THE MIXTURE DISTRIBUTION POLYTOMOUS RASCH MODEL USED TO ACCOUNT FOR RESPONSE STYLES ON RATING SCALES: A SIMULATION STUDY OF PARAMETER RECOVERY AND CLASSIFICATION ACCURACY

摘要: Title of Document: THE MIXTURE DISTRIBUTION POLYTOMOUS RASCH MODEL USED TO ACCOUNT FOR RESPONSE STYLES ON RATING SCALES: A SIMULATION STUDY OF PARAMETER RECOVERY AND CLASSIFICATION ACCURACY Youngmi Cho, Doctor Philosophy, 2013 Directed By: Professor Jeffrey R. Harring George B. Macready Department Human Development and Quantitative Methodology Response styles presented in rating scale use have been recognized as an important source systematic measurement bias self-report assessment. People with the same amount a latent trait may some cases be victims biased test scores due to construct’s irrelevant effect response styles. The mixture polytomous Rasch model has proposed tool deal style problems. This can used classify respondents different into classes provides person estimates that corrected for style. study investigated how well partial credit (MPCM) recovered parameters under various testing conditions. Item responses characterized extreme (ERS), middle-category (MRS), acquiescent (ARS) on 5-category Likert ordinary (ORS), which does not involve distorted use, were generated. results suggested ARS could almost perfectly differentiated from other response-style while correct differentiation between MRS ORS was most difficult attain followed by ERS respondents. classifications more when small proportions within sample. Under simulated conditions where ten-items sample size 3000 there reasonable item thresholds parameter obtained. As structure became complex, increased size, length, balanced mixing proportion needed order achieve level recovery accuracy. Misclassification impacted overall accuracy estimation. BIC found effective data-model fit statistic identifying number this modeling approach. model-based correction score explored up four classes. Problems estimation including non-convergence, boundary threshold estimates, label switching discussed.