There is an extensive history of arguments made by psychometricians that underlie this position. Edwards (2001) noted that researchers have long appreciated the need to avoid heterogeneous items: If such an item predicts a criterion, one will not know which aspect of the item accounts for the covariance. The same reasoning extends to tests: If a test includes multiple dimensions, one cannot know which dimensions account for the test's covariance with measures of other constructs. If one uses single scores from multidimensional tests, one has simply moved the heterogeneity problem from the item level to the scale level (Smith et al., 2003). Hough and Schneider (1995); McGrath (2005); Paunonen and Ashton (2001); and Schneider, Hough, and Dunnette (1996), among others, have all noted that use of scores of broad measures often obscures predictive relationships. Indeed, studies comparing prediction using specific facets of broad personality dimensions with prediction using scores on the dimensions themselves show that prediction is improved when one represents each facet individually (Paunonen, 1998; Paunonen & Ashton, 2001). Essentially, one gives oneself the chance to study the separate and incremental roles of each dimension involved in one's measures, rather than averaging across the different dimensions before predicting.
