Social class segregation in upper-secondary school choice in Canada

UNIVERSITY OF THE BASQUE COUNTRY - UPV/EHU FACULTY OF ECONOMICS Social class segregation in upper-secondary school choice in Canada A thesis submitted by Pedro Iraburu for the Master in Economics: Empirical Applications and Policies Supervised by: Petr Mariel 30th of July, 2020 Bilbo Abstract The aim of this work is to analyze how the spoken language at home influences the parental decision on school choice in Canada. There is a vast literature on the factors affecting school choice, but very little research has been made on bilingual or multilingual countries. We take data from the 2012 PISA questionnaires to estimate the probability of choosing a determinate type of school, discriminating by instructional language and financing. The results show clearly that the impact of bilingualism dilutes otherwise very important socio-economic variables. Keywords: School choice, bilingualism, multinomial logit. 1 INDEX 1 Introduction 4 2 Literature Review 6 2.1 The Influence of Official Minority Languages . . . . . . . . . . . . . . . . . 6 2.2 Previous Studies on School Choice . . . . . . . . . . . . . . . . . . . . . . . 8 2.3 A different approach: bilingualism . . . . . . . . . . . . . . . . . . . . . . . . 10 3 The Data Analysis 12 3.1 Description of the Database . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 3.2 Division of the provinces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 4 Methodology 17 4.1 Theoreticalframework.............................. 17 4.2 Interpretation of the model: discrete changes in probabilities . . . . . . . . 18 5 Empirical Results 20 5.1 Ourmodels.................................... 20 5.2 Relevance of the variables: Wald test . . . . . . . . . . . . . . . . . . . . . 20 5.3 Discrete changes in the Probabilities . . . . . . . . . . . . . . . . . . . . . . 23 5.3.1 First model: Group I of provinces .................... 23 5.3.2 Second model: Group II of provinces .................. 25 6 Concluding Remarks & Further Research 30 References 32 Appendix 36 2 List of Figures 1Change in probability with respect to the benchmark family, Group I . . . . 28 2Change in probability with respect to the benchmark family, Group II . . . . 29 List of Tables 1Population by mother tongue and geography, 2016 .............. 5 2Number of Surveyed Students ......................... 12 3Summary statistics of the explanatory variables ................ 13 4Group I: Distribution of public, private, English and French school. ..... 15 5Group II: Distribution of public, private, English and French school. ..... 16 6Wald tests for the significance of the explanatory variables (I). ........ 21 7Wald tests for the significance of the explanatory variables (II). ....... 22 8Multinomial logit: first model estimation .................... 36 9Binary logit: second model estimation ..................... 37 3 1 Introduction The goal of the thesis is to cast light on the factors affecting the parental school choice in Canada, while taking into account the role of bilingualism. Canada is divided into 10 provinces and 3 territories. As these last are quite small, they were not included in the Programme for International Student Assessment (PISA) study (OECD, 2012), which we make use to extract the data necessary for our analysis. There exists a wide array of languages coexisting in Canada, including many indigenous ones. In spite of this, only 209,570 individuals speak aboriginal languages in private (Statistics Canada, 2016) and thus, were not considered in our analysis. The two main languages in Canada are English and French, although the last one is not as widely spread. According to Statistics Canada (2020), about 25% of Canadians use French as their home language, which is consistent with our findings from the PISA 2012 data (see Table 3). About 25% of Canadians use French as their home language (Statistics Canada, 2020), which is consistent with our findings from the PISA 2012 data (see Table 3). As Table 1 shows, the population in Canada is polarized, with many provinces sparsely populated and sharing a low percentage of the total. Most population of Canada is concentrated in Ontario and Quebec, each one being the main representative of their respective official language: English and French. Quebec concentrates the majority of francophones along with New Brunswick, being it the only territory that has both languages as official. According to the Section Sixteen of the Canadian Charter of Rights and Freedoms (Constitution Act, 1982), English and French are the two