How Can We Use Mixture, Multi-Process, and Other Multi-Dimensional Item Response Theory Models to Account for Midpoint and Extreme Response Style Use in Personality Assessment?

• Activity Number: 54 - Record Linkage, Data Integration, and Improving Survey Measurement
• Type: Contributed
• Date/Time: Sunday, August 8, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Michael Lucci*
• Companies: University of Pittsburgh at Greensburg
• Keywords: Personality Subscales/Assessment; Midpoint/Extreme Response Styles; Multi-process (IRT tree) Model; Mixture Item Response Model; Multi-dimensional Item Response model; K-means clustering

Abstract

Since survey respondents may view item response options differently, accounting for midpoint (MRS) and extreme response style (ERS) use is important to accurately estimate the latent trait. This study investigated how five different IRT models for addressing ERS and MRS performed for three different personality subscales (Anxiety, Openness to Experience Feelings, and Compliance) from the German version of Costa and McCrae's NEO Personality Inventory-Revised. The mixture graded response and mixture partial credit models were compared with three multidimensional IRT models: the Multi-process (M-PM), Multidimensional Partial Credit (MPCM), and Multidimensional Nominal Response (MNRM) models.

Response process traits of the M-PM differed from response style traits of the other models. The two and three class mixture models, the two and three dimensional MNRM and MPCM, and the two process model for intensity ERS and direction fit better than standard IRT models. ERS accounted for more item response variability than MRS. The MPCM is suggested to questionnaire users to account for ERS and MRS due to the number of estimated parameters and amount of explained variability in item responses.