What are the Factors of Success at University? A Case Study in Belgium
*Université libre de Bruxelles, Centre for Knowledge Economics (CKE), ECARES (European Center for Advanced Research in Economics and Statistics), e-mail: earias{at}ulb.ac.be
Université libre de Bruxelles, ECARES (European Center for Advanced Research in Economics and Statistics), Institut de Recherche en Statistique and CKE, e-mail: cdehon{at}ulb.ac.be
 |
Abstract |
|---|
 Top Abstract 1 Introduction2 Review of the...3 Data and methodology4 Empirical results5 ConclusionAppendixReferences |
|---|
By using a unique data set containing the entire newly enrolled student population at the University of Brussels, this case study aims to be the first complete analysis of the determinants that influence the student's; path at university in Belgium. We analyse the probability of succeeding the first year at university in Brussels taking into account individual characteristics, prior schooling and socioeconomic background. Our results show that the socioeconomic background of the student influence success in a significant way. More specifically, the mother's; level of education and the father's; occupational activity seem to predominate. We observe also a difference in performance between students coming from different high school programs. Indeed, students coming from one of the two high school systems existing in Belgium's; French Community ("traditionnel" and "rénové"), present non-homogenous results at the end of their first year. In addition and in contrast with some of the literature findings, Belgians and foreigners have the same first year performances if we take into account their socioeconomic environment. Moreover, the same results are obtained when we look at European and non-European students. Nevertheless, when we distinguish foreign students with respect to their level of integration, our analysis shows the existence of a "European elite" that comes to Belgium looking for a diploma and that do much better in their first year than Belgian students.
Key Words: Academic achievement • logit models • socioeconomic factors.
|
1 Introduction |
|---|
|
TopAbstract1 Introduction2 Review of the...3 Data and methodology4 Empirical results5 ConclusionAppendixReferences |
|---|
A child's academic path is the result of a comprehensive set of choices made successively from birth to adulthood. Parents and children make decisions in a given economic environment defined by the government (taxes, public spending and regulations). This implies that some educational choices are not made by the individual himself but by the government and his family. Evaluating to which extent these two agents will influence children's achievement has been widely studied in the empirical literature. Interestingly, the results obtained depend on how academic achievement is defined. These different definitions of success depend on the stage of the academic path at which they are measured. For example, the scores obtained in different kinds of test (reading test, mathematical tests, social test, etc.) or the probability of high school completion are early measures of success whether the number of years of schooling is a global measure at the end of the academic path of each individual. This article focuses on a particular stage (not studied so far): achievement during first year at university. We identify which factors influence success at university through the case study of Belgium's French Community (BFC).
Belgium is a federal state where the communities are competent for the educational system. This french speaking community offers an unique framework for the analysis of success at university: an important part of higher education is financed through public funds so that all universities have very low, common entry fees and no entry barriers (there is no entry exam1). As a result, almost 60 percent of the secondary student population that finishes the general high school system,2 enrolls at university. However, during the first year very high rates of failure and drop out are observed, increasing the cost of publicly financed mass higher education. In this framework, we analyse success at university through the first year because it is considered as an information "key point" about student success.
The problem is that until now the existing studies in Belgium's French Community have either lacked socioeconomic information about the students (Droesbeke, Hecquet and Wattelar 2001), or precise evaluation methods of how the different factors interact (Demeulemeester and Rochat 1995 and Alaluf et al. 2003). With the help of a new database collected by the Université libre de Bruxelles (ULB), this article is a first step towards a complete analysis of students’ path in a Belgian university. This implies that the higher education system in Belgium's French Community is analysed through the student population of the ULB, a case that offers two main advantages. On the one hand, even if all universities have the same entry fee and no entry exam, in 2001, the ULB was the one that recruited the most first year students across the eight existing universities in this french speaking community.3 On the other hand, unlike the rest of Belgium's French Community, Brussels has two different high school systems (the "Rénové" and the "Traditionnel") co-existing at the same time, an unique setting to analyse the influence of the high school system on success at university.
In this framework, we address three main questions. First, in this mass higher education system, does the socioeconomic status of the family still influence the probability of succeeding the first year at the ULB? Second, even taking into account familiy's characteristics, do we observe large differences between students coming from a specific high school program? And third, are these effects the same for natives and foreigners? Our findings suggest that the socioeconomic background of the student clearly influences the probability of succeeding the first year at the ULB. However, even if couples’ characteristics are closely related, the parents do not seem to have the same channel of influence. The educational level of the mother is statistically closer to success than that of the father but when it comes to professional activity, the opposite is observed. In addition, the students that come from the "Traditionnel" system do better during their first year. These effects must be interpreted with caution given that high school choices might be endogenous to the model. Finally, there is no significant difference in success between natives and foreigners4 if we take into account their socioeconomic background. Differences do arise when we look at the "type of immigration" of the foreign students at the ULB, since students that come alone to Belgium to enroll at university are more successful during their first year than Belgian students.
