The use of value-added models for the identification of schools that perform “against the odds”

Value-added (VA) modeling aims to quantify the effect of pedagogical actions on students’ achievement, independent of students’ backgrounds. VA modeling is primarily used for accountability and high-stakes decisions. To date, there seems to be no consensus concerning the calculation of VA models. Our study aims to systematically analyze and compare different school VA models by using longitudinal large-scale data emerging from the Luxembourg School Monitoring Programme. Regarding the model covariates, first findings indicate the importance of language (i.e., language(s) spoken at home and prior language achievement) in VA models with either language or math achievement as a dependent variable, with the highest amount of explained variance in VA models for language. Concerning the congruence of different VA approaches, we found high correlations between school VA scores from the different models, but also high ranges between VA scores for single schools. We conclude that VA models should be used with caution and with awareness of the differences that may arise from methodological choices. Finally, we discuss the idea that VA models could be used for the identification of schools that perform “against the odds”, especially for those schools that have positive VA scores over several years.