Ph.D. Dissertation Defense Meeting

Name: Eunbee Kim

Date: Friday, July 5th, 2024

Time: 11:30 AM - 1:30 PM

Location: Link

Dissertation Chair/Advisor:

Dr. Susan Embretson—School of Psychology, Georgia Institute of Technology

Dissertation Committee Members:

Dr. James Roberts—School of Psychology, Georgia Institute of Technology

Dr. Rick Thomas—School of Psychology, Georgia Institute of Technology

Dr. Mark Himmelstein—School of Psychology, Georgia Institute of Technology

Dr. Michael Hunter—Human Development and Family Studies, Pennsylvania State University

Title: Applications of the Partial Credit Model (PCM) Accounting for Extreme Response Styles (ERS) with a Constrained Weight Parameter across Multiple Scales

Summary: It is worth noting that the structural dependence between Extreme Response Style (ERS) and the measured trait is intrinsic to measurements using rating scales. Respondents with high or low trait levels tend to endorse more extreme options compared to those with middle trait levels. Polytomous models with thresholds, such as the partial credit model (PCM), have been extended to account for ERS tendency (ERS-PCM) by incorporating weight parameters multiplied by thresholds. Given that the weight parameter estimates in the ERS-PCM are affected by structural dependence, this study proposes the application of the ERS-PCM with the ERS tendency constrained to be constant across scales (ERS-PCM-C).

In Study 1, the ERS-PCM-C is applied to Big Five personality measurements. The findings show that the modified ERS-PCM-C reduces structural dependence for the weight parameter estimates and yields different trait level adjustments, resulting in a better model fit compared to the ERS-PCM. Study 2 examines parameter recovery for the ERS-PCM-C under various simulated conditions, differing in the number of scales, the mean trait level within a scale, and the relationship between ERS and the trait. The results demonstrate that the ERS-PCM-C recovers weight parameter estimates and the true correlations between ERS and the trait more accurately, with less bias compared to the ERS-PCM. In the ERS-PCM, estimation bias for weight parameters stems from structural dependence. The findings illustrate that the ERS-PCM-C improves weight parameter estimation by employing multiple scales, leading to different trait level adjustments than the ERS-PCM. The suggestion of models with constrained weight parameters highlights potential problems arising from the structural dependence between ERS and the trait, thereby increasing the applicability and generalizability of IRT models with ERS extensions.