Assessing the Measurement Invariance of Factors that are Related to Students' Science Achievement across Gender in TIMSS-R Turkey Sample

The purpose of this study is to investigate the factors that are related to students' science achievement and assessing the measurement invariance of these factors across gender. This study examined the TIMSS data for Turkish students with the sample size of 7841 through the analysis of Structural Equation Modeling (SEM). First, it is designed a model over the factors from student questionnaire affecting student's science achievement with Structural equation modeling. It was understood that the designed model's goodness fit index output is acceptable. Then four forms of invariance was assessed for each latent variables progressively and assessed MI. The analyses basically begin by fitting a proposed model to the data for each sample considered separately with none of the parameters constrained to be equal across groups(configural invariance). This unconstrained model serves as the baseline model. Subsequently, in a hierarchical fashion, more stringent constraints are placed on the model by specifying the parameters of interest to be constrained across the groups (e.g., factor loadings, factor intercorrelations, error variances). The variables were examined using comparative fit index (CFI) difference test between the more restrictive invariance form and the basic form to determine whether the model and the individual parameter estimates are invariant across the samples. The MI results revealed that there were invariance problems across gender in TIMSS. All latent variables at least have metric invariance, but none of them have strict invariance across groups.