official languages of the State, meaning that both have equality of status in the Parliament, courts and Government of Canada. At the province level this is different, as usually only one is considered the official, although some of the provided public services are available in both of them. 4 Given all of this, all schooling across Canada is implemented in both languages, but in different proportions depending on the province. The majority of it is publicly funded (see Table 2). The educational system is divided into primary, secondary and post secondary education, although it varies on the territory, as it is under exclusive provincial jurisdiction (CICIC, 2020). For example, education is compulsory up to the age of 17 years all throughout the country, except for Ontario, Manitoba and New Brunswick, where it is 18 years. In regard to the 2012 PISA results, Canada stands as one of the countries with the highest scores, being in the top 10 of the best performances in mathematics, reading and sciences (Brochu, Deussing, Houme & Chuy, 2013). Among all the territories in Canada, Quebec was placed in the top-performing participants globally in paper-based mathematics. In sciences, students in British Columbia and Alberta performed better than the country’s average. Table 1: Population by mother tongue and geography, 2016 Province Population Official Lang. Speakers by Mother tongue (%) English French Ontario 13,312,870 English 68.2 % 4.0 % Quebec 8,066,555 French 8.1 % 78.0 % British Columbia 4,598,415 English 70.0 % 1.4 % Alberta 4,026,650 English 75.4 % 2.0 % Manitoba 1,261,620 English 72.6 % 3.4 % Saskatchewan 1,083,240 English 83.2 % 1.5 % Nova Scotia 912,300 English 91.4 % 3.4 % New Brunswick 736,285 Both 64.8 % 31.9 % New F. and L. 515,680 English 97.1 % 0.5 % Prince Edward Is. 141,020 English 91.1 % 3.6 % Source: Statistics Canada. 5 2 Literature Review During the last century, many authors have studied the variables and factors affecting the school choice, but only few have considered the coexistence of more than one language in their models. Bilingualism in Canada is the key issue to understand the current situation and how it has affected its education system throughout the centuries (Vaillancourt, 2012). In Canada, language and religion have been always intrinsically related, as 70% of the French speakers in 1871 were Catholic. This fact led to a historical indirect protection of the language through religion, which started with the establishment of the Constitution Act of 1867, whose section 93, was introduced as a measure of protecting those Catholic Communities. This stated that the current and future school systems of the religious minority (which could be either protestant or catholic), existing in a province, should be protected from any potential harm from its provincial government; and gave authority to the Government to intervene to protect such minority. 2.1 The Influence of Official Minority Languages The introduction of the Constitution Act led to situations such as the one in New Brunswick, which had a very large percentage of Catholics in 1871, and which nowadays share both languages as official ones. In Ontario, although having a very small linguistic minority, there exists an above average constitutional protection, as half of the population spoke French at the time. This protection was necessary, as there have been always a unwillingness of English-Canadians to collaborate in such purpose, giving the francophones in Canada an “added incentive to pursue nationalistic language policies, with the possibility of some consequences that English Canadians may not always appreciate”, as Breton stated in his 1978 paper, “Nationalism and Language Policies”. As this system did not consider the language but the religious rights, in 1982 the “Canadian Charter of Rights and Freedom” was redacted, where the protection of minority languages educational rights got introduced into the legal system. Villancourt (2012), analyses the measurement of the costs and benefits of these official languages policies in 6 Canada, and explains that, although the coexistence of these two languages do not have a positive direct effect, it greatly increases the person welfare, as the public services are available in their mother’s tongue. Apart from this, Canadians may also have an increased utility from other types of benefits, such as the obvious cultural value increment of having more than one official languages, and indirect monetary effects. Christofides and Swidinsky (2008), found that men outside of Quebec (where the majority is francophone) with