EXAMINATION OF DIMENSIONALITY AND LATENT TRAIT SCORES ON MIXED-FORMAT TESTS

The aim of the present study is to examine various item types utilized to measure success in mathematics in terms of dimensionality and latent trait scores. The data collection instruments utilized in the present study were the student questionnaires and the mathematics achievement tests developed to measure 4th and 8th grade students' mathematics success in TIMSS 2015. It is assumed in the current study that two different dimensions are formed: the combination of MC and CR items, forming the “math ability” and the CR items, forming the “CR ability”. To determine the dimensionality of latent trait of math ability, three different IRT models – unidimensional, within-item and between-items – were used. It was found that the within-item model displayed a better fit, when compared to the unidimensional model. Moreover, the within-item dimensional model showed better fit according to AIC and BIC as well. In the unidimensional and within-item models, the talent parameter predictions were similar. While the effect of the variables of sense of school belonging and students' confidence in mathematics on the primary trait were significant, the home resources for learning variable also had a significant impact within 8th grade.