This article is organized as follows: in Section 2 we provide a brief review of the literature's main findings about the socioeconomic determinants of children's general academic achievement. Section 3 discusses the data, the variables chosen and the methodology that will be used to analyse them. Section 4 presents the results of the empirical analysis and Section 5 concludes.
|
2 Review of the literature |
|---|
|
TopAbstract1 Introduction2 Review of the...3 Data and methodology4 Empirical results5 ConclusionAppendixReferences |
|---|
The theoretical framework of the parental decision making process and its influence on children's educational attainment has been studied through the theory on family behavior. In the seminal paper by Becker and Tomes (1976), family is viewed as a production unit that generates utility for all of its members. Their results showed that parents influence their children in three different ways. First, through the endowment of ability transmitted directly to their children at birth. Second, under the assumption that parents care for the success of their children, when they make specific expenditures in order to influence their level of human capital.5 Third, when deciding on other factors than the allocation of resources, like for example location, family structure and fertility that will also affect the environment in which children grow. However, given that family decides not only on goods inputs but also on time inputs, Arleen Leibowitz (1974) introduces a different channel of influence of parental background. The author argues that parents can influence their children's attainment through a more behavioral effect since the quality of time inputs is positively influenced by the educational level of the parents.
This theoretical link between parental decisions (or home characteristics) and children's educational success gave rise to an important line of empirical economic research. Most of the empirical studies concentrate on the analysis of this specific set of explanatory variables. Contrarily, our aim is to do a global analysis of the factors that influence success during first year at university. The problem is that, theoretically, there is an infinite number of channels through which parents can influence their children. Therefore, in order to determine which variables have been identified as the most important to explain academic attainment, we review the empirical international literature on academic attainment.
In the international literature, different approaches can be found as there is no unique definition of children's academic achievement. The most frequently used definitions depend on the stage of the academic path at which they are measured: the scores obtained in different kinds of test (reading test, mathematical tests, social test, etc.) or the probability of high school completion are the early measures; the number of years of schooling is a global measure at the end of the academic path of each individual. Given that in our article we analyse a new dependent variable (success at higher education), it is interesting to test all the proposed variables. For example, the articles using the scores obtained at different tests put forward a common set of variables that explain success. Indeed, Murnane et al. (1981) and Sammons (1995) argue that gender and ethnic origin are important to explain differences in test scores but the skills of the mother also play a significant role in children's achievement. This finding is in agreement with Blau (1999), who shows that income effects are small compared to other individual characteristics such as race, gender or mother and household characteristics.
More global measures of achievement have been used, such as the probability of completing high school or even the total number of years of schooling. In this type of studies, family structure plays a significant role in explaining success. Indeed, Ermisch and Francesconi (2001) and Ermisch, Francesconi and Valin (2003) showed that individuals who experience single parenthood as children, have significantly lower attainments (defined as the number of years of schooling). In the same way, Manski et al. (1992) found that the probability of graduating from high school increases if the student lives in an intact family. Finally, as noted by Haveman and Wolfe (1995) in their survey on American studies, living in a single parent family has a negative impact on achievement, independently of the measurement used. Another common factor of papers using these two types of measures for achievement is that ethnic origin is not significantly associated with neither high school completion nor with the level of schooling attained when background characteristics are included in the model (Haveman and Wolfe 1995; Cameron and Heckman 2001). It is important to notice that some articles include both types of measures of achievement, like the relevant work of Altonji, Elder and Taber (2005) and Evans and Schwab (1995). They are mainly focused on the effects of catholic schools and their results show that attending a catholic school has a positive impact on both high school completion and years of schooling. Some dissimilarities do exist: studies using the number of years of schooling tend to put forward the effect of family characteristics on child outcomes. In almost every study, parental human capital is statistically significant but are both parents equally important? Many authors determine that the number of years of schooling of the mother is more closely related to school achievement than that of the father (Blau 1999; Ermisch and Francesconi 2001; Black, Devereux and Salvanes 2005).
We argue that the existing definitions of academic success may not be able to capture all the factors that influence higher education achievement since it is a particular and later stage on the academic path. Given that further years of schooling requires the completion of a cycle, we cannot distinguish students that wanted to invest on higher education but failed from those that chose not to invest at all. This point is of special interest in the countries were the costs of enrollment are low as in Belgium's French Community.6 According to the Global Higher Education Ranking 2005,7 Belgium's French community has the second cheapest educational costs among several industrialized countries in terms of tuition and costs of books and study materials.8 If we consider enrollment as an experiment, in Belgium's French community the experience is a less expensive private decision. In addition to low fees, there is no entry exam. As a result, almost 60 percent of the secondary student population that finishes the general high school system,9 enrolls at university. However, very high rates of failure and drop out are observed during the first year of university. Thus, this first year is considered as an information "key point" for understanding the determinants of success at university since most of the students that drop out seem to be discouraged during their first year. Studying this new dependant variable may reveal new factors that cannot be captured by the existing measures of achievement.
Previous studies have analysed this question in Belgium's French community (Demeulemeester and Rochat 1995; Droesbeke, Hecquet and Wattelar 2001; Alaluf et al. 2003) but had some serious drawbacks. For example, some use a small and undefined sub-sample increasing the risk of having a sample selection bias that they do not control for. Others do not include any information about students’ socioeconomic background and it only analyses an aggregate rate of success. Finally, conclusions are only based on qualitative analysis since the methodology used is strictly descriptive. Thus, a multiple analysis of academic achievement needs both an appropriate measure of achievement and a large variety of variables explaining success. Several features distinguish our research from the studies discussed above. First, a bigger sample of students for whom we have socioeconomic information about their families and a broad range of variables that accounts for prior schooling. Second, we analyse a new dependant variable and we control for all the variables highlighted as important in the international literature.
