Reducing the length of a Goldberg Based Personality Questionnaire using Item Response Theory & Confirmatory Factor Analysis : a thesis presented in partial fulfilment of the requirements for the degree of Masters of Science in Psychology at Massey University, Albany, New Zealand

Abstract

Objectives: This study seeks to reconstitute an existing personality questionnaire by identifying the items that capture the best quality information as measured through item Response Theory (IRT). This process will reduce the length of this measure and increase its measurement precision. Method: A polytomous IRT model (Graded Response: Samejima, 1969) will be used to assess the psychometric properties of each item in this questionnaire and produce item level graphs in order to select the best three items for each of the 26 first-order factors. Confirmatory Factor Analysis (CFA) will be used to assess the model fit and unidimensionality before and after the IRT selections are made. This will illustrate the improvement gained through both the deletion of redundant items and the selection of high-quality items. Results: This questionnaire was reduced from 246 items down to 78 items with three high-quality items identified for each of the 26 first-order factors. The model fit considerably improved through this selection process and the reduction of information was minimal in comparison to the amount of items that were deleted. Conclusions: This study illustrated the power of using IRT for test development. The item selections are not only of benefit for the organisation that supplied the data for this study, but also the original developers as well as any other users of these items as they are freely available via an online source.