA test. The key finding from this analysis is that, for children who are resident in Australia, the level of inequality in access to secondary education for children of their own gender in their country of ancestry is highly statistically in determining their own mathematics, science and reading scores. In the case of a migrant 15 year old female from a country with GPI one standard deviation below that of the average country, there is an observed reduction in performance on the 2015 PISA of 10.0 points on the mathematics test, 8.4 marks on the reading test and 10.7 marks on the science test. To place this in context, in Australia the overall gender gap for the 2015 PISA mathematics component is 6 points in favour of boys, the reading gap is 32 points in favour of girls, while the science gap is 2 points in favour of boys. From this, it would appear that the Equity measure captures important negative effects on the mathematics and science tests, other things being equal, with the impact on reading being more moderate in view of the very strong advantage girls enjoy on this component. Further, the magnitude and significance of the Equity measure for second generation migrant students is similar to that seen among first generation migrant students. This is somewhat unexpected given the traditional view of migrant assimilation, in which it takes time for migrant families to adapt to the cultural context of their host societies (between first and second generations), but aligns with the literature stemming from Portes and Zhou (1993) which posits an increasingly complex relationship of interaction between migrant and host communities in recent decades. The important point, however, is the fact that the result holds for second generation migrants provides strong evidence that the effect is one that is transmitted through cultural effects, since all second generation children have had access to the same (Australian) education system. A surprising finding is that these effects are in fact much stronger for boys. Much of the literature around how gender roles influence academic performance, and indeed our own motivation in undertaking this analysis, concentrates on how culturally constructed gender identities inhibit the achievement of girls, notably in mathematics and science. Dronkers and Kornder (2015) observe that higher degrees of gender inequality in migrants original countries impact on PISA scores for both migrant male and female students. For the students included in this analysis, less than half of the relevant countries of origin had under-representation of girls in secondary education: 68 countries as opposed to 74 in which boys were under-represented in secondary school enrolments. It must be noted, however, that it is girls who face the most extreme cases of inequality in countries such as Somalia, Central African Republic, South Sudan and Afghanistan and Zhambia with enrolment rates less than 60% of that for boys. However, lower rates of participation in education in the country of origin are estimated to have roughly three times the detrimental effect on boy s test scores than on girls.

Exactly how cultural background impacts upon academic perform is, of course, an important question that cannot be addressed with these data. A further finding from the analysis may shed some light on this question. The most robust effects were obtained for the Equity variable when it is measured on the basis of the country of birth of the parent of the same gender as the student. This is suggestive of student performance being influenced by cultural stereotypes passed on through the parents as role models, possibly mediated by the student s preferences, expectations and aspirations being based on what they observe of their mother in the case of daughters; and of what boys observe of their father. References Akerlof, G. A., and Kranton, R. E. (2000), Economics and Identity. The Quarterly Journal of Economics, Vol. 115, No. 3, pp. 715 753. Bedard, K. and Dhuey, E. (2006), The Persistence of Early Childhood Maturity: International Evidence of Long-Run Age Effects, The Quarterly Journal of Economics, vol. 121, no. 4, pp. 1437-1472. Dronkers, J. and Kornder, N. (2015), Can gender differences in educational performance of 15-yearold migrant pupils be explained by societal gender equality in origin and destination countries? Compare, Vol. 45, No. 4, pp. 610 634. Ferna ndez, R. (2011), Does Culture Matter?, in J. Benhabib, A. Bisin and M. O. Jackson (eds), The Handbook of Social Economics, Volume 1A, Amsterdam: North Holland, pp. 481-510. Ferna ndez, R. and Fogli, A. (2005), Culture: an Empirical Investigation of Beliefs, Work, and Fertility, NBER Working Paper No. 11268. Ferna ndez, R., Fogli, A. and Olivetti, C. (2004), Mothers and Sons: Preference Formation and Female Labor Force Dynamics. Quarterly Journal of Economics, Vol. 119, No. 4, pp. 1249 99. Figlio, D., Giuliano, P., Özek, U. and Sapienza, P. (2016), Long-Term Orientation and Educational Performance, IZA Discussion Paper No. 10147 (August), IZA: Bonn, Germany. Fryer, R.G. and Levitt, S.D. (2010), An Empirical Analysis of the Gender Gap in Mathematics, American Economic Journal: Applied Economics, Vol. 2, No. 2, pp. 210-240 Guiso, L., Monte, F., Sapienza, P. and Zingales, L. (2008), Culture, Gender and Math, Science, Vol. 320, No. 5880, pp. 1164-1165, DOI: http://dx.doi.org/10.1126/science.1154094. Guiso, L., Sapienza, P. and Zingales, L. (2006), Does culture affect economic outcomes?, Journal of Economic Perspectives, Vol. 20, No. 2, 23-48. Halpem, D., Wai, J. and Saw, A. (2005), A Psychobiosocial Model: Why Females Are Sometimes Greater Than and Sometimes Less Than Males in Math Achievement in Gender Differences in Mathematics, A. M. Gallagher and J. C. Kaufman, (eds), Cambridge University Press: New York, pp. 48 72.