|
3 Data and methodology |
|---|
|
TopAbstract1 Introduction2 Review of the...3 Data and methodology4 Empirical results5 ConclusionAppendixReferences |
|---|
3.1 The database
Our research on the determinants of success in the first year at university is based on a new data set granted by the ULB authorities. At the beginning of the academic year 1997–1998, the ULB launched a survey based on a non-compulsory sociological test filled in by newly enrolled students at inscription, in order to get information about the student's socioeconomic background. The experience was repeated during the academic year 2001–2002 and as a result, the ULB created a database that contains 5822 individuals from two different generations.10 Thus, the data offers several research projects: the factors of drop out, the probability of succeeding after the first year, the influence of ability (through grades of the entry exam at the faculty of applied sciences). As said before, we start by studying what we called a "key point" of information, the first year at university. In this case study, the truncated sample that is analysed is exclusively composed of first year students at the ULB that attended a Belgian's French Community high school and that filled in the sociological questionnaire at enrollment (2531 students).
3.2 The variables and some descriptive statistics
In this section, we briefly describe the different types of explanatory variables included in the model: those that account for individual characteristics, those for prior schooling and those that measure socioeconomic factors. We also control for the year of the first enrollment at university (1 if 2001, 0 if 1997) and for the field chosen by the student (human sciences, science, health sciences) to account for differences across fields and across time. Table 1 presents the variables included in the model, as well as some descriptive statistics of the explanatory variables.11
|
Individual characteristics
Student's gender and ethnic origin are two individual characteristics that do not depend on personal choices but that are part of the determinants of children's success. Droesbeke, Hecquet and Wattelar (2001) found that success rates at university in Belgium's French Community are persistently higher for women than for men. Thus, we include a dummy variable that accounts for gender differences (1 if female, 0 if male) and expect to have a coefficient that has a positive sign in accordance with Belgium's French Community characteristics. Indeed, Table 1 shows a difference in success rates of female and male students of 7.69 percentage points. Concerning ethnic origin at the ULB, previous results show that nationality does influence student's success at university (Demeulemeester and Rochat 1995) and indeed, in our sample, the difference in the success rate reaches 9.55 percent between Belgian and foreign students (in favor of Belgian students). However, in the international literature, this debate boils down to whether ethnic origin or nationality differences help to explain educational attainment, even after controlling for differences in socioeconomic background in a multivariate model.
Prior schooling
The empirical model includes three variables that account for student's high school path prior to university. The first one is relative to repetition during high school. Theoretically, in Belgium's French Community a student should finish high school on the academic year that started 17 years after his date of birth12 (student is "on time"). The dummy created controls for the number of years that the student is "late" with respect to his peers of the same generation ("on time", 1 year "late", 2 or more years "late").
Droesbeke, Hecquet and Wattelar (2001) finds that high school repetition is relevant in explaining success at university in the case of Belgian students. We expect this variable to have a negative impact on success increasing with the number of years failed during high school. Second, a variable that takes into account some of Belgium's French Community institutional characteristics (1 Rénové, 0 Traditionnel). In this community, two educational systems co-exist at the same time, the "Rénové" and the "Traditionnel". In the former type, students follow a smaller amount of compulsory hours that in the latter type i.e. optional hours per discipline are much more important. In practice, the "Traditionnel" schools are known as being "difficult" schools and in our sample, students coming from this type of school have a mean rate of success much higher than the students from the "Rénové" (i.e. a difference of 14.05 percent). Third, given that students get to university with large differences on the disciplines they follow during high school and even in intensity for disciplines took in common, the empirical model accounts also for the intensity of the "math profile" (less than 3 hours per week, 4 or 5 hours per week, 6 or more hours per week) and of the "latin and greek profile" (1 if latin and greek classes, 0 if never took latin and greek) chosen by the student.
Socioeconomic background
As stated in the review of the literature, parental choices can influence academic achievement. However, the socioeconomic environment in which children develop cannot be defined in one dimension. This is why we include four different variables that will account for home environment. First, parents’ academic attainment, measured by the higher educational level attained by each parent (primary school, high school, higher education non-university, university). The second variable analyses a different type of choice made by the parents, that is the parents occupational activity ("farmer", "low or medium level employee", high level employee, professor, liberal or independent, unemployed/no profession). We also include a proxy for the level of income of the households that captures if the student paid a reduced fee because of low income (1 if low income reduction, 0 if not). Indeed, several studies showed that money inputs can have an influence on success (Blau 1999). Finally, we include a proxy for household structure: with whom the student has lived before university (both parents, single parent, alone etc.). For all variables, the difference in the success rates between each category and the control dummy variable are displayed in Table 1.
3.3 Methodology and the sample selection bias
In this article, we study success for first year students enrolled at the ULB for the first time. The dependant variable is defined as follows: either you succeed or you fail your first year.13 Thus, success is analysed by a logit model. Let yi be our dependant dichotomous variable such that yi = 1 if the i th individual succeed his first year and yi = 0 if he failed. In this model the probability of success on the first year can be expressed as:
|
|
However, our specification includes some socioeconomic variables that are only available for the individuals that completed the sociological survey. This implies that we have a dependent variable (success or failure in first year) studied in a truncated sample. The seminal work of Heckman (1979) proposed a solution to avoid a potential severe bias of the estimates in the context of attrition. In our case, the missing data does not concern the dependant variable (success is observed for every student on the database) but some independent variables missing for individuals that did not filled in the sociological survey.14 This type of selection is not as serious as selection by the dependent variable but it cannot be ignored. Indeed, it could still affect the randomness of the sample and thus yield biased coefficient estimates. This is why, before interpreting the empirical results of the estimated model, we need to make sure that the truncated sample is still representative of the studied population. In order to facilitate comprehension, in what follows yi will be used to express the random variable as well as the observation.
