Electronic Thesis and Dissertation Repository

Degree

Doctor of Philosophy

Program

Psychology

Supervisor(s)

Dr. John P. Meyer

Abstract

There continues to be debate about the dimensionality of important psychological constructs (e.g., anxiety; job satisfaction). The standard procedures for determining whether a construct is unidimensional or two-dimensional have been to (a) conduct factor analyses and (b) test for differences in the nomological network of correlations with other variables. I argue, and demonstrate empirically, that these approaches can sometimes mislead researchers to draw incorrect conclusions. In Study 1, I examined how item valence (i.e., favorability of item content) can affect factor analyses and nomological network analyses with two separate samples of undergraduate students. Results consistently showed that item valence can induce regular- and reverse-keyed items to load on separate factors in factor analyses, and that it can systematically bias construct correlations in favor of the two-dimensional interpretation of a construct. Multi-trait multi-method analyses demonstrated that the item valence accounts for close to 10% of the item variance, and this effect cannot be entirely explained by social desirability. In Study 2, I again tested the effect of item valence in a job satisfaction measure. In addition, I examined careless responding as an alternative explanation to the item valence effect among working adults. Results not only replicated item valence effect in Study 1, but it also showed that careless responding can amplify, but cannot explain entirely, the apparent two dimensionality results caused by valence bias in both factor analytic and nomological network analyses. I suggested several remedies for the valence problem, including the minimization of valence in item design, the use of reverse-keyed items for construct measurement, and the use of objective behavioral measures in nomological network investigations.


Share

COinS