The Effects of Sample Size and Guessing on Parameter Recovery in IRT Modeling of Aphasia Test Data

摘要

In this simulation study we sought to identify the most appropriate IRT measurement model for aphasia tests requiring 2-alternative forced-choice responses, exemplified by the Pyramids and Palm Trees Test. We also sought to estimate the minimum sample size necessary for estimating these models, under assumptions based on relevant empirical data. The results suggest that incorporating the assumption of correct guessing into the model improves performance. However, none of the models tested performed particularly well in any of the sample size conditions, likely because the test was very easy for most respondents, and guessing had a very large influence on performance.