The special case presented here is close to the one described in Wooldridge (2001). We estimate a partially observed bivariate probit model with sample selection and thus, instead of having a standard discrete dependant variable model, we can rewrite the model as:
|
|
and u follow by assumption a bivariate normal distribution with correlation 
. Note that the problem including endogenous explanatory variables in the equation of success is a very difficult one and for the moment, no answer exists in the literature. Unfortunately, we think that the effect of the high school variables may be due to spurious correlation between the choice of a high school system and unobserved characteristics.15 This is the reason why we use independent variables in these two equations as only the variables for which we are sure about their exogeneity to the model, that is: |
|
The estimates of this binary response model can be obtained through a two-step procedure. The first step is to get estimates of 
by doing a probit regression on the selection equation. The second step requires finding the density of y conditional on x and z = 1 that will be used to compute the likelihood function of the sample and obtain our maximum likelihood estimates (MLE) of β and . According to Wooldridge (2001):
|
|
|
|
|
|
We can use the Wald test under the null hypothesis H0 : = 0 to determine if attrition is random. If we do not reject the null hypothesis of zero correlation, the model estimates can be derived with a traditional probit/logit model. With the help of a specialized econometric program, we computed the ML estimates and the results of the Wald test which do not reject the null hypothesis (p = 0.3013), that is, the hypothesis that the error terms of the two equations are independent.17 Therefore, the truncated sample of students that filled in the sociological survey constitutes a faithful representation of the student population at the ULB and thus, unbiased estimators can be obtained by standard regression methods.
|
4 Empirical results |
|---|
|
TopAbstract1 Introduction2 Review of the...3 Data and methodology4 Empirical results5 ConclusionAppendixReferences |
|---|
This section focuses on the analysis of the empirical results. Our findings deal with two different aspects of student life that influence success at university: family environment and high school choices. Table 2 presents the results of the logit estimation on the sample composed of first year students at the ULB that attended a high school in Belgium's French Community and that filled in the sociological questionnaire at enrollment. The first estimated model (Model 1) focuses on the personal characteristics and family background of the student. The second model (Model 2) reveals the estimated results of the full model, which includes the variables of the first model plus all of the variables that account for prior schooling. As discussed below, the two remaining models (Models 3 and 4) are used to determine which specification is more appropriate to deal with the fact that the couple's socioeconomic variables provide the same information about home environment. For each model, the coefficient estimates are displayed in the first column and the odds ratios in the second column. At the end of this section, we also discuss more in detail the case of immigration and success at university on a larger sample that includes students who attended other high school systems (international or other systems).
|
Family characteristics and children's success
The empirical results show that the family environment influences the probability of success for students at the ULB and we present evidence against the existence of a unique channel of influence: parental choices can affect their children's success in multiple ways (even beyond compulsory schooling). In the first model, the sign of the coefficient associated with having a father with a university diploma is positive (with respect to a father with a primary school diploma). This implies that students whose father has a higher educational level are more successful during their first year at university. However, to evaluate the magnitude of the difference in their performances, it is interesting to interpret the odds ratio. Considering Table 2, we see that a student whose father holding an university diploma has twice the odds of succeeding of a student whose father only attended primary school. The three remaining models will be analysed more in detail given that they take into account the past schooling characteristics of the students.
The second specification contains all of the variables considered in the first model plus the high school choices. The results show that all socioeconomic characteristics and family structure variables become non-significant. This is probably due to a multicollinearity problem. Indeed, as mentioned before, intuition leads to believe that the father's level of education will tend to be very close to the level of education of the mother. In order to evaluate to which extent these two variables are related, we performed a multiple correspondence factor analysis (MCFA) on the socioeconomic variables.18 The results show that the levels of education of each parent are indeed located very close to one another. Therefore, these two variables capture the same kind of information about the cultural environment at home. This implies that one variable could be removed without any significant explanatory loss, leaving us with a more parsimonious model and avoiding the multicollinearity problem. However, to find out which variable to remove from the model, we must evaluate its overall significance as a factor of success. This information is not provided by student test scores in Table 2 given that this test evaluates the individual significance of each level of education with respect to the reference group (for example: the influence on the probability of success of having a mother with an university diploma with respect to having a mother with a primary school diploma). The test we need is one that tells us whether having a mother with any diploma other than primary school has an impact on success or not, i.e. a Wald test:
|
|
The same reasoning applies to another important aspect of the house environment that has to be taken into account: the occupational activity of the parents. The analysis of this variable is identical to that of the educational level of the parents. Indeed, the professional status of the mother and the father are closely related and thus capture the same kind of information, as shown by the results of the MCFA. The difference in this case is that even if they capture almost the same information about the household socioeconomic status, the professional activity of the father is closer to success. According to the results of the Wald test (Table A1 in the Appendix), the mother's professional status is not significant to explain success in the full model, whereas the father's occupational activity is significant. Again, for all of these reasons, we should also exclude the professional activity of the mother (as done for the father's level of education) from our final specification (Model 4).
