Wolfgang Schneider, Andreas Helmke

• A combined criterion involving the regression slopes of pretest-posttest achievement scores and achievement gain scores was used to classify similar types of classrooms. Mathematics achievement differences among 632 fifth graders were analysed in a longitudinal design and explained in a structural equation framework provided by LISREL, separately for four types of classrooms. The results replicated the findings of an earlier study (Schneider & Treiber, 1984) in that the local nature of achievement models could be demonstrated. That is, theA combined criterion involving the regression slopes of pretest-posttest achievement scores and achievement gain scores was used to classify similar types of classrooms. Mathematics achievement differences among 632 fifth graders were analysed in a longitudinal design and explained in a structural equation framework provided by LISREL, separately for four types of classrooms. The results replicated the findings of an earlier study (Schneider & Treiber, 1984) in that the local nature of achievement models could be demonstrated. That is, the structural components of the causal models could not be generalized across the four groups of classrooms. The inclusion of a second grouping criterion (i. e., achievement gainJ proved useful in that a better model fit was always obtained for classrooms with high achievement gains. As a global model test ignoring group and classroom membership did mask the differential validity of the achievement model in the various subgro.ups, the need for multilevel approaches was emphasized.…