| Gregory Palardy
University of California, Santa Barbara
A comparison of hierarchical linear and multilevel structural equation growth models and their application in school effectiveness research
FINAL REPORT:
This study consists of two parts. The first is a comparison of multilevel growth models in the regression framework with multilevel latent growth curves. It is demonstrated that under certain model restrictions those two methods produce equivalent parameter estimates. It is also shown that multilevel latent growth models are a more flexible platform which affords numerous additional modeling options not available in the multilevel regression framework. The usefulness of those options in the context of educational research is evaluated.
The second part uses a nationally representative sample of high school students to investigate the effects schools have on student learning using multilevel latent growth curve models. Among the findings is that several school process measures, which school site personnel have considerable control over, have sizable effects on student learning. Moreover, those processes mitigate the effects that school structure and student composition have on learning, factors which schools do not have control over. The results of the school effects study are used to develop a model for evaluating the relative effectiveness of schools using the residuals from multilevel latent growth curve models. A comparison of results from this method with alternatives suggests that a longitudinal outcome of student learning is the most fundamentally important aspect of a sound school evaluation procedure. Also important is the use of statistical controls for student inputs. Failing to integrate both of those aspects in the evaluation model can result in highly misleading estimates of the effectiveness of schools. Several policy recommendations are offered.
Back to Funded Dissertation Grants Page |