In the first model, the results of the t-test show that the coefficients related to the father's education and to the mother's profession are significant. Intuitively, it may seem contradictory to exclude them from the model, as it will be done if we follow the conclusions derived from the MCFA and the Wald test. This is why we also estimated Model 3, to show that first of all, the Pseudo-R2 of Model 4 is higher than the one of the third estimated model. This implies that the education of the mother and the professional activity of the father have a larger explanatory power than the father's diploma and the mother's occupation. Second, we also compared the full model to the two restricted models (Model 3 and model 4) by means of a likelihood ratio test. The results of the fourth model show that we do not reject the null hypothesis (at the 5 percent level) that the set of parameters associated with the education of the father and the occupation of the mother are null, leading to the conclusion that the unrestricted model is not more informative than the restricted model (p = 0.2626). The opposite is observed in the case of the third model, given that the likelihood ratio test rejects the null hypothesis that the occupation of the father and the education of the model are null (p = 0.0029).
The educational level of the mother seems to be the most influent on academic achievement while the occupational activity of the father affects his children success at university. In general, most studies conclude that the mother has a stronger effect on the academic path of their children. The question that arises is whether this influence is due to time inputs given by the mother through child care as suggested by Leibowitz (1974) and Murnane et al. (2001) or due to inherited endowments as stated by Behrman and Rosenzweig (2002). As far as professional activity is concerned, the fourth model shows that it is having a father who is a professor that influences the most academic achievement since the odds of succeeding the first year at the ULB are 53 percent higher than those of a student whose father is a workman or a farmer. It is important to highlight that the occupation and the educational level of the parents are significant in the presence of our proxy of the household level of income, which is the dummy for paying a reduced fee at university or not (lowfee). Unfortunately, lowfee does not capture the difference between households with middle and high levels of income but it still captures the influence of having a really low income. Thus, material inputs are not the only channel through which parents can influence their children's success given that in our model, education and occupational activity are significant.
Finally, our findings regarding the influence of the family structure are different from those found in the literature. As mentioned before, in their survey on American studies, Haveman and Wolfe (1995) claim that living in a single parent family has a negative impact on achievement, independently of the measurement used.20 However, the studies that get significant results about family structure are either the ones focused on high school completion or on years of schooling. In our article, students that live with a single parent do not display significant differences in achievement at university with respect to those living with both parents. This result is not surprising since we can assume that students that get to university are often living away from their parents and are becoming more independent individuals. Thus, they should be less affected by the family structure at home when they are at university than during high school (since a large majority of students live with their parents).
Decisions made during high school also matter
The results of Model 4 in Table 2 reveal that success of ULB first year students is also related to their high school path. First of all, in accordance with other Belgian empirical studies, students that have repeated during high school have lower rates of success during the first year at university. In our final specification, a student that is 1 year "late" has 52 percent smaller odds of succeeding than a student that never failed in his academic path. The difference in the odds of succeeding can go up to 77 percent for students that are 2 or more years "late", i.e. those that met multiple failures during their scholarship. This result is particularly important in the case of Belgium's French Community given the extremely high rates of repetition in schools. Indeed, the last report from the education minister revealed that in 2006, only 50 percent of the students that graduate from high school are on time.21 Thus, an important part of student population has a lower probability of succeeding even before getting to university. The second variable capturing high school choices was the type of school where the student attended (between the two types of systems that co-exist in Brussels) and it is statistically significant in explaining success at the ULB. A student that attended a school of the type "Traditionnel" has 81 percent higher odds of succeeding first year at university than a student of the "Rénové" type of school.
Which factors could explain this difference in performance? As mentioned before, the "Rénové" type of school has a smaller amount of compulsory hours leaving the student a broader choice of disciplines to follow and even the frequency of each subject. In order to know the educational profile of the students that attend this particular system, the ULB recorded the number of hours received by each student in disciplines like mathematics and latin and greek. Our final model includes these variables and the results show that both variables are highly significant. Students that received any amount of hours of latin and greek lessons are more successful during their first year (higher odds of succeeding of 80 percent) than students that did not received any. In the same way, a student that attended 6 or more hours of mathematics per week (Strong Math Profile) has more than twice the odds of succeeding first year than a peer that attended 3 or less hours per week. Furthermore, controlling for differences in the mathematical or latin and greek profile of the students brought evidence against the ULB common belief that students in science have higher rates of success than students in other domains. In the first model, we can see that a student enrolled in the domain of science has 49 percent higher odds of succeeding relative to a student enrolled in the domain of human sciences. However, in the other models this effect no longer holds when we account for differences in prior schooling. More precisely, this effect will remain significant until we include the math or latin and greek profile variables.
Nonetheless, in this section, we use a standard logit model that implicitly assumes that explanatory variables are exogenous. This assumption is easily verified for a majority of variables (i.e. gender, socioeconomic factors, family structure) but can be questionable for others variables like the attended high school system ("Traditionnel" versus "Rénové"), the success obtained in past schooling (i.e. the number of years "late") or the type of profile chosen (mathematical or latin). For example, the positive effect of the "Traditionnel" system can be due to spurious correlation between the choice of a high school system and unobserved characteristics (Altonji, Elder and Taber 2005) for the case of catholic school). In human sciences with respect to laboratory sciences, causal effects are more difficult to identify because experiments are generally not randomized in the sense that the individuals exposed to one "treatment" (in this case, attending the "Traditionnel" system) can differ systematically from individuals exposed to the other "treatment" (the "Rénové" system). In our regression, we can suspect that our four high school variables could be endogenous so their estimates could be biased. More research is needed to solve this issue but meanwhile, it is important to notice that the estimates of the remaining variables are stable no matter the specification chosen.
