Accounting for class effect using the TIMSS 2003 eighth-grade database: Net effect of group composition, net effect of class process, and joint effect

Abstract

In the scientific literature, the debate around the school and class effect is polarized, because the variance between schools and classes is explained either by school and class process or by group composition. The first aim of this article is to shed light on this debate and move beyond it by reviewing qualitative and quantitative studies showing how managerial and pedagogical process are connected to group composition. After that, a technique of models rotation is applied to the TIMSS 2003 dataset of 4 OECD countries to estimate the net effect of group composition, the net effect of class process, and the joint effect of these 2 dimensions. These 4 education systems (Belgium Flemish, England, The Netherlands, and the United States of America) have been selected because they are characterized by a strong between-classes variance. The results suggest that the joint effect of group composition and class processes, like class climate, explains a significant part of the between-class variance. The implications of these results for educational effectiveness research (EER) are discussed.

Notes

1. All the variables are described in detail in Appendix. For more information about the variables measured in TIMSS 2003 survey, see Martin (2005).

2. The variable sex is a dummy variable coded 0 for girl and 1 for boy.

3. Considering that sex is a dummy variable, we used phi-Cramer to compute the associations between sex and the other background variables.

4. Only the correlations between EdExpect and Pared in Flemish Belgium, the USA, and England exceed .40.

5. We averaged those items because a principal components analysis applied on those six items yielded a single-factorial solution in the USA and in Flemish Belgium, and a two-factorial solution with a second factor explaining less than 10% of the total variance in The Netherlands and in England. The average (rather than a factorial score) is used, since the factorial solution varies between the educational systems.

6. For more information about the construction of the indexes in TIMSS 2003 survey, see Martin (2005).

7. Because of the intensity of the correlations between the composition variables—and Multicolinearity problem, the effect of these variables has to be considered as a whole effect, regardless of the differentiated impact of each composition variable.

8. Except for the language spoken at home within the English educational system and for the educational resources within the Flemish one.