Exploiting TIMSS and PIRLS combined data: Multivariate multilevel modelling of student achievement

What is it about?

New paper on the achievement of Italian pupils using large-scale assessment surveys. The analysis is carried out with multivariate multilevel modeling.

Featured Image

Why is it important?

In official TIMSS and PIRLS reports, for any country the outcomes in Reading, Math and Science are analyzed separately by means of multilevel models. We propose a multivariate multilevel approach, where the three scores are treated as a multivariate outcome measured for each student (level 1), where students are nested within classes (level 2). This approach allows us to gain further insights with respect to the univariate analysis, as we can estimate the residual correlations between pairs of outcomes at both hierarchical levels, which is important to make a comprehensive picture of student achievement and educational effectiveness. Moreover, a multivariate model enables to test whether the coefficient of an explanatory variable is identical across outcomes, for example, whether gender differences in achievement are the same for Reading and Math.