혼합척도에서의 표현효과 탐지와 통제 방안: 제약된 요인혼합모형(CFMM)의 적용

초록

Wording effect refers to the systematic bias that occurs when respondents answer questionnaires based on item phrasing rather than content in scales with mixed positive and negative items. Since wording effects compromise research reliability and validity, researchers have traditionally employed factor analysis models to control for them. However, traditional factor analysis models are limited by their assumption of population homogeneity and their explanation of wording effects for the entire population. To address these issues, this study introduces and evaluates two control methods based on Constrained Factor Mixture Modeling (CFMM), which incorporates population heterogeneity. These approaches model wording effect response groups by differentially constraining parameter estimates such as factor loadings for positive and negative items across latent classes. The analysis examines Rosenberg's (1965) Self-Esteem Scale, comprising five positive and five negative items, using data collected from 5,271 respondents who participated in an online survey on “Social Development of MZ Generation in the COVID-19 Era.” The results first identified latent subgroups that respond heterogeneously to positive and negative items and determined their demographic characteristics. Second, the study compared the performance of two constrained factor mixture model approaches for identifying these wording effect response groups and verified their utility in terms of reliability and validity, with the more complex Steinmann model yielding better results than the Arias model. Finally, the practical implication of the results were discussed.

키워드