Undergraduate Physics Test Dimensionality and Conditional Independence: Perspective from Latent Traits Model Package of R Language

The National Open University of Nigeria (NOUN) physics test items developed to assess test-takers proficiency in elementary mechanics, heat, and properties of matter (PHY 101) are intrinsically multidimensional. However, we observed that the current method used by the University to score test-takers proficiency in the course is premised on classical test theory, which has been criticized in the literature for its limitations, and a measurement model for unidimensional tests. Consequently, using an inappropriate model to model test-takers responses to items would adversely affect their true proficiency in the course. This study assessed the dimensionality and conditional independence of physics test items of NOUN using the latent traits model (ltm) package in the R language. Ex-post facto research was used as a non-experimental design. Test-takers responses to the thirtyfive PHY 101 multiple-choice items across NOUN study centers in the thirty-six states of Nigeria were retrieved from the Directorate of Examination of Assessment (DEA) and used for the study, with empirical reliability of 0.80. A modified parallel analysis test and Yen Q3 statistics implemented in R language software, version 4.0.2, were used to analyze 978 test-takers responses. Study results revealed that NOUN physics items had more than one dominant dimension to explain test-takers performance. In addition, residual correlations of ten item pairs were severely over 0.20. Therefore, PHY 101 is intrinsically multidimensional. Multiple latent variables determine test takers' responses. To ensure the appropriateness of the test items developed by the lecturers, we recommend setting up a psychometric unit within the DEA.