A comparison of three approaches to constructing item parcels to improve subject-to-parameter ratios in confirmatory factor analysis

Abstract

A confirmatory factor analysis (CFA) of the 100 items of the Revised Eysenck Personality Questionnaire (EPQ-R) on a sample of 205 South African refugees failed to satisfactorily fit the EPQ-R's intended 4-factor structure. The lack of fit is possibly due to the low subject-to-parameter (n:k) ratio the dichotomous item responses. Three procedures were examined for combining items into parcels to alleviate these problems. Within each of the Psychoticism, Extraversion, Neuroticism, and Lie scales, typically three items were parceled by (a) judged content, (b) highest intercorrelation, and (c) randomly, CFAs of these three methods for constructing items parcels showed equivalent, if modest, fits of the 4-factor model to the data. The Psychoticism indicated that the other three factors appeared to reasonably measure their intended constructs. Arguments related to simplicity and the a priori explication of meaning are used to advocate to application of content parcels.