Abstract
The proportion of low-achieving children in a class can affect the progress of individual pupils in that class. Having a large proportion of low achievers in a class could slow down growth in cognitive achievement but, might also boost such growth, due to the effects of specialist teaching geared to low achievers' needs. In a longitudinal study of 8,684 pupils aged 9 – 12, cognitive and social-emotional development were analysed. In classes with a comparatively large number of pupils with migrant parents, the increase in pupils' mathematics achievement was smaller. In classes with a comparatively large number of pupils with parents with a low level of education, the increase in pupils' achievement in Dutch language proficiency was smaller. No relationship was found between class composition and either well-being at school or academic self-concept. The effects of class composition appeared to be differential for different groups of pupils.
Notes
1. The research was supported financially by the Netherlands Organisation for Scientific Research (NWO).
2. Pupils are defined as migrant for the purposes of this study, if at least one of their parents was not born in The Netherlands.
3. As different maths tests were used for the pretest and the posttest, a conversion formula was used to equate both scores. Nevertheless, the results might have been distorted by the change of test. Therefore, we conducted the analyses for maths a second time, using data from the two most recent measurements in the PRIMA cohort study (different pupils from those in the sample of our study, but again grade 4 and 6), in which only one maths test is used (from the National Institute for Educational Measurement or Cito). Most of the current findings were replicated.
4. Some pupils were taught in classes combining two or more grades. For these pupils, the class composition was determined based solely on their classmates in the same grade (grade 4). Results of separate analyses for pupils taught in combined-grade classes and pupils in single-grade classes turned out to be similar.
5. A model with change scores is, in fact, similar to a regression model in which the regression weight of the pretest is fixed at the value 1.
6. The deviance of the model being compared was subtracted from the deviance of the reference model. This difference is χ2-distributed, with the difference in degrees of freedom as new value for degrees of freedom.
7. Of course, the extreme points of the x-axis, 0 and 100% do not apply, as these classes contained no migrant pupils and no Dutch background, respectively.