Author(s): Dupriez, Vincent; Dumay, Xavier
Pages: 29 p.
In recent years, several authors have described the school composition effect as a methodological artefact, suggesting that it results from two major categories of methodological bias: model underspecification and predictor unreliability. The main purpose of this article is to discuss these methodological considerations and test empirically the impact of model specification on the magnitude of the school composition effect, based on reading performance at primary school in French-speaking Belgium. The results show that the school composition effect remains significant even after controlling for pupils’ initial performance, socio- cultural capital and non-cognitive dispositions, although the effect size vary greatly when these individual parameters are successively introduced. The second objective is to examine covariance between school composition and several organizational variables and their joint effect on school performance. The second set of analyses is intended to question the conceptual nature of the school composition effect, establishing whether it is direct or indirect.