Evaluating the Contribution of Schools to Pupils' Progress in Educational Outcomes through a Bivariate Multilevel Model

In the last few decades, school value-added studies have largely demonstrated the effects of socio-economic and demographic characteristics of schools and pupils on standardised tests performance. Traditionally, contextualised value-added models (CVA) have been employed to assess variation arising from the schools and pupils in standardised tests of Mathematics, Language, Science, etc. in separate univariate models. However, recent studies show analyses of schools’ and pupils’ performance significantly benefit from additional model complexity, including the specification of unexplored levels of variation, either nested or non-hierarchical, such as between classrooms or teachers, between neighbourhoods and local authorities, as well as carry-over effects from primary schools.

Using data from the 2004 and 2006 SIMCE database (Chilean National Pupil Database), an extended 5-level bivariate cross-classified CVA model estimating simultaneously the variation in Mathematics and Spanish Language between primary schools, between classrooms (within secondary schools) and between local authorities was implemented. This is in addition to the variation between pupils and secondary schools. This extended CVA model for progress in Mathematics and Spanish Language proved effective in reducing bias present in the traditional univariate value-added models, while controlling for the correlation between the two subjects. The model also controls for the fixed effects of prior attainment, household income, gender, grade repetition, school type and average school socio-economic status, as well as the random effects of prior attainment and gender.

This research intends to contribute to school value-added modelling by analysing additional relevant structural effects and by considering the relationship between two of the most relevant educational outcomes. The analysis of these data shows that using school averages or even 2-level CVA model is incomplete for the purposes of informing school-level effects, and therefore parental choice and school accountability due to overestimation of school effects and pupils' heterogeneity.