Comparing predictive validity in a community sample: High–dimensionality and traditional domain–and–facet structures of personality variation

Abstract

Prediction of outcomes is an important way of distinguishing, among personality models, the best from the rest. Prominent previous models have tended to emphasize multiple internally consistent “facet” scales subordinate to a few broad domains. But such an organization of measurement may not be optimal for prediction. Here, we compare the predictive capacity and efficiency of assessments across two types of personality–structure model: conventional structures of facets as found in multiple platforms, and new high–dimensionality structures emphasizing those based on natural–language adjectives, in particular lexicon–based structures of 20, 23, and 28 dimensions. Predictions targeted 12 criterion variables related to health and psychopathology, in a sizeable American community sample. Results tended to favor personality–assessment platforms with (at least) a dozen or two well–selected variables having minimal intercorrelations, without sculpting of these to make them function as indicators of a few broad domains. Unsurprisingly, shorter scales, especially when derived from factor analyses of the personality lexicon, were shown to take a more efficient route to given levels of predictive capacity. Popular 20$^{th}$–century personality–assessment models set out influential but suboptimal templates, including one that first identifies domains and then facets, which compromise the efficiency of measurement models, at least from a comparative–prediction standpoint. © 2020 European Association of Personality Psychology