A unifying account and an empirical study of spurious multidimensionality in psychological measures

It is common knowledge that psychological measures are hardly ever unidimensional. Some authors explain the multidimensionality by additional construct-irrelevant factors, such as social desirability, careless responding, or verbal abilities. However, we focus on spurious multidimensionality, caused by imperfections in the measurement model, and provide an unifying account of why it arises. We hypothesized that if the latent factor and at least two groups of items have the same non-linear relationship, it will require modeling an additional, apparent “method” factor to account for the non-linear relationship, even though the items are fundamentally unidimensional. An empirical study (N = 11,979 people, 51% women) supported our expectations. Items sharing a similar pattern of misfit disrupted unidimensionality and tended to form an additional factor, but this factor still contained construct-relevant variance. Our findings pinpoint under which conditions spurious factors arise and underscore how it is essential to check whether the relationship between a latent variable and its indicators is modeled correctly to avoid spurious multidimensionality, interpret spurious factors substantively, or needlessly remove items that seemingly disrupt unidimensionality.