Trait specificity refers, in the context of a measure of a given personality trait, to variance that is not due to a broad personality factor but that is reliable or valid variance.
Examining Trait Specificity
Some omnibus personality inventories assess many narrow, “facet”-level personality traits that load on a few broad personality factors. One way to examine the trait specificity of a facet-level variable is to focus on its reliable variance, by comparing its internal-consistency reliability with its correlations with other facets that define the same factor. For example, if a given facet scale has a reliability of 0.75 and correlates 0.45 with the other facets of its factor, then this would suggest that trait specificity represents about 75 − 45 = 30% of that facet scale’s variance (or about 30/75 = 40% of its reliable variance). (This amount would actually be somewhat lower to the extent that the facet is also associated with the facets of any other factor.)
Another way to examine the trait specificity of a facet-level variable is to focus on its (convergent) valid variance, as approximated most commonly by the agreement between self-reports and observer reports from persons who know each other very well. This approach involves comparing a facet scale’s self/observer convergent correlation with that facet’s cross-source correlations with other facets that define the same factor. For example, if self-reports on a given facet scale correlate 0.40 with observer reports on the same scale but correlate 0.24 on average with observer reports on other facet scales that define the same factor, then this would suggest that trait specificity represents about 40 − 24 = 16% of that facet scale’s variance (or about 16/40 = 40% of its convergent valid variance).
Broad Versus Narrow Traits in Prediction
Researchers have disagreed as to whether trait specificity can contribute to the prediction of outcome variables, beyond the predictive validity (i.e., criterion validity) provided by the common factor variance shared by related traits (e.g., Ashton et al. 2014; Salgado et al. 2013). For the practical purpose of using personality measures to predict outcome variables, the issue is this: Is predictive validity essentially maximized by using a few broad scores representing the major personality factors? Or, instead, is predictive validity considerably increased by using additional scores representing narrow traits that define the broad factors?
In some investigations of this issue, researchers use regression techniques to compare the predictive validity of many narrow traits against that of a few broad factors. In such investigations, researchers typically use adjusted multiple correlations to avoid capitalizing on chance associations in a given sample. A more theoretically oriented approach avoids relying on regression equations and instead compares the predictive validity of one or more broad dimensions with that of one or more narrow traits, where these personality variables are selected a priori as being conceptually related to the outcome variable (e.g., Ashton et al. 2014).
One methodological issue in comparing narrower trait measures against broad factor measures is that of the relative length of the measures. If the researcher wishes to assess personality broadly – as might be desired when a diverse array of outcomes is to be predicted – then the narrow traits can be assessed as constituent facet-level scales of broad factor-level scales, and the former will necessarily be much shorter. If instead the researcher wishes to assess only certain aspects of personality that are conceptually relevant to one outcome, then it may be appropriate to administer (and compare) a long narrow measure against an equally long broad measure.
In some studies, narrow traits have shown clear superiority over broad dimensions in the prediction of theoretically related outcome variables (e.g., Ashton et al. 2014). This result indicates that the use of broad personality measures exclusively will sometimes produce lower criterion validity than will the use of both broad and narrow measures. However, some outcome variables might be better predicted by broad dimensions than by narrow traits, and the optimum predictor variable(s) can be selected on the basis of empirical evidence from large samples similar to those in which the outcomes will be predicted.
Measures of various personality traits typically contain some reliable or valid variance that is not accounted for by broad personality factors. This trait specificity may in some cases contribute to the prediction of theoretically relevant criterion variables.