Date of this Version
Beck, M. F. (2023). The examination of nonparametric person-fit statistics as appropriate measures of response bias in ordered polytomous items [Unpublished doctoral dissertation]. University of Nebraska-Lincoln
Survey research is ubiquitous within the social sciences; however, surveys are vulnerable to response biases. Response biases introduce construct-irrelevant variance into survey responses, which degrades the accuracy of conclusions drawn through the use of surveys. Nonparametric person-fit statistics have been shown to accurately identify response biases in dichotomous response data but are not well studied in polytomous response data. This study examines the accuracy of nonparametric person-fit statistics in polytomous response data. A 6 x 4 x 4 x 2 simulation study was conducted, with type of aberrancy (6), number of response options (4), dimensionality (4), and test length (2) as factors. The sensitivity, specificity, positive predictive value, and negative predictive value for U3, the normed number of Guttman errors, and HTi were calculated using a bootstrapped cutoff. Findings indicate that these person-fit statistics with a conservative cutoff had excellent specificity but poor sensitivity.
Advisor: Kurt F. Geisinger