Abstract:
The development and validation of achievement tests have been founded on the classical test model. Unfortunately, the classical test model has been unable to solve adequately a number of testing problems during its history. Consequently, several alternative test models have been suggested. A frequently cited alternative to the classical test model is the latent trait model. The study therefore was aimed at developing and establishing the psychometric properties of a physics achievement test using the one-parameter, the two-parameter latent trait models and the classical test model. Altogether, 2215 male and female senior secondary school physics students participated in the study. The instrument for the study consisted of 60 multiple - choice items on a physics achievement test, developed by the researcher. Reliability and validity for each item and for the whole test were established according to the one-parameter and two-parameter latent trait models. Analysis was performed using PROX and regression techniques on the Microsoft Excel Visual BASIC computer programme. The chi-square goodness-of-fit test was used to investigate how well the one-parameter latent trait model fits the PAT items, while factor analysis was used to establish the unidimensionality of the final version of the PAT items. The result showed that the overall reliability co-efficient (KR-20) of the PAT was 0.89; the fit to the model was good, there was no significant relationship between easiness values obtained from two contrasting score groups. The physics achievement test measured one single trait for physics ability.