Multivariate Behavioral Research, 50:136, 2015 C Taylor & Francis Group, LLC Copyright  ISSN: 0027-3171 print / 1532-7906 online DOI: 10.1080/00273171.2014.988995

Abstract: Measurement Invariance Prior to Covariate Adjustment Kiran Khurshid and William F. Chaplin Department of Psychology, St. John’s University Meredith and Millsap (1992) argued that it is only after measurement invariance is established that efforts to explain the group difference are warranted. In this study, we illustrate the advice of Meredith and Millsap by considering an observed sex difference on a measure of cognitive failures. We found a gender difference on a measure of selfreported cognitive failures [CFQ] (Broadbent et al., 1982) in the Eugene-Springfield community sample (N = 756; 322 men and 434 women). Specifically, the mean for women was one third of a standard deviation (d = .31) higher than the mean for men (t(754) = 3.86, p < .001). We hypothesized that this gender difference might be explained by a gender difference on neuroticism. Specifically, personality variables, such as increased anxiety, sensitivity, or hyperviligance about symptoms would contribute to women’s higher scores on the CFQ. Before testing this hypothesis, we sought to ensure that the sex difference was not illusory because of a lack of measurement invariance between men and women. We undertook an analysis of differential item functioning (DIF) using Wald tests (Lord, 1977) as implemented in IRTPRO 2.1. We fit a graded response model to the 25 CFQ items and found nine items on the cognitive failures scale for which the Wald tests indicated significant DIF. Eight of the nine I would like to express my gratitude to my SMEP sponsor, Dr. William F. Chaplin, for his invaluable feedback and support. Correspondence concerning this abstract should be addressed to Kiran Khurshid, Department of Psychology, St. John’s University, 8000 Utopia Parkway, Queens, New York, NY 11439. E-mail: Kiran. [email protected]

items exhibited uniform DIF, and one item exhibited both uniform and non-uniform DIF. On average, the thresholds for women were shifted slightly higher than the thresholds for men. The mean difference in the DIF-adjusted expected a posteriori (EAP) scores was 0.39 higher for women than men. Thus, the gender difference was not a measurement artifact. Controlling for neuroticism reduced the standardized difference between men and women on the CFQ from .35 to .29. However, the difference was still statistically significant (t(753) = 4.15, p

Abstract: Measurement Invariance Prior to Covariate Adjustment.

Abstract: Measurement Invariance Prior to Covariate Adjustment. - PDF Download Free
565B Sizes 1 Downloads 8 Views