School tracking and educational inequality: a comparison of 12 education systems in Switzerland

Abstract

Abstract Using data from the super-sample of the 'PISA Suisse' 2003 assessment, this article examines the relationship between the characteristics of education systems (made up of homogeneous or heterogeneous tracks) and their consequences in terms of effectiveness and equity. Our results indicate that it is not so much the official structure of tracks as the ways in which tracking is really organised – and, in particular, the degree of segregation that tracking methods involve – which make it possible to explain inequalities among students. They also show that some education systems are more inegalitarian than others and that the factors leading to inequalities can vary significantly from one Swiss canton to another. Finally, through multilevel analyses, we demonstrate that when the individual and aggregate characteristics of students are taken into account, the type of track a student attends is of only limited significance for explaining inequalities, which suggests that the effects of tracking are in fact linked to the nature of the population educated in the tracks. Acknowledgements This paper was translated by Ashley Riggs, University of Geneva. We thank the Federal Office of Statistics (OFS – Office Fédéral de la Statistique) for providing us with the data from 'PISA Suisse' 2003. Notes A reform is currently underway. The 15 cantons that have ratified the 'HarmoS Concordat' have until 2015 to harmonise the structures and objectives of compulsory education. Switzerland is made up of 26 cantons with a total population of about 7.5 million. The PISA survey in Switzerland has two parts. The first is an international assessment of a sample of 15-year-old students, the same as the assessment conducted in all participating countries. In addition, there is a 'super-sample' which makes it possible to compare the different cantons and which involves students in 9 ème (the ninth year), or the last year of lower secondary school (students between 13 and 15 years old). The latter assessment serves as the basis for our analysis. Tracking refers to the practice of sorting students into different classes according to their level of competence. It may also be referred to as 'streaming'. A precise description of the modalities of sample construction, the administration of tests and questionnaires and the creation of the databases are provided in the following publications: OECD (Citation2004, Citation2005a) and OFS/CDIP (Citation2005). More detailed information about the construction of estimated scores in the PISA survey can be found in the following publications: PISA 2003 Data Analysis Manual (OECD Citation2005a) and PISA 2003 Technical Report (OECD Citation2005b). Analyses have been conducted using the scores in reading and in science and the results are very similar to those obtained for mathematics. We use the term 'ability grouping', but it may also be referred to as 'setting'. In cantons with an integrated system (Jura and Ticino), the degree of social differentiation of ability groups is calculated. In cantons with an integrated system (Jura and Ticino), the degree of academic segregation of ability groups is calculated. The proportion of variance explained by the cantonal programme (Rho * 100) is calculated by dividing the level-2 variance by the total variance (level 1 + level 2). In model 1, for example, Rho is equal to 3500/(4362.3 + 3500) = 0.445. All the continuous variables (individual ESCS, average ESCS by track, age, average age by track) are standardised (average 0, standard deviation 1). Introducing the proportion of girls per track in the multilevel model showed that this variable is not significant for explaining scores. The decrease in deviance (–2 log L) makes it possible to test the significance of a regression model in relation to another, less complete model. The decrease in deviance follows a Chi2 distribution with n degrees of freedom, n representing the number of additional parameters to estimate when switching from one model to another. For example, between model 3 and model 4, the decrease in deviance is equal to 6 (200,167–200,161) for three degrees of freedom (the variable 'track' added to model 5 has three parameters). The Chi2 distribution shows that the decrease in deviance should reach at least 7.83 to be significant at the 95% confidence level.