Immigration and higher education
The most important result about student's personal characteristics is that nationality appears as not significant in explaining student's success at university. This is observed when we include in the model the variables that account for the socioeconomic status of the family (on the sample of students from a high school in BFC). As already mentioned, the literature has been trying to find out if differences in school attainment can partially be explained by ethnic origin even if we take account of the differences in socioeconomic backgrounds. In contrast with Demeulemeester and Rochat (1995), we show that this is not the case. Nowadays, being a foreigner at the ULB is not statistically significant in explaining success if we account for the student's socioeconomic status. However, do all foreign students have the same characteristics once they get to university? For example, we could control for the country of origin to take into account ethnic differences. Surprisingly, the results remain unchanged since there are no significant differences in first year performances between European and non-European students22 (Model 1 in Table 3).
|
A solution for this problem could be to identify another key element that differentiates immigrants, like for example the level of integration of the student i.e. to characterize if he is a first or a second generation immigrant. This information is not explicitly available in the database but we created a proxy using the country of residence of the parents of all foreign students. The resulting variable is structured as follows: the student can either be Belgian (control dummy), foreigner with no parents in Belgium (first generation immigrant), foreigner with only one parent in Belgium and foreigner with both parents in Belgium (at least second generation). Replacing the variable Belgian by this more detailed variable implied changing the variable for the high school type, given that the first sample only included students enrolled in a Belgium's French Community high school (the amount of mathematics and latin and greek was only available for students in a school that was part of a Belgium's French Community educational system). The new high school type variable captures if the student went to a "Rénové" high school (control dummy), "Traditionnel" school or an "International or other" school (international being a school in Belgium that is not part of the community's system and other meaning school in another country or graduation through other particular paths).
The second model in Table 3 presents the results obtained using the detailed version of the "Belgian" variable and they reveal that foreigners that are alone in Belgium have higher odds of succeeding than Belgian students. The opposite is observed with foreigners living with both parents in Belgium since they have lower probability of succeeding their first year at university than Belgian students. This result is in agreement with the conclusions from our first estimated model. Given that some foreigners do better and other do worst than Belgian students, if we use an aggregated variable for nationality, it is normal that Belgians and foreigners have on average the same odds of succeeding. Note that the effect of being a foreigner alone in Belgium becomes significant when we take into account socioeconomic differences, meaning that this group has a lower socioeconomic status with respect to Belgian students. Furthermore, being a foreigner with one parent in Belgium has no significant impact in the probability of success. Our results show that foreigners alone in Belgium at university are definitely a type of immigration specific to university: students that come to get their higher education diploma. The motivation to succeed for these students can be different from students that immigrated with their parents and that did not explicitly chose to be there. However, does this effect holds for all types of foreigners? For example, is it likely that European and non-European students alone in Belgium have the same performances? As shown by the third model in Table 3, the "elite immigration effect" is only valid for European students.
Finally, it is interesting to note that our results are in line with the recent study of foreign students in secondary education made by Jacobs, Rea and Hanquinet (2007) using the PISA database for Belgium. One of their main conclusions is that even if we take into account socioeconomic differences, foreign students have poorer success profiles than their Belgian peers. Intuitively, a large majority of foreign high school students do not move to Belgium alone but belong mainly to our groups "foreigner with 1 parent" or "foreigner with both parents in Belgium" and for these groups we observe the same kind of results. Again, we see that the factors that influence success at university need to be analysed separately. A different type of immigration arises at this stage (immigrate to study) and since it influences positively the odds of succeeding, in aggregate it makes being a foreigner not different from Belgians in terms of the probability of succeeding their first year at university.
|
5 Conclusion |
|---|
|
TopAbstract1 Introduction2 Review of the...3 Data and methodology4 Empirical results5 ConclusionAppendixReferences |
|---|
By using a unique dataset containing the entire newly enrolled student population at the ULB, our research aimed to be the first complete analysis of the determinants that influence the university path of the student. We decided to start by studying what we call an information "key point", the analysis of student success on the first year at university. This first year is of great importance because of the high rates of repetition and drop out observed during this year across universities in Belgium's French Community. Several features distinguish our research from the existing literature. First, a bigger sample of students for whom we have socioeconomic information about their families and broad range of variables that account for prior schooling. Second, we analyse a new dependant variable and we control for all the variables highlighted as important in the international literature.
What influences students success on the first year at the ULB? We show that the educational level of the parents has a positive impact on the probability of success. In agreement with the literature, the mother's schooling is more importantly related to their children's success. However, we went a step further and reveal that if we control for the parental occupational activity, the father's profession is more important to student success than the profession of the mother. In any case, we still do not know if the effects are different because of old beliefs about child care (a more educated mother raises the quality of time inputs) or because of inherited abilities. Furthermore, as opposed to the studies on the factors of high school completion, students living in a single parent family do not have a different success profile at university than those in intact families. The difference could come from the fact that at university, students are young adults that are learning to be independent and living with only one parent will not affect their achievement at university. Finally, if we look at Belgians and all foreign students, they have the same success profile if we account for differences in the socioeconomic environment. However, detailing the profiles of foreign students showed that this result comes from the fact that some students belong to an immigrating "European elite". This particular group has higher odds of succeeding their first year than Belgian students, while the opposite is observed for foreign students living with both parents in Belgium. We think that this first generation students present in the higher education system immigrated with a particular goal (getting a degree) and thus face a particular motivation (or pressure?) that the others foreign students do not have.
