Abstract:
The objectives of this research were 1) to analyze of rater errors in problem-based learning (PBL) using many-facet Rasch model, 2) to develop and validate the quality rating causal model in PBL, 3) to test the invariance of the quality rating causal model in PBL across tutor-, peer- and self-raters, and 4) to develop and test the quality rating model in PBL. The sample consisted of instructors and students of the institutes of higher education providing PBL. Research instruments were a set of rater characteristics questionnaires and a tutorial-based assessment form. Descriptive statistics, MFRM, SEM, MI, ANCOVA, and MANCOVA were employed to analyze the data. Research results were as follows: 1. In summary, three rater groups tend to have very little rater errors (Measure in Logits ranged from -0.75 to 0.47). Peer-raters tend to have the most lenient rater while tutor-raters tend to have the second severe raters next to self-raters. All ratings were considered accuracy (Infit MNSQ ranged from 0.97 to 1.05). 2. The quality rating causal model in PBL was comprised of the following two latent variable, such as 1) rater context measured from five observed variables consisting of rater goals, ability for rating, conscientiousness, raters motivation and accountability, and 2) quality rating measured from three observed variables consisting of perception of PBL standards, comparison process, and rater error. The model was valid and fit to the empirical data (chi-square = 22.444, df = 19, p = .263, GFI = 0.985, AGFI = 0.971 and RMR = 0.035). 3. The quality rating causal model in PBL indicated invariance of model form among different rater groups, namely tutor-raters, peer-raters, and self-raters, but the model indicated variance of the factor loading of each indicators. 4. After a try out session of the quality rating model in PBL which was made available on manuals, it was found that the rater error of the experimental group was less than that of the control group at the statistically significant level .05.