French language skills, tended to be disproportionally represented in higher paying occupations. Breton (1998), analysed census data from 1971-1991, concluding that there were significant returns to bilingual language skills in the estimated wage equations. Being a bilingual country have also its detriments, as the coexistence of the two languages can be very expensive regarding bureaucracy, as a consequence of the multilingualism (Fidrmuc, 2011;Pons-Ridler & Ridler, 1990). The school system in Canada differs from other bilinguals’ countries, such as Belgium or Switzerland, where the instruction language of the school depends on the region, and there is no real free choice. Similar to Basque Country, in Canada the parents can choose the language of education for their children. This concept is explored by Pons-Ridler & Ridler (1990), stating that the establishment of bilingualism in Canada, respecting the school choice, follows the “Personality Principle”. According to that, the individual customer of Government services will be attended in the language of his choice. The authors also state that territoriality would be a cheaper alternative, in a similar way to Belgium and Switzerland, as it would save a lot of education expenditure. Languages can be seen both as a means of communication and as part of a cultural identity as established by CASLT (2016). For instance, English is widely used in the business as a communication tool, and no cultural heritage is derived from its use in such environment. Specifically, CASTL (2016, pp. 62) states that, “Language, however, transmits not only meaning in a strict, terminological sense, but also commonly held moral 7 values, judgments of other social or political references”. It is quite clear that the francophones in Canada have historically shared a common heritage, given how it was tied to the Catholic Community. 2.2 Previous Studies on School Choice Manski and Wise (1983) are considered the first authors to model choice in the educational environment, but now the literature about school choice is abundant. For example, Nakhaie (2000) found that the parental education and occupation have significant consequences for their children’s educational attainment. The study used log odd ratios to analyse two national representative samples of Canadians surveyed in 1985 and 1994. Likewise, Borgers et al (1999), use an extended logit model to analyse data from Veldhoven, in the Netherlands, which has two Catholic, one Protestant and two public schools. They found that the religion of the household, the size of the classroom, and the distance to the schools, were the most relevant variables. This is proved once again by M¨ uller, Haase and Seidel (2012), by analyzing German schools data with a multivariate analysis to find that there exists an increasingly competitiveness among all schools, as a result from the possibility of free school choice. Their approach explicitly accounts for spacial substitution, as the authors considered that this patterns between school locations exist, setting a different approach from the previous literature, which is usually focused on racial mix, tuition fees, and travel-to-school distance. Following the same steps, H´ etu (1991) uses a multivariate analysis model to analyse gathered data from 1,894 French speaking students in Quebec, at grade 5, 10 and 12, where 12.7% assisted private schools, to gather different conclusions, such as: “[...] parents’ assessment of the ‘quality’ of teaching in public and private institutions probably corresponds to a particular teaching style, moral values and cultural preferences rather than merely to school performance measured by grades and failures.” (H´ etu, 1991, pp. 496) This clearly shows that the culture is a key factor for parents when considering different school types. 8 their socio-demographic characteristics, we apply a MNLM in which four different type of school represent the explained variable. The codification for each outcome, as computed in the model, is respectively: 1 for the Public English, 2 for the Private English, 3 for the Public French, and 4 for the Private French. In the second group, we discarded individuals that attended to private schools, as the number of observations for this type of schools was insufficient to perform the analysis. Accordingly, in this second group we created a binary logit model with only two outcomes: Public English or Public French. In each subset, we can find a province expected to have a higher influence on the language component, given its francophone population: Quebec in the first, and New Brunswick in the second. In this second subset, the dependent variable will take value 1 if the instructional language of the school is French, and 0 if English. Table 4: Group I: Distribution of public, private, English and French school. Quebec Manitoba Saskatchewan British Columbia Type n % n % n % n % Public English 1,165 34.1 1,481 80.5 1,588 90.7 1,409 86.4 Private English 293 8.6 94 5.1 74 4.2 73 4.5 Public French 1,713 50.2 204 11.1 32 1.8 92 5.6 Private French 243 7.1 60 3.3 56 3.2 57 3.5 Total Private 536 15.7 154 8.4 130 7.4 130 8.0 Source: OECD PISA 2012, Canada. 