personality and job performance: the importance of narrow traits

15
Personality and job performance: the importance of narrow traits MICHAEL C. ASHTON Department of Psychology, University of Western Ontario, London, Ontario N6A 5C2, Canada Summary In a recent discussion of the bandwidth–fidelity dilemma in personality measurement for personnel selection, Ones and Viswesvaran (1996) concluded that ‘broader and richer personality traits will have higher predictive validity than narrower traits’. In this paper, the arguments made by Ones and Viswesvaran in favor of the exclusive use of broad personality dimensions are discussed. New data are presented that contradict Ones and Viswesvaran’s claim of the existence of a general, integrity-related personality factor, and that show two narrow measures—the Responsibility and Risk Taking scales of the Jackson Personality Inventory—to have higher validities than the Big Five dimensions with respect to job performance criteria based on self-reported workplace delinquency in a sample of 127 entry-level employees. # 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289–303 (1998) Introduction 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 eective 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 CCC 0894–3796/98/030289–15$17.50 Received 1 February 1996 # 1998 John Wiley & Sons, Ltd. Accepted 25 July 1996 JOURNAL OF ORGANIZATIONAL BEHAVIOR, VOL. 19, 289–303 (1998)

Upload: michael-c-ashton

Post on 06-Jun-2016

219 views

Category:

Documents


5 download

TRANSCRIPT

Page 1: Personality and job performance: the importance of narrow traits

Personality and job performance:the importance of narrow traits

MICHAEL C. ASHTON

Department of Psychology, University of Western Ontario, London, Ontario N6A 5C2, Canada

Summary In a recent discussion of the bandwidth±®delity dilemma in personality measurementfor personnel selection, Ones and Viswesvaran (1996) concluded that `broader and richerpersonality traits will have higher predictive validity than narrower traits'. In this paper,the arguments made by Ones and Viswesvaran in favor of the exclusive use of broadpersonality dimensions are discussed. New data are presented that contradict Ones andViswesvaran's claim of the existence of a general, integrity-related personality factor, andthat show two narrow measuresÐthe Responsibility and Risk Taking scales of theJackson Personality InventoryÐto have higher validities than the Big Five dimensionswith respect to job performance criteria based on self-reported workplace delinquencyin a sample of 127 entry-level employees. # 1998 John Wiley & Sons, Ltd.

J. Organiz. Behav. 19: 289±303 (1998)

Introduction

Ones and Viswesvaran (1996) recently discussed the bandwidth±®delity dilemma in personnelselection. They presented several arguments in support of the exclusive use of broad, as opposedto narrow, personality variables in personnel selection, and reiterated their view (Ones, Schmidtand 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 aprelude to the description of the present study, which is intended to test some of Ones andViswesvaran's arguments.

One reason why Ones and Viswesvaran preferred broad measures of personality over narrowermeasures is that the former are more reliable. They gave examples from two personality testmanuals to demonstrate that scales measuring the `Big Five' traits have higher reliabilities thando the narrower subscales they comprise. However, these results follow directly from psycho-metric theory. Any group of positively intercorrelated subscales will produce a composite scalewhose reliability exceeds that of the average of those subscales (see Nunnally, 1978, p. 249). Butdespite this increase in reliability, it does not follow that the broad, composite scale must be amore e�ective predictor of a given criterion than are all of its constituent subscales. The questionto be asked is whether or not the improved reliability derived by aggregating the subscalesprovides a gain in validity that outweighs the loss in validity due to the dilution of variancespeci®c to certain subscales which relates to the criterion of interest. When the subscales of abroad composite are highly correlated, the validity of that composite is unlikely to be less than

CCC 0894±3796/98/030289±15$17.50 Received 1 February 1996# 1998 John Wiley & Sons, Ltd. Accepted 25 July 1996

JOURNAL OF ORGANIZATIONAL BEHAVIOR, VOL. 19, 289±303 (1998)

Page 2: Personality and job performance: the importance of narrow traits

the validity of any of its constituent subscales. However, when the subscales are more weaklyrelated, each possessing a substantial proportion of unique, non-error variance, the risk ofdecreased validity may be great.

A study that demonstrated the potential loss in validity that may occur as a result of using onlybroad personality traits was that of Paunonen (1993), who found that various self-reportbehavioral criteria were better predicted by lower-level traits than by the Big Five. However, Onesand Viswesvaran (1996, p. 623) dismissed Paunonen's ®ndings as the result of several methodo-logical `errors', including an `extremely small sample size and high capitalization on chance, poornature 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 numberof items for the narrow personality scales)'. But Paunonen's results were con®rmed in asubsequent study (Ashton, Jackson, Paunonen, Helmes and Rothstein, 1995), to which none ofOnes and Viswesvaran's criticisms apply.

