Dichotomous or polytomous data: which is best when analyzing job analysis ratings using item response theory?

Job analysis surveys typically involve rating scales (e.g., frequency, importance) with multiple response options. These data are typically analyzed with descriptive statistics (e.g., mean, standard deviation, percentages). Recently, Item Response Theory (IRT) has been explored as a technique for analyzing job analysis data. At times, when analyzing this data using IRT, the polytomous ratings are collapsed to only two data points. This thesis reviews the consequences of treating job analysis data as polytomous versus dichotomous for IRT analysis. The thesis uses archival data that was not previously analyzed with IRT. Focusing on common task statements from three entry-level job positions, it was demonstrated that there are certain advantages to keeping the data polytomous. Some advantages include increased clarity of how the individuals rated the tasks and better overall item fit. However, the dichotomized ratings do sufficiently answer the question of whether or not the task is part of the job.