Flexible Latent Trait Metrics: An Application of the Filtered Monotonic Polynomial Item Response Model

Although item response theory (IRT) analysis is widely used to analyze test data, the scores that are reported and analyzed are often transformations of the IRT trait estimates. Transforming the IRT latent metric affects not only the reported scores but properties of the model such as test information. If the metric transformation is nonlinear, interval-level inferences are also affected. One theoretical way to avoid the problems associated with nonlinear score transformations is to build the IRT model directly on the reported metric. However, until now there has been no formal framework in which to build an IRT model on a user-specified metric or to nonlinearly transform an IRT model on one metric to another metric. In this talk, I propose that the filtered monotonic polynomial item response model (FMP; Liang & Browne, 2015), which includes the familiar two-parameter model, can be used to model both linear and nonlinear metric transformations. Specifically, I show that the IRT models that result from taking both linear and nonlinear metric transformations of FMP models can also be expressed as FMP models.

Leah Feuerstahler is an IES postdoctoral fellow at the UC Berkeley Graduate School of Education and the BEAR Center. Her research interests include theoretical and applied issues in item response theory and latent variable modeling. Her work appears in peer-reviewed journals such as Multivariate Behavioral Research and Applied Psychological Measurement. Leah received her PhD in Quantitative Psychology from the University of Minnesota in 2016 and an MA in Statistics from the same university in 2015.

Date: 
Tuesday, September 19, 2017 - 2:00pm
Building: 
Tolman
Room: 
2515