15 Table 5: Group II: Distribution of public, private, English and French school. Nova Scotia Alberta New Brunswick Ontario Type n % n % n % n % Public English 1,064 86.6 1,582 89.0 820 59.6 2,069 63.4 Private English - - - - 28 1.7 48 1.5 Public French 165 13.4 166 9.3 805 48.7 1,147 35.1 Private French - - 29 1.6 - - - - Total Private - - 29 1.6 28 1.7 48 1.5 Source: OECD PISA 2012, Canada. 16 4 Methodology In this section, we will set the theoretical foundation and the description of the discrete choice models used in our analysis: the multinomial and binary logit. 4.1 Theoretical framework As our dependent variable is a categorical one with 2 or 4 outcomes - depending on the modelwe apply a discrete choice model to analyse the likelihood of an individual picking a specific alternative. Consequently, we model the parent’s school choice by a MNLM in provinces where four type of schools are available (English Private, English Public, French Private, French Public) and binary logit model in provinces with only two alternatives (English Public, French Public). For the first subset, we have estimated a MNLM, for the following four outcomes: Public English,Private English,Public French and Private French, coded in the model as y=1, 2,3,4, respectively. Equations (1) and (2) represent the probability of outcome min a MNLM given socio-demographic variables of the i-th family (xi).βare the corresponding parameters to be estimated. For the sake of identification, in equation (1) the parameters for one category (1 in our case) are set to zero. Pr(yi=1|Xi) = 1 1+∑J j=2exp(X0 iβj),(1) Pr(yi=m|Xi) = exp(X0 iβm) 1+∑J j=2exp(X0 iβj)for m <1.(2) This equations will be the basis for the Maximum Likelihood estimator, which will be the method used for the estimation of the parameters, for both models. 17 For the second subset, we have estimated a binary logit, which only has two outcomes instead: Public English and Public French. The dependent variable ywill take value 0 if the instructional language of the school is English, or 1 if it is French. Equation (3) presents the formula for the probability of outcome one (Long, 1997). Pr(y=1|xi) = F(x0 iβ) = exp(x0 iβ) 1+exp(x0 iβ),(3) This equation represents, therefore, the probability of choosing French as the instructional language given socio-demographic characteristics of the i-th family (xi), and βare the corresponding parameters to be estimated. The reference level is outcome zero, that is, a Public English school. 4.2 Interpretation of the model: discrete changes in probabilities As Long (Long, 1997, pp. 164) states, there are many parameters in a MNLM. Although the statistical significance is important, the magnitudes and the direction of the effects cannot be interpreted directly.There are several methods of interpreting the results of the estimation, such as predicted probabilities and partial changes. In this research, we have used discrete change to measure the changes in the probabilities. it can be applied to continuous and dummy variables. The predicted probability of the dependent variable y, being equal to the outcome m given the vector of covariates xiis: c Pr(y=m|xi) = exp(x0 iˆ βm) ∑J i=1exp(x0 iˆ βj)(4) 18 By changing the xkparameter from xSto xE,ceteris paribus, the change in the predicted probability is: ∆ c Pr(y=m|x) ∆xk = c Pr(y=m|x,xk=xE)− c Pr(y=m|x,xk=xS).(5) The magnitude of the change in probabilities depends on the amount of change in xk, the starting value of xkand the other variables (Long, 1997). Most of the time, the other x−kvariables are held at their mean values, with all the dummies at either 1 or 0. How much should xkvary? If it is a dummy, the choice is pretty straightforward, as it is a discrete change from 0 to 1, or vice versa. If on the contrary, the chosen variable is continuous, the amount of change depends on your analysis and on the nature of it. In our case we use a standard deviation change but this issue is further discussed in section section 5.3. Usually the best available tool for this type of analysis is a discrete change plot, as it quickly summarizes all the information. In this type of graph, the horizontal axis represents the magnitude of the positive or negative change in the probability for each outcome. These are represented by letters. The vertical axis includes all the variables of the model (see fig. 1 and fig. 2). This analysis tool is suitable for both binary and multinomial logits. 