Educational leadership and pupil achievement: The choice of a valid conceptual model to test effectsin school effectiveness research

Abstract

School effectiveness research often uses multilevel models in which only direct effects of characteristics of schools on pupil achievement are modelled. Recently, more attention is given to conceptual models that assume indirect and antecedent effects. In this paper, we elaborate on these models and show that the findings from school effectiveness research are influenced by the conceptual model underlying the analyses. We do this by focusing on the effect of “integrated leadership” on 2 outcome measures: mathematics achievement and mother tongue achievement. Four different conceptual models will be tested by means of multilevel structural equation modelling. Therefore we use data from a school effectiveness research in Flanders (850 fourth graders and 847 sixth graders within 47 schools). The results of these analyses show that the conclusion whether integrated leadership has an effect or not on pupil achievement, is dependent on the choice of the conceptual model.

Notes

1. For a more extensive description of conceptual models about the relationship between goals, leadership, and achievement, we refer to Hallinger and Heck (2002).

2. In Belgium, there are three official languages (Dutch, French, and German). Therefore, the variable “language spoken at home” is not as such a good proxy-variable for the fact that a pupil comes from an ethnic minority group.

3. This is the model fit for the most parsimonious antecedents model. Testing the model that included all possible relationships between context factors and integrated leadership and academic climate, revealed that for only one of these relationships the parameter estimate differed significantly from zero. Given the fact that we are in fact not really interested in the best fitting model but in the comparison between estimates within different models, we describe this more parsimonious model as the best-fitting model. This is done because most researchers would take the same step and use this data-driven parsimonious model as the best-fitting model.