9:45AM - 11:15AM
Mudd Building | Room 833
To θ or not to θ: A simulation on the validity of Item Response Theory (IRT)
The code for the simulation study can be found by clicking here. The slides for this presentation may be found by clicking here. Find the references for the talk and some FAQs here.
The talk will review some of the results from a broader simulation study exploring the effects that poorly functioning items have on accuracy (i.e., optimal test length, estimate-to-total correlations, estimate-to-outcome correlations, false-positive, and false-negative rates) across a multitude of testing frameworks (i.e., unweighted summed scores, α-optimization, factor analysis, and Item Response Theory).
We look forward to meeting you on Thursday!
Findings suggest that all simulated methods perform quite adequately. Additionally, the testing frameworks may differ in their ability to identify poorly functioning items. Specifically, IRT and factor analysis perform best when tests are longer (e.g., 40- to 80-items) and α-optimization when tests are shorter (e.g., 20-items). Finally, IRT typically generates drastically shorter tests without necessarily compromising accuracy in any outcome that was assessed. This may have implications in test development and construction that require parsimonious scales (i.e., short-forms, clinical applications where time is limited, etc.).