There has been a general belief in school effectiveness research that schools have a larger impact on their students growth than on their students outcomes at a certain point in time. This belief emanates mainly from the research results in which the school effect on student initial status for mathematics has been found to be about three times less than the school effect on learning rates or students progress over time. Several studies have prompted growth in student outcomes over time to gain great acceptance among many educational effectiveness researchers as the most appropriate criterion for assessing school effectiveness. The investigation of such changes in students outcomes has dramatically boosted the number of longitudinal studies in educational effectiveness researchin the last two decades. In addition to this, researchers now understand that cross-sectional designs underestimate the impact of schools and that these designs do not provide the proper framework for studies on school effectiveness. The use of repeated measures data make multilevel growth curve models an invaluable statistical tool in educational research. This is because a multilevel growth curve model estimates changes in student outcomes more accurately by taking into account the hierarchical nature of the data. Befitting results are not only appealingto researchers but also to policy makers and parents who both want a meticulous education for their citizens and children respectively. The main aim of this dissertation is to improve the statistical methods applied by educational effectiveness researchers in order to have more credible results. In this context, school effect estimates from traditional methods and the proposed methods of this dissertation are compared to argue persuasively for the need for more advanced techniques whenusing growth curve models. Such techniques will not only be applicable to educational effectiveness research in but to educational research as a whole and all other research fields interested in growth curve modelling. The school effect estimates on student status and student growth areused for different types of student outcomes like well-being, mathematics, and language achievement. Manuscript 1 defines clearly how the school effect on students growth can be estimated using multilevel growth curve models with more than two levels. It also shows how themanner of coding time affects these estimates. Manuscript 2 introduces techniques to properly handle multilevel growth curve models with serialcorrelation at higher levels beyond level 1, while Manuscript 3 introduces a new multilevel growth curve model which can be used to model growth data with two or more levels of serial correlation simultaneously. Because most studies of school effects on students growth have focused only on one effectiveness criterion, which is problematic given that schooleffects are only moderately consistent over different criteria. Moreover, the consistency issue has seldom been studied through multivariate growth curve models; Manuscript 4 introduces a model that can handle multivariate multilevel growth data with an unequal number of measurement occasions. Data from the LOSO-project (the Dutch acronym for Longitudinal Research in Secondary Education) and the SiBO-project (the Dutch acronym for School Career in Primary School) are used to answer theresearch questions of this dissertation. The main software used is SAS 9.2, MLwiN 2.02 and Mplus 6.1. This dissertation shows clearly how the choice of a time coding affects school effect estimates and their interpretation. It also recommends that the choice of a time coding should not only be based on the ease of interpretation andmodel convergence. The results show that school effects on students well-being and language achievement in secondary school are greater for student growth than for student status. This work also indicates that the common assumption of serially uncorrelated level 1 residuals usually fails and therefore the need for appropriate modelling of this serial correlation is invaluable. These results demonstrate how modelling of serially correlated residuals at level 1 or level 2 has a huge payoff on schooleffects estimates. Because of the increasing popularity of multilevel growth curve models as a flexible tool for investigating longitudinal change in students outcomes, this study investigates some covert issues inmethodology resulting from repeated measures data structure. A complex double serial correlation multilevel growth curve model is developed andthe results of this model show great improvement in school effects estimates compared to those of models without double serial correlation correction. This dissertation also investigates the school effects on pupils growth in both mathematics and reading comprehension (and their relation) in primary schools taking previous changes in mathematics into account through a bivariate transition multilevel growth curve model. The results show that stronger growth in mathematics tends to associate with stronger growth in reading comprehension. Earlier growth in mathematics isalso found to predict subsequent growth in reading comprehension.
展开▼