مرکز منطقه ای اطلاع رساني علوم و فناوري - Examining Measurement Invariance with Different Methods: Example of Pisa 2012

Abstract :

In this study, it is aimed to examine results about measurement invariance of the same data set by using different methods in the way of statistical assumptions. Also the methods requiring the normality assumption and the methods not requiring this assumption were examined whether they show different results or not when the normality assumption cannot be supplied. In accordance with this aim, analyses were carried out by using mean and covariance structures invariance analysis under roof of the structural equation modeling and multi-group latent class analysis under roof of the latent class analysis on the data set Openness for Problem Solving Scale consisting of five items, subscale of PISA 2012 of Turkey and Finland samples. For mean and covariance structures method, Lisrel 8.72 program and for the multi-group latent class method Latent Gold 5.1 program were used. Steps of stepwise testing method starting with configural invariance and ending with testing strict invariance step were followed for mean and covariance structures invariance method. After determining number of class, heterogeneous, partial homogenous and homogenous models were tested for multi-group latent class analysis. Before the analyses, normality of each item was examined by using Kolmogorov Smirnov test. None of the items demonstrated normal distribution for all subsamples and the whole group. According to the results of mean and covariance structures invariance analysis requiring normality assumption, strict invariance was accepted. In other words measurement invariance was supplied. But for multi-group latent class analysis not requiring normality assumption, partial homogenous model was accepted. Partial homogeneous model is equal to weak invariance in steps of mean and covariance structures invariance analysis. According to the results, findings of measurement invariance changed when methods differing from the point of assumptions were used. In this context, certainly the assumptions considered necessary for methods must be examined and choosing the suitable method must be considered by bearing in mind the structures of latent and observed variables.