Rasch Modeling of Survey Data

Rasch measurement is a methodology that can be used to obtain equal-interval measures of person ability from test or survey responses (Boone, 2016; Engelhard Jr, & Wang, 2024; Wilson, 2023). Consider a situation where you are administering a locus of control instrument to a group of college students. You want to be able to obtain estimates that work like a ruler—equal differences represent equal amounts of the measured trait (equal-interval). This makes the scores useful for further statistical analyses. What procedure might help to generate robust estimates of each person’s ability? Rasch measurement is a form of latent trait modeling that can be applied to both dichotomous items (e.g., correct/incorrect) and polytomously-scored tests or surveys (e.g. level of agreement measuring confidence in one’s ability to conduct a t-test), and generates sample-independent, equal-interval estimates of a person’s ability on a target latent trait. Rasch measurement seeks to provide robust, psychometrically sound person estimates by simultaneously generating item difficulty estimates and person ability estimates on a common linear scale. In a testing context, item difficulty estimates refer to how difficult a knowledge assessment item was to answer correctly when scored dichotomously or how difficult it was for a higher score to be assigned to a polytomous item when scored using an ordinal grading scale. The resulting estimates can be evaluated for model fit, reliability of separation, and also can provide diagnostic information regarding individual response patterns. Rasch measurement is widely applicable and can be used in numerous disciplines.