Prior schooling also appeared to be an important element for student success. Important achievement differences exist between the two types of high school systems in Belgium's French Community: the "Traditionnel" and the "Rénové". The students that attended the former have higher odds of succeeding than the "Rénové" students. In addition, a student that has met several failures during high school has almost twice as high odds of failing their first year of university than a student that finished "on time". This result is of special interest in the case of Belgium's French Community where high school repetition concerns a majority of the student population. Finally, analyzing in detail the "math-intensive" profile of the students erased the previous belief about the better performance of students enrolled in the faculty of applied sciences, given that the difference comes exclusively from their high school profile. Individual endogeneity tests were used to check for the doubtful exogeneity of the prior schooling variables and they revealed that indeed the high school system variable and the "late" variable are endogenous to the model. Then, the estimated effect of these two variables has to be interpreted with caution. We are conducting further research on the endogeneity problem, which clearly needs to be solved in order to identify causality effects of the high school variables. Meanwhile, it is important to highlight that the estimates of the exogenous variables are incredibly stable (as shown by the preliminary sensitivity analysis we have conducted).
Finally, we can see that this first step was somewhat revealing in itself. It also raised new interesting questions that deserve some special attention in the future. For instance, it would be interesting to study the sub-sample of students enrolled in the faculty of applied sciences. Using their scores at the entry exam as a proxy for ability, we could determine if the influence of the parent's educational level is due to inherited ability or to time inputs. In addition, we could analyse the effects of failure on the future academic path of the student, by analysing for example, the factors that influence the drop out and the re-orientation. Similarly, what about the student that succeeds his first year? We could find out if the elements that influence success during the first are the same that could help the student until the completion of his higher education studies. Finally, our results showed that there are high achievement differences between students determined not only by the type of high school attended but also by the options chosen.
|
Appendix |
|---|
|
TopAbstract1 Introduction2 Review of the...3 Data and methodology4 Empirical results5 ConclusionAppendixReferences |
|---|
Table A1 Joint dummy tests of the socioeconomic variables in the full model (Model 2)
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
Footnotes |
|---|
First of all, we would like to thank Françoise Thys-Clément for her support and her precious comments as well as ULB authorities for granting the database. We also thank all our ULB colleagues and the participants of the ECARES seminars in particular Laurent Bouton, Quentin David, Denis Herbaux and Vincenzo Verardi for their insightful comments. We thank the participants of the CESifo Venice Summer Institute and the Belgian Statistical Society for stimulating discussions. Finally, we would like to thank our anonymous referee for his very useful comments. All remaining errors are our own.
1 Except for the faculty of applied sciences. 
2 Note that this fact does not concern students from the technical or the professional high school system.
3 According to Droesbeke et al. (2005), 28.91 percent of all first year students were enrolled at the ULB, with respect to 23.3 percent for the Université Catholique de Louvain (UCL) and 20,59 percent for the Université de Liège (ULg).
4 In the sample composed of students that attended a Belgium's French Community high school.
5 This second influence channel would not exist if capital markets were perfect, if parents knew exactly the initial endowment of their children and if debts could be passed to the next generation. In this setup, parents could borrow money for each child subject to their own ability that they will refund when they get to the labor market. If these economic conditions are not fulfilled, the level of investment in children's human capital will enter the maximization problem of the family and hence, will depend on parental choices.
6 Belgium is a federal state where the organization of the educational system is the competence of each community. In the French community an important part of higher education is financed through public funds that considerably lower the share of the private contribution.
8 The countries included are: Sweden, Finland, The Netherlands, Belgium (Flemish Community), Ireland, Belgium (French Community), Austria, Germany, France, Italy, Canada, Australia, Unites States, United Kingdom, New Zealand and Japan.
9 Does not concern students from the technical or the professional high school system.
10 Repeaters are excluded i.e. only the freshman enrolled for the first time were taken into account.
11 By a confidentiality clause agreement with the ULB authorities, we can only publish variations in rates of success not the actual levels. That implies that no descriptive statistics of the dependant variable will be presented.
12 There are 12 years of compulsory schooling that starts at the age of 6.
13 Given that during first year, students may take some time to adapt to the new educational system and so may have a second session but to succeed anyway, we decided for now not to detail the dependant variable on whether a student succeeded with a second session or not.
14 This does not mean that the student chose not to enter university, this only means that the student chose not to fill in the survey at inscription.
15 See Section 4 for more details of this matter.
16 Note that the non-singularity of the information matrix is sufficient to obtain locally identified parameters for partially observed bivariate probit model (Poirier, 1980). Then the identification problem is solved as soon as the exogenous variables exhibit sufficient variation over the sample, which is the case in our study.
17 We have also performed the Wald test excluding the instrument (Brussels) of the second equation, and the conclusion is the same (p = 0.4230).
18 The graphical representation of the MCFA can be found in the Appendix.
19 The results for the joint tests can be found in the Appendix (Figure A1).
20 For more information on family structure, see also Ermisch and Francesconi (2001) and Ermisch, Francesconi and Pevalin (2003) or Manski et al. (1992).
