Multilevel and agent-based modelling in the analysis of differential school effectiveness

Multilevel Models (MLM) have pioneered the analysis of hierarchical data, with two or more levels. Agent-Based Models (ABM) are also used to analyse social phenomena in which there are two or more levels involved. This paper addresses the integration between MLM and ABM. To provide a basis of comparison, we focus on differential school effectiveness analysis, where MLM has been well studied, using data from the London Educational Authority's Junior Project. A MLM is fitted and an ABM of pupils' educational attainment using a social network structure is built. We reports the results of both models and compare their performances in terms of predictive power. Although the fitted MLM outperforms the proposed ABM, the latter still offers a reasonable fit and provides a causal mechanisms to explain differences in school performance that is absent in the MLM.