19 5 Empirical Results This section contains all of the obtained results from our analysis, from the definition of our model to the interpretation of our estimations. 5.1 Our models As explained in section 3.2, we have split the database as well as our analysis into two, as there were two regions clearly defined by its private schooling percentage. Given that we include dummy variables for provinces in the two models the set of explanatory variables differ slightly. Nevertheless, the key variables for the interpretation family wealth and the spoken language in the household are common in the two models. For the estimation of the model, we adopt the Rprogramming language, more specifically, the VGAM and NNET packages. Moreover, as we cannot interpret the coefficients directly, due to the non-linear nature of the model, we interpret our results using discrete changes in probabilities, as explained in section section 4.2. Posteriorly, both formulas are shown. Hence, for the first group of provinces, we have defined a MNLM as the tool to analyse the variables that drives Canadian parents to choose a specific type of school, in a bilingual environment. Our dependent variable, type of school, has four categories: public English, private English, public French and private French. As stated above, Table 3 presents the summary statistics for the explanatory variables, divided in continuous or dummy categories. This variables were based on the previous research done by Mariel and Vega-Bayo (2015,2018a,2019), regarding the subject. For the second group of variables we make use of a binary logit, where we estimate the probability of selecting a French speaking school compared to a English one. 5.2 Relevance of the variables: Wald test The estimation outcomes of binary logit and MNLM are presented in Tables 8 and 9in the Appendix. In order to test the significance of all the parameters associated with the 20 outcomes of both models, we have chosen the Wald test to analyze the significance of the coefficients. The null hypothesis of this tests, is that a specific explanatory variable does not have an effect on the explained variable. It is, therefore, a joint test that involves three parameters in the MNLM. Table 6 and Table 7 show the results of this test. Most of them are significant at the 1% level, although three variables in the binary logit are not relevant at 5% significance level. We are keeping them as they are relevant to the first model, for comparison purposes: Siblings,the educational level of the father and two-parent family. Table 6: Wald tests for the significance of the explanatory variables (I). Variable χ2statistic p-value Mother not working 42.3 <0.01 *** Two-parent family 13.3 <0.01 *** Siblings 31.1 <0.01 *** Grandparents living with the family 15.3 <0.01 *** Cultural possessions 128.9 <0.01 *** Educ. level mother 21.7 <0.01 *** Educ. level father 17.3 <0.01 *** Family Wealth 108.4 <0.01 *** Highest parental occup. status 122.7 <0.01 *** Home language not English 1242.2 <0.01 *** Quebec 284.3 <0.01 *** Sasketchwan 63.2 <0.01 *** British Columbia 58.7 <0.01 *** ***, **, * denotes significance at the 1%, 5% and 10% level respectively. 21 Table 7: Wald tests for the significance of the explanatory variables (II). Variable χ2statistic p-value Mother not working 26.8 <0.01 *** Two-parent family 13.3 <0.952 Siblings 0.1 0.335 Grandparents living with the family 35.9 <0.01 *** Cultural possessions 116.5 <0.01 *** Educ. level mother 9.8 <0.01 *** Educ. level father 1.9 0.167 Family Wealth 7.3 <0.01 *** Highest parental occupp. status 36.8 <0.01 *** Home language not English 1247.3 <0.01 *** Ontario 66.8 <0.01 *** New Brunswick 114.4 <0.01 *** Alberta 41.4 <0.01 *** ***, **, * denotes significance at the 1%, 5% and 10% level respectively. 22 5.3 Discrete changes in the Probabilities This last section of the empirical results includes the overview of the discrete changes in the probabilities and thus, it represents the key result for the interpretation of our model. Results include, for each group of provinces (there are two, as seen in section 3.2), an analysis for each explanatory variable. The theory behind this methodology is explained in section 4.2. As previously mentioned, there are both dummy and continuous variables, and each type follows a distinct technique when studying its influence. In the first case, the change is intuitive: we will switch the initial value of the parameter from 0to 1or vice versa, depending on the benchmark. For the latter, we will add and subtract the standard deviation to the mean. This is not arbitrary, the logic behind this is that this shift represents not a unit, but a relevant change of the variable. Group I is the first group of provinces for which the interpretation will be presented, that is, those that showcased enough data for private schooling analysis. Afterwards comes Group II, where we only discriminated the schools by language. The outcomes will be illustrated in fig. 1 and fig. 2. The tables containing the coefficients and the std. errors for each group can be found in the appendices. 5.3.1 First model: Group I of provinces This group is represented by Quebec, Manitoba, Saskatchewan and British Columbia. Quebec is the second biggest province, both financial and population wise (Statistics Canada, 2016). It is also the principal francophone territory. British Columbia and Manitoba are not as big as the latter, but they are not small either, being the third and the fifth provinces by populace (see Table 1). 23 The setting of the characteristics of the benchmark family should not be an arbitrary decision, as it can ease or make more difficult the subsequent interpretation. Having analyzed the magnitude of the explanatory variables, we set the benchmark characteristics as follows: (a) Family living in Manitoba, as Quebec, Saskatchewan and British Columbia had a bigger differential effects when introduced as dummies (therefore these three dummies are set to zero). (b) The home language is English, as it is the most common outcome. (c) The continuous variables are set to the mean, these includes the family wealth, the occupational status, and the educational level for both mothers and fathers. (d) The remaining dummy variables have been set to the most common outcome, with siblings and with no grandparents living in the family unit. Figure 1 shows the changes in probability for a zero-one, or standard deviation change, for each variable depending on its nature. The Rlanguage did not have an implemented library for this purpose (unlike Stata), so everything in the process is self coded. Starting with the continuous variables, we can appreciate that that their effect is generally not big. The cultural possessions,occupational status and the educational level of both parents show little to none influence over the probability changes. In spite of this, the family wealth parameter gains weight when we take in account the differential effects of the provinces. In both Saskatchewan and British Columbia the gain is noticeable, but not as huge as in Quebec. We can appreciate that the probability of choosing a public school (1and 3) decreases, implying that for richer families, private schools are preferred. For Saskatchewan, public French schools have a higher probability of being chosen, this could come because of the interest of richer households of their children speaking more 24 Overall, the estimation of the parameters from both models are robust. The Wald test confirms that all variables are relevant across both models. The probability of choosing schools where French is the instructional language is affected the most by those variables related to the home language. This proves that bilingualism is the main driver of school choice. Among the provinces, Quebec and New Brunswick uncover the greater influence of French when compared to others. Besides, the second most important group of parameters are those related to family wealth. Once again, Quebec and Ontario showcase bigger differential effects, which is logical, given that they are among the richest territories in Canada (Statistics Canada, 2016). Having analyzed all of the variables, a few stand out for their lack of relevance, such as siblings and the educational level of both parents, which are relevant in other papers, such as in the Mariel et al study of Japan (2019). In conclusion, similar to some studies in Basque Country (Mariel et al Spanish study, 2015), we have gathered for Canada compelling evidence of bilingualism diluting the socio-economic segregation previously found in the school choice. It would be interesting to analyze other bilingual or multilingual countries. Nonetheless, we need to take into account that not many countries boast from the unremarkable situation of Canada or the Basque Country. In order to properly understand the bilingualism effect on the school choice, the instructional language must not be restricted by territory (e.g. Belgium, where the language depends on the area in which you study). There are 55 officially bilingual countries in the world (University of Ottawa, 2020), so there is plenty of potential research ahead. Regarding Canada, it would also be alluring to use other databases to test our findings. Statistics Canada, is the official institute of statistics of the state, and it contains enough studies to carry a similar analysis to the one we have conducted, such as the Classification of Instructional Programs (Statistics Canada, 2016). Studying school choice from another point of view could be interesting: as a compromise between fulfilling the preferences of the households (people will tend to segregate on their own), or making an effort to eradicate the existing segregation on a social, racial, or cultural level. 31 References [1] Bifulco, R., & Ladd, H. F. (2006). School choice, racial segregation, and test-score gaps: Evidence from North Carolina’s charter school program*. Journal of Policy Analysis and Management, 26(1), 31-56. [2] Borgers, A., Oppewal, H., Ponj´ e, M., & Timmermans, H. (1999). Assessing the Impact of School Marketing: Conjoint Choice Experiments Incorporating Availability and Substitution Effects. Environment and Planning A: Economy and Space, 31(11), 19491964. [3] Breton, A. (1978). Nationalism and Language Policies. The Canadian Journal of Economics, 11(4), 656. [4] Breton, A. (1998). Economic approaches to language and to bilingualism. Ottawa: Canadian Heritage. [5] Brochu, P., Deussing, M., Houme, K. & Chuy, M. (2013). Measuring up: Canadian results of the OECD PISA study: The performance of Canada’s youth in science, reading and mathematics: 2012 first results for Canadians aged 15. Council of Ministers of Education, Canada. [6] Calsamiglia, C., Haeringer, G., & Klijn, F. (2010). Constrained School Choice: An Experimental Study. American Economic Review, 100(4), 1860-1874. [7] Christofides, L. N., & Swidinsky, R. (2010). The Economic Returns to the Knowledge and Use of a Second Official Language: English in Quebec and French in the Rest-ofCanada. Canadian Public Policy, 36(2), 137-158. [8] CICIC, 2020. Ministries/departments responsible for education in Canada. Retrieved from the CICIC webpage. 32 [9] Constitution Act, 1982. Retrieved from the Government of Canada webpage. [10] Economic Advantages of Bilingualism. Literature Review (2016). Canadian Association of Second Language Teachers. [11] Farre, L., Ortega, F., & Tanaka, R. (2018). Immigration and the public–private school choice. Labour Economics, 51, 184-201. [12] Fidrmuc, J. (2011). The Economics of Multilingualism in the EU. Research Handbook on the Economics of European Union Law. [13] Freynet, N., & Cl´ ement, R. (2015). Bilingualism in minority settings in Canada: Integration or assimilation? International Journal of Intercultural Relations, 46, 55-72. [14] Ganzeboom, H. B., Graaf, P. M., & Treiman, D. J. (1992). A standard international socio-economic index of occupational status. Social Science Research, 21(1), 1-56. [15] H´ etu, C. (1991). Educational reform and tradition: The case of public and private high schools in Quebec. International Journal of Educational Research, 15(5), 483-497. [16] Lankford, R., Lee, E., & Wyckoff, J. (1995). An Analysis of Elementary and Secondary School Choice. Journal of Urban Economics, 38(2), 236-251. [17] Long, J. S. (1997). Regression Models for Categorical and Limited. Dependent Variables, SAGE Publications, Thousand Oaks. [18] Maddaus, J. (1990). Parental Choice of School: What Parents Think and Do. Review of Research in Education, 16, 267. [19] Manski, C. F., & Wise, D. A. (1983). College choice in America. Cambridge, Mass. u.a.: Harvard Univ. Pr. 33 [20] M¨ uller, S., Haase, K., & Seidel, F. (2012). Exposing Unobserved Spatial Similarity: Evidence from German School Choice Data. Geographical Analysis, 44(1), 65-86. [21] Nakhaie, M. (2000). Social origins and educational attainment in Canada: 1985 and 1994. Review of Radical Political Economics, 32(4), 577-609. [22] OECD. (2012). Programme for International Student Assessment (PISA). [23] Pons-Ridler, S., & Ridler, N. B. (1989). The territorial concept of official bilingualism; A cheaper alternative for Canada? Language Sciences, 11(2), 147-158. [24] Prieto, L. M., Aguero-Valverde, J., Zarrate-Cardenas, G., & Maarseveen, M. V. (2018). Parental preferences in the choice for a specialty school. Journal of School Choice, 13(2), 198-227. [25] Ripley, B. & Venables, W. (2020). Feed-Forward Neural Networks and Multinomial Log-Linear Models. R package, version 7.3-14. [26] Statistics Canada (2020). Table 15-10-0003-01. Population by mother tongue and geography, 1951 to 2016. Retrieved from the Statistics Canada webpage. [27] Statistics Canada (2016). Aboriginal Population Profile. 