Analysing Differential School Effectiveness Through Multilevel and Agent-Based Modelling

Authored by Elio Marchione, Nigel Gilbert, Mauricio Salgado

Date Published: 2014-10-31

Sponsors: National Commission for Scientific and Technological Research of Chile

Platforms: NetLogo

Model Documentation: Other Narrative Pseudocode

Model Code URLs: Model code not found

Abstract

During the last thirty years education researchers have developed models for judging the comparative performance of schools, in studies of what has become known as “differential school effectiveness”. A great deal of empirical research has been carried out to understand why differences between schools might emerge, with variable-based models being the preferred research tool. The use of more explanatory models such as agent-based models (ABM) has been limited. This paper describes an ABM that addresses this topic, using data from the London Educational Authority's Junior Project. To compare the results and performance with more traditional modelling techniques, the same data are also fitted to a multilevel model (MLM), one of the preferred variable-based models used in the field. The paper reports the results of both models and compares their performances in terms of predictive and explanatory power. Although the fitted MLM outperforms the proposed ABM, the latter still offers a reasonable fit and provides a causal mechanism to explain differences in the identified school performances that is absent in the MLM. Since MLM and ABM stress different aspects, rather than conflicting they are compatible methods.

Tags

Agent-based modelling Social networks Performance Differential School Effectiveness Peer effects Teacher Expectation Bias multilevel modelling Mechanisms Educational-achievement Teacher expectations