A similar study, by Reynolds and Nichols (1977), found opposite results, with narrower scalesbeing outpredicted by the factors that comprise them. But the personality questionnaire used inthe 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 enoughunique variance to outpredict factor-level combinations of those scales. Goldberg (1977) foundthat 33±36 per cent of the CPI scale intercorrelations exceeded 0.40, while only 9±13 per cent ofscale intercorrelations among scales of the Personality Research Form (PRF; Jackson, 1989), thetest 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 therelationships between personality and job performance, and argued that these results show thesuperiority of broad traits. One such result was the ®nding by Barrick and Mount (1994) that theBig Five conscientiousness factor was more predictive of job performance criteria than were anyof its constituent lower-level traits. However, this result does not establish the superiority, ingeneral, of broader measures, for two reasons. First, the measures of those aspects of con-scientiousness considered by Barrick andMount (1994) might not have contained a large enoughproportion of unique, non-error variance to produce widely di�ering correlations with criteria.Many personality inventories are developed without any attempt to maximize discriminantvalidity or to minimize the in¯uence 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 achievementorientation or planfulness, would be better predictors than conscientiousness of overall perform-ance 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, ®rst, that integrity testsmeasure a very broad construct that subsumes several of the Big Five dimensions. This was basedon their ®ndings that integrity tests correlated positively with measures of three Big FivedimensionsÐconscientiousness, agreeableness, and emotional stabilityÐwhich also correlatedpositively with each other. The positive manifold of correlations among these constructs wasinterpreted as evidence in favor of a general factor of personality, comparable in scope to the gfactor of mental ability tests. Next, Ones et al. (1994) had compared results of their previousmeta-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 tojob performance criteria than did any of the Big Five dimensions. These results were consideredby Ones and Viswesvaran (1996) as additional support for their argument that broader traits arenecessarily more valid predictors of job performance.

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

290 M. C. ASHTON

Page 3: Personality and job performance: the importance of narrow traits

Despite the apparent predictive validity of integrity tests, as reported by Ones et al. (1993),there are reasons to doubt the existence of a general factor of personality de®ned by the con-structs measured by integrity tests. First, if personality scales measuring conscientiousness,agreeableness, emotional stability, and overall integrity really were substantially intercorrelated,then one would expect most omnibus personality inventories to contain a large integrity factor.But despite the vast number of personality inventories in use and the myriad studies of theirfactor structures, a general, integrity-related factor rarely, if ever, emerges. Another reason todoubt the general, integrity-based factor posited by Ones et al. (1994) is that their data suggestthat variance unique to integrity tests is largely responsible for the criterion validity of those tests.If there were an important general factor, then the validity of integrity tests should largely be dueto variance shared with Big Five dimensions. But Ones et al.'s (1994) results showed that thiswas not the case; partialling out the Big Five had little e�ect on the validity of integrity tests,suggesting that variance unique to those tests was in large part responsible for their validity.

Ones and Viswesvaran (1996) also argued that most job performance criteria are factoriallycomplex, and thus require factorially complex predictors in order that maximum validity may beachieved. Therefore, they concluded, broad and general personality measures will be the bestpredictors of those criteria. However, there is no guarantee that the narrower traits that are thebest predictors of a criterion will necessarily be subsumed within the same higher-order construct.One example of this distinction, from a context unrelated to job performance, was reported byAshton et al. (1995). Ashton et al. found that a criterion of `fun-seeking' behavior (e.g. party-going, sports-playing, alcohol consumption, etc.) was much better predicted by a unit-weightedcombination of the sociability and impulsivity aspects of extraversion and low conscientiousness,respectively, than by those broad factors themselves. Optimal prediction was not achieved byusing broad measures corresponding to the Big Five factors, but instead by combining subscalesrepresenting di�erent Big Five factors. A major disadvantage of relying entirely on broad person-ality measures is that one is prevented from extracting from the broad dimensions those narrowerfacets that have the strongest theoretical and empirical relationships with the criteria of interest.

The purpose of the present study is to test Ones and Viswesvaran's (1996) claim that broaderpersonality variables are necessarily better predictors of job performance criteria, and also todetermine whether or not there exists a general factor of personality de®ned by the constructsmeasured by integrity tests. The method that will be used to accomplish this is to administer avariety of personality scalesÐincluding markers of the Big Five and measures of integrity-relatedtraitsÐto a sample of respondents for whom some job performance-related criteria are available.If Ones and Viswesvaran's arguments are correct, then two results should occur. First, when thepersonality scales are factor-analyzed, a large general integrity factor should emerge. Second, thevalidity of the personality factor scores (such as the Big Five) for predicting job performancecriteria should exceed the validity even of those narrow personality scales that bear substantiverelationships to the same criteria.

The sample of respondents selected for this study was drawn from an undergraduate studentpopulation at a Canadian university. This population was chosen partly for reasons of con-venience, but partly because of the typical employment history of such individuals. MostCanadian high school or university students are employed full-time during the summer months,and many are employed part-time during the academic year, mainly at entry-level jobs, for whichintegrity tests are most commonly used in selection (Ones et al., 1993). Examples of jobscommonly held by students include waiter/waitress, cashier, grocery store clerk, gas stationattendant, groundskeeper, children's camp counsellor, farmhand, and factory worker.

Owing to the great di�culty and expense of obtaining job performance ratings directly fromthe many former employers of these respondents, it was decided to use respondents' self-reports

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

PERSONALITY AND JOB PERFORMANCE 291

Page 4: Personality and job performance: the importance of narrow traits

of previous workplace delinquency as a job performance criterion. Because these self-reportswere to be obtained anonymously and in a research context unrelated to employment, they wereexpected to be accurate and frank. The use of self-reports of workplace delinquency as estimatesof job performance was justi®ed by the results of the Michigan