Hofstede, G. (2001), Cultures consequences: comparing values, behaviours, institutions and organizations across nations, second edition, Sage: Thousand Oaks, California. Li, I.W. and Miller, P.W. (2009), Academic performance and graduate outcomes: does age matter? in New Research In Education: Adult, Medical and Vocational, Nova Science Publishers, Inc., New York Marginson, S., Tytler, R., Freeman, B., & Roberts, K. 2013. Securing Australia s future: STEM country comparisons. Melbourne: Australian Council of Learner Academies (ACOLA). Marks, G. (2007), Accounting for Gender Gaps in Student Performance in Reading and Mathematics: Evidence from 31 Countries. Oxford Review of Education, Vol. 34, No. 1, pp. 89-109. Nollenberger, N., Rodríguez-Planas, N. and Sevilla, A. (2016), "The Math Gender Gap: The Role of Culture." American Economic Review, 106(5): 257-61. OECD Organisation for Economic Cooperation and Development (2009), PISA Data Analysis Manual: SAS Second Edition, OECD Publishing, Paris. DOI: http://dx.doi.org/10.1787/9789264056251-en OECD (2016), PISA 2015 Results (Volume I): Excellence and Equity in Education, OECD Publishing, Paris. DOI: http://dx.doi.org/10.1787/9789264266490-en OECD (2017a), About PISA, available at: http://www.oecd.org/pisa/aboutpisa/ OECD (2017b), PISA International Data Explorer, available at: http://piaacdataexplorer.oecd.org/ide/idepisa Portes, A. (1998), Social Capital: Its Origins and Applications in Modern Sociology, Annual Review of Sociology, Vol 24, pp. 1-24 Portes, A. and Zhou, M. (1993) The New Second Generation: Segmented assimilation theory and its variants, The Annals of the American Academy of Political and Social Science, Vol. 530, pp. 74-96. Reuben, E., Sapienza, P., and Zingales, L. (2014), How stereotypes impair women s careers in science. Proceedings of the National Academy of Sciences, Vol. 111, No. 12, pp. 4403-4408. Sen, A. (1985), Commodities and Capabilities, Amsterdam: North-Holland. Thompson, E.A., Thompson, A.L. and Fromer, J.E. (eds.) Resiliency in ethnic minority families, Vol I: Native and immigrant American families, Centre for Excellence in Family Studies: The University of Wisconsin: Maddison, Wisconsin. pp. 199-220. Tienken, C.H. (2017), Understanding PISA Results, Kappa Delta Pi Record, Vol. 53, No. 1, pp. 6-8.

Triandis, H. C. (2000), Cultural syndromes and subjective well-being, in Diener, E. and Suh, E.M. (eds) Culture and Subjective Well-Being, MIT Press: Cambridge, pp. 13-36. Warner, W.L. and Srole, L. (1945) The Social Systems of American Ethnic Groups, New Haven: Yale University Press. World Bank (2017), School Enrolment, Secondary (Gross), Gender Parity Index, available from: http://data.worldbank.org/indicator/se.enr.seco.fm.zs Zimmerman, M.A., Ramirez, J., Washienko, K.M., Walter, B. and Dyer, S. (1994), Enculturation hypothesis: Exploring direct and positive effects among native American youth, in McCubbin, H.L., Thompson, E.A., Thompson, A.L. and Fromer, J.E. (eds.) Resiliency in ethnic minority families, Vol I: Native and immigrant American families, Maddison Wi., Centre for Excellence in Family Studies, University of Wisconsin, 199-220.