Improving Measurement via Item Response Theory

Abstract

In my reaction to the major contribution by Mallinckrodt, Miles, and Recabarren, I endorse their recommendation that the use of item response theory (IRT) be increased. Advantages of IRT are numerous, with most resulting from the fact that IRT models typically take a more realistic view of how item responses are related to underlying traits than classical test theory. However, I raise concerns regarding their advocacy of the Rasch IRT model, which arguably takes an overly simplistic view of how items and traits are related. Many alternative IRT models exist, including ones based on the ideal-point measurement philosophy (most IRT models use the dominance model). I recommend that researchers seek to determine which IRT model best fits their items—including the larger question of whether a dominance versus ideal-point approach is preferable—and avoid assuming that one model will always perform the best.