thirt: An R package for the Thurstonian Item Response Theory model using Markov chain Monte Carlo estimation
Phuong Linh Le Nguyen · 2022
Personality assessments are becoming increasingly popular outside of the academic settings, including hiring and promotion decisions. In these high-stakes settings, traditional personality questionnaires using Likert scales tend to incur response biases, in which participants may easily manipulate their responses to fit a desirable personality profile. Forced-choice questionnaires prevent biased responding by asking participants to rank different items each describing different constructs, which might be equally as socially desirable. However, these assessments provide ipsative scores that cannot be compared between individuals. The current paper describes thirt, an R package created specifically to simulate and estimate parameters under the Thurstonian Item Response Theory model for multi-dimensional forced-choice questionnaires. We will (1) explain the motivation behind forced-choice questionnaires for personality assessments, (2) describe the fundamentals of item response theory models in general and specifications of the current model, (3) describe the data generation model and Markov chain Monte Carlo estimation algorithm behind thirt, and (4) present simulation results that compares parameter recovery and performance between thirt and the R package thurstonianIRT which relies on existing general-purpose algorithms to fit the model. This paper provides evidence that by creating an estimation algorithm specific to this model, we are able to successfully and efficiently recover important parameters.