Measuring Personality When Stakes Are High: Are Graded Paired Comparisons a More Reliable Alternative to Traditional Forced-Choice Methods?

Abstract

In graded paired comparisons (GPCs) two items are compared using a verbally anchored, multi-point rating scale. GPCs can measure non-cognitive constructs. GPCs are expected to reduce faking/response tendencies compared with Likert-type scales and to produce more reliable and less ipsative trait scores than traditional-binary forced-choice response formats. To investigate the statistical properties of GPCs, we conducted a set of 288 simulations in which we varied six independent factors and additionally implemented conditions with algorithmically optimized item combinations. Traits were estimated using Thurstonian IRT models. Under ideal conditions, good reliabilities and low ipsativity of trait estimates were achieved. However, under conditions similar to those in high-stakes assessments, reliabilities did not consistently exceed the conventional threshold. Moreover, there was evidence for ipsativity. In these conditions, more response categories and optimized combination of items led to higher reliabilities and nearly fully normative trait scores. In sum, this simulation informs about the psychometric properties of GPCs under different conditions and makes specific recommendations as to how these properties can be improved. We discuss the generalizability of our results to response processes in selection situations.