The Superiority of Big Five Personality Traits

Ones and Viswesvaran (1996) recently discussed the bandwidth-fidelity dilemma in personnel selection. They presented several arguments in support of the exclusive use of broad, as opposed to narrow, personality variables in personnel selection, and reiterated their view (Ones, Schmidt and Viswesvaran, 1994) that there exists a very broad, integrity-related factor of personality. Several points raised by Ones and Viswesvaran (1996) are examined and criticized below, as a prelude to the description of the present study, which is intended to test some of Ones and Viswesvaran's arguments. One reason why Ones and Viswesvaran preferred broad measures of personality over narrower measures is that the former are more reliable. They gave examples from two personality test manuals to demonstrate that scales measuring the 'Big Five' traits have higher reliabilities than do the narrower subscales they comprise. However, these results follow directly from psycho- metric theory. Any group of positively intercorrelated subscales will produce a composite scale whose reliability exceeds that of the average of those subscales (see Nunnally, 1978, p. 249).

But despite this increase in reliability, it does not follow that the broad, composite scale must be a more effective predictor of a given criterion than are all of its constituent subscales. The question to be asked is whether or not the improved reliability derived by aggregating the subscales provides a gain in validity that outweighs the loss in validity due to the dilution of variance specific to certain subscales which relates to the criterion of interest. When the subscales of a broad composite are highly correlated, the validity of that composite is unlikely to be less than use subject to the validity of any of its constituent subscales. However, when the subscales are more weakly related, each possessing a substantial proportion of unique, non-error variance, the risk of decreased validity may be great. A study that demonstrated the potential loss in validity that may occur as a result of using only broad personality traits was that of Paunonen (1993), who found that various self-report behavioral criteria were better predicted by lower-level traits than by the Big Five. However, Ones and Viswesvaran (1996, p. 623) dismissed Paunonen's findings as the result of several methodo- logical 'errors', including an 'extremely small sample size and high capitalization on chance, poor nature of the criteria involved, [and] problems of reliability both in the criteria and the predictors (note the smaller number of items for the Big Five dimensions versus the relatively larger number of items for the narrow personality scales)'. But Paunonen's results were confirmed in a subsequent study (Ashton, Jackson, Paunonen, Helmes and Rothstein, 1995), to which none of Ones and Viswesvaran's criticisms apply. A similar study, by Reynolds and Nichols (1977), found opposite results, with narrower scales being outpredicted by the factors that comprise them. But the personality questionnaire used in the Reynolds and Nichols (1977) study was the California Psychological Inventory (CPI; Gough, 1957), a test whose scales are highly intercorrelated, and thus unlikely to contain enough unique variance to outpredict factor-level combinations of those scales.

Goldberg (1977) found that 33-36 per cent of the CPI scale intercorrelations exceeded 0.40, while only 9-13 per cent of scale intercorrelations among scales of the Personality Research Form (PRF; Jackson, 1989), the test whose scales were used in the Ashton et al. (1995) study, exceeded that same value. Ones and Viswesvaran (1996) also summarized the results of several meta-analyses of the relationships between personality and job performance, and argued that these results show the superiority of broad traits. One such result was the finding by Barrick and Mount (1994) that the Big Five conscientiousness factor was more predictive of job performance criteria than were any of its constituent lower-level traits. However, this result does not establish the superiority, in general, of broader measures, for two reasons. First, the measures of those aspects of conscientiousness considered by Barrick and Mount (1994) might not have contained a large enough proportion of unique, non-error variance to produce widely differing correlations with criteria. Many personality inventories are developed without any attempt to maximize discriminant validity or to minimize the influence of response styles, particularly desirability responding. Also, even if conscientiousness were a better predictor of job performance, averaged across jobs, it might nevertheless be true that certain aspects of conscientiousness, such as achievement orientation or planfulness, would be better predictors than conscientiousness of overall performance in certain types of jobs.

Ones and Viswesvaran (1996) also argued that the results of an earlier paper (Ones et al., 1994) were further evidence of the superiority of broad traits. They suggested, first, that integrity tests measure a very broad construct that subsumes several of the Big Five dimensions. This was based on their findings that integrity tests correlated positively with measures of three Big Five dimensions-conscientiousness, agreeableness, and emotional stability-which also correlated positively with each other. The positive manifold of correlations among these constructs was interpreted as evidence in favor of a general factor of personality, comparable in scope to the g factor of mental ability tests. Next, Ones et al. (1994) had compared results of their previous meta-analysis (Ones, Viswesvaran and Schmidt, 1993) with those of Barrick and Mount's (1991) meta-analysis, and showed that integrity tests had much higher validities with respect to job performance criteria than did any of the Big Five dimensions. These results were considered by Ones and Viswesvaran (1996) as additional support for their argument that broader traits are necessarily more valid predictors of j