Data and Method

In order to test these research hypotheses, cross-national comparative micro-level data as well information on the two macro-level variables thought to moderate the impact of parental education was needed. As is most often the case with secondary data not expressly designed to answer a given research question, I had to incur a set of trade-offs, as described below.

Individual-Level Data

The individual-level data are taken from the ESS, which is a large-scale cross-national survey financed by the European Science Foundation. It endeavours to track attitudes in 30 (mostly) European countries and has been carried out every two years since the year of 2002. In the following analysis, the first five waves (from 2002 to 2010) have been pooled to obtain sufficient sample sizes and to make it possible to compare both between higher education systems and over time (see ESS 2010, 2011). Non-OECD members were excluded, as well as countries that did not participate in the majority of the survey rounds (Italy and Turkey), leaving a sample size of 22.

As indicated in the previous section, I am interested in estimating the probability of university enrolment after secondary school. Accordingly, the population of the analyses is limited to the five-year age group after the official age of leaving secondary school in each country of the sample. By focusing on this age group, the goal is to adequately capture the competing options presented to young people after completing secondary schooling.

The dependent variable, therefore, would ideally be a dummy indicating whether the survey respondent at the time of the interview was a student in higher education. Perhaps surprisingly, most cross-national surveys including the ESS do not contain an item indicating university enrolment. I was therefore forced to deduce the information from two auxiliary items. First, I relied on a question asking respondents what their main activity was. Those who answered ‘education’ (as opposed to paid employment, apprenticeships or unemployment) to this item and at the same time had already obtained the qualifying degree in secondary education [as indicated by International Standard Classification of Education (ISCED) scheme] were coded as being higher education students. The dependent variable is thus a binary variable dubbed Student.[1] This coding procedure may confound the analysis to a degree, but it does clearly distinguish between entering vocational or tertiary education, thus eliminating the most obvious source of bias. As described, the expectation is that the propensity to take up higher education is first and foremost structured by the level of parental education. While many studies exclusively focus on the father’s education, my own preliminary analyses suggest that the level of education obtained by the mother also significantly and independently contributes towards study propensity.[2] Therefore, an index comprised of the highest level of education of both parents, again as indicated by the ISCED scheme, was constructed. In this coding scheme, zero points were given to parents who obtained a higher education degree. Consequently, for the completion of upper secondary education, one point was allocated; for the completion of lower secondary schooling, two points were given; and for respondents whose father or mother do not hold a degree, the maximum score of three points was assigned. Results for both parents were then added to each other, resulting in an index ranging from 0 (both parents hold higher education degrees) to 6 (neither parent holds a secondary degree). This variable—called Parental Education—is the central independent variable of the analysis. Because the effect of the different combinations of parental education may be non-linear, this variable in all but one model is treated as categorical, meaning that coefficients have to be interpreted in relation to a predefined reference category—which in this case means the value of the index equals 0. Overall, diminished odds for the propensity to study are expected for all subsequent values of the index. Moreover, effect sizes should increase as higher values are compared against the combination of both parents holding a higher education degree.

In addition to the central independent variable, a set of control variables was included in the models. First of all, dummy variables for each year under investigation (2002, 2004, 2006, 2008 and 2010) were introduced as covariates. This might have captured systemic differences between years, but was mainly done to absorb unobserved heterogeneity associated with each respective year. This is especially important with regard to the macro-level variables introduced below and is designed to inspire confidence that their effects are not artefacts of concurrent developments over time. In addition, the migration background of an individual may be related to social background and have an independent impact on the propensity to study. The concept of migration background is captured by two variables. The first— Foreign Born—indicates whether a respondent was born in a country other than the one she is living in (= 1). The second—Migration Background—harkens back to the importance of parental status and differentiates between both parents having been born in the respective country (= 0), one parent born elsewhere (= 1) and both parents as immigrants (= 2). Finally, a dummy indicating the gender of the respondent (Female = 1) is included.

  • [1] Individuals who at the time of the survey indicated they already held a higher education degreewere also coded into the Student category.
  • [2] In fact, in multilevel random intercept models, effect sizes and significance levels of father’s andmother’s education levels as explanatory variables were almost indistinguishable.
 
Source
< Prev   CONTENTS   Source   Next >