2016 Census. Retrieved from the Statistics Canada webpage. [28] Statistics Canada (2016). Language Highlight Table. 2016 Census. Retrieved from the Statistics Canada webpage. [29] Statistics Canada (2016). Table: 36-10-0487-01. Gross domestic product (GDP) at basic prices, by sector and industry, provincial and territorial . Retrieved from the Statistics Canada webpage. 34 [30] Statistics Canada (2016). Classification of Instructional Programs (CIP) Canada 2016. Retrieved from the Statistics Canada webpage. [31] University of Ottawa (2020). Compendium of Language Management in Canada (CLMC). Retrieved from the University of Ottawa webpage. [32] Vaillancourt, F. (2012). Official language policies of the Canadian provinces: Costs and benefits in 2006. Vancouver, BC: Fraser Institute. [33] Vega-Bayo, A., & Mariel, P. (2015). School choice in the Basque Country: Public, government-dependent and private schools with different languages of instruction. International Journal of Educational Research, 74, 13-25. [34] Vega-Bayo, A., & Mariel, P. (2018). School Choice across Different Regions of Spain. Revista Hacienda P´ ublica Espa˜ nola, 227(4), 11-36. [35] Vega-Bayo, A., & Mariel, P. (2018). A Discrete Choice Experiment Application to School Choice. Revista Hacienda P´ ublica Espa˜ nola, 230(3), 41-62. [36] Vega-Bayo, A., Sanko, N., & Mariel, P. (2019). Social class segregation in uppersecondary school choice in Japan. Universidad del Pa´ ıs Vasco, Departamento de Econom´ ıa Aplicada III. [37] Yee, T. & Moler, C. (2020). Vector Generalized Linear and Additive Models. R package, version 1.1-3. 35 Appendix I Table 8: Multinomial logit: first model estimation Dependent variable: Private English Public French Private French (1) (2) (3) Mother not working 0.439∗∗∗ −0.292∗∗∗ 0.226∗∗ (0.100) (0.083) (0.109) Two-parent family 0.268∗0.108 0.617∗∗∗ (0.155) (0.107) (0.179) Siblings −0.471∗∗∗ 0.133 0.738∗∗∗ (0.135) (0.108) (0.190) Grandparents 0.187 −0.560∗∗∗ −0.589∗∗ (0.207) (0.170) (0.278) Cultural Possessions 0.483∗∗∗ −0.217∗∗∗ 0.229∗∗∗ (0.057) (0.043) (0.059) Mother Educ. level 0.219∗∗∗ 0.035 0.127∗∗ (0.052) (0.033) (0.051) Father Educ. level 0.149∗∗∗ −0.029 0.055 (0.042) (0.027) (0.042) Family Wealth 0.522∗∗∗ 0.330∗∗∗ −0.104 (0.113) (0.087) (0.148) Occup. Status 0.031∗∗∗ 0.010∗∗∗ 0.016∗∗∗ (0.003) (0.002) (0.003) Home Language not English 0.231 2.413∗∗∗ 0.203 (0.338) (0.175) (0.383) Quebec 1.453∗∗∗ 0.266 1.142∗∗∗ (0.204) (0.201) (0.208) Sasketchwan −0.223 −1.678∗∗∗ −0.099 (0.267) (0.331) (0.263) B. Columbia −0.435 0.109 −0.439 (0.267) (0.202) (0.278) Home Language not English: Quebec −0.668∗1.886∗∗∗ 1.093∗∗∗ (0.377) (0.239) (0.411) Home Language not English: Saskatchewan 0.323 0.145 −0.628 (0.575) (0.432) (0.830) Home Language not English: B. Columbia 0.289 −1.356∗∗∗ 0.265 (0.444) (0.287) (0.509) Family Wealth: Quebec −0.064 −0.752∗∗∗ −0.173 (0.132) (0.102) (0.167) Family Wealth: Saskatchewan −0.153 −0.129 0.070 (0.165) (0.198) (0.206) Family Wealth: B.Columbia −0.160 −0.709∗∗∗ 0.417∗∗ (0.166) (0.147) (0.200) Constant −7.067∗∗∗ −3.636∗∗∗ −6.228∗∗∗ (0.406) (0.252) (0.403) Akaike Inf. Crit. 10,458.460 10,458.460 10,458.460 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 Benchmark: Public English schools. 36 Appendix II Table 9: Binary logit: second model estimation Dependent variable: Public French Mother not working −0.371∗∗∗ (0.069) Two-parent family 0.025 (0.087) Siblings −0.052 (0.084) Grandparents −0.752∗∗∗ (0.133) Cultural Possessions −0.356∗∗∗ (0.034) Mother Educ. level 0.090∗∗∗ (0.028) Father Educ. level −0.022 (0.023) Family Wealth 0.342∗∗∗ (0.113) Occup. Status 0.011∗∗∗ (0.002) Home Language not English 2.744∗∗∗ (0.227) Alberta −0.425∗∗ (0.187) N. Brunswick 0.333∗∗ (0.170) Ontario 1.411∗∗∗ (0.134) Home Language not English: Alberta −0.778∗∗∗ (0.286) Home Language not English: N. Brunswick 2.225∗∗∗ (0.295) Home Language not English: Ontario −1.408∗∗∗ (0.241) Family Wealth: Alberta −0.299∗∗ (0.142) Family Wealth: N. Brunswick −0.407∗∗∗ (0.152) Family Wealth: Ontario −0.288∗∗ (0.119) Constant −3.388∗∗∗ (0.216) Akaike Inf. Crit. 6,444.261 Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 Benchmark: Public English schools. 37