21 Ministère de la Communauté française and l'Entreprise des Technologies Nouvelles de L'information et de la Communication (2006), "Les indicateurs de l'enseignement", Ministère de la Communauté française, pp. 44.
22 We also checked for ethnic origin by controlling for the continent of origin of the student and in the same way, we found no significant difference in first year performances.
|
References |
|---|
|
TopAbstract1 Introduction2 Review of the...3 Data and methodology4 Empirical results5 ConclusionAppendixReferences |
|---|
-
Alaluf M, Imatouchan N, Marage P, et al. Les filles face aux études scientifiques. Réussite scolaire et inégalités d’orientation. (2003) Editions de l’Université de Bruxelles.
Altonji JG, Elder TE, Taber CR. Selection on Observed and Unobserved Variables: Assessing the Effectiveness of Catholic Schools. The Journal of Political Economy (2005) 113(1):151–84.[CrossRef]
Becker GS, Tomes N. Child Endowments and the Quantity and Quality of Children. The Journal of Political Economy (1976) 84(4). Part 2: Essays in Labor Economics in Honor of H.Gregg Lewis, S143–62.
Behrman JR, Rosenzweig MR. Does Increasing Women's Schooling Raise the Schooling of the Next Generation. The American Economic Review (2002) 92(1):323–34.[CrossRef][ISI]
Black SE, Devereux PJ, Salvanes KG. Why the Apple doesn't Fall Far: Understanding Intergenerational Transmission of Human Capital. The American Economic Review (2005) 95(1):437–49.
Blau DM. The Effect of Income on Child Development. The Review of Economics and Statistics (1999) 81(2):261–76.[CrossRef][ISI]
Cameron SV, Heckman JJ. The Dynamics of Educational Attainment for Black, Hispanic, and White Males. The Journal of Political Economy (2001) 109(3):455–99.[CrossRef]
Demeulemeester J.-L, Rochat D. Impact of Individuals Characteristics and Sociocultural Environment on Academic Success. International Advances in Economic Research (1995) 1(3):278–87.[CrossRef]
Droesbeke J-J, Hecquet I, Wattelar C. "La population étudiante: description, évolution et perspectives". Bruxelles: Editions de l’Université Libre de Bruxelles and Editions Ellipses (2001).
Droesbeke J-J, De Kerchove A-M, Lambert J-P, et al. "Enquête sur les trajectoires étudiantes a l’entrée de l’enseignement supérieur de la Communauté Française de Belgique", Colloque francophone sur les sondages (24–27 may), Societé Française de Statistique (2005).
Ermisch J, Francesconi M. Family Matters: Impacts of Family Background on Educational Attainments. Economica (2001) 68(270):137–56.[CrossRef][ISI]
Ermisch J, Francesconi M, Pevalin DJ. Parental Partnership and Joblessness in Childhood and their Influence on Young's People Outcomes. Journal of Royal Statistical society (2003) 167(1):69–101.
Evans WN, Schwab RM. Finishing School and Starting College: Do Catholics Schools make a Difference? The Quarterly Journal of Economics (1995) 110(4):941–74.[CrossRef]
Haveman R, Wolfe B. The Determinants of Children's Attainments: A Review of Methods and Findings. Journal of Economic Literature (1995) 33(4):1829–78.[ISI]
Heckman JJ. Sample Selection Bias as a Specification Error. Econometrica (1979) 47(1):153–62.[CrossRef][ISI]
Jacobs D, Rea A, Hanquinet L. "Performances des élèves issus de l'immigration en Belgique selon l'étude PISA: une comparaison entre la Communauté française et la Communauté flamande". (2007) Bruxelles: Fondation Roi Baudouin.
Leibowitz A. Home Investment in Children. The Journal of Political Economy (1974) 82(2). Part 2: Marriage, Family Human Capital and Fertility, S111–31.
Manski C.F, Sandefur GD, McLanahan S, et al. Alternative Estimates of the Effect of Family during Adolescence on High School Graduation. Journal of the American Statistical Association (1992) 87(417):25–37.[CrossRef][ISI]
Ministere de la Communaute Francaise and l Entreprise des Technologies Nouvelles de l Information et de la Communication. "Les indicateurs de l’enseignement". (2006) française: Ministère de la Communauté.
Murnane RJ, Maynard RA, Ohls JC. Home Resources and Children's Achievement. The Review of Economics and Statistics (1981) 63(3):369–77.[CrossRef][ISI]
Poirier DJ. Partial Observability in Bivariate Probit Models. Journal of Econometrics (1980) 12(2):209–17.[CrossRef][ISI]
Sammons P. Gender, Ethnic and Socioeconomic Differences in Attainment and Progress: A Longitudinal Analysis of Student Achievement Over 9 Years. British Educational Re-search Review (1995) 21(4):465–85.
Usher A, Cervenan A. Global Higher Education Ranking 2005 (2005) Toronto: Educational Policy Institute.
Wooldridge JM. Econometric Analysis of Cross Section and Panel Data (2001) The MIT Press, Cambridge, MA.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  F. van der Ploeg and R. Veugelers Towards Evidence-based Reform of European Universities CESifo Economic Studies, June 1, 2008; 54(2): 99 - 120. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
G. Kempkes and C. Pohl Do Institutions Matter for University Cost Efficiency? Evidence from Germany CESifo Economic Studies, June 1, 2008; 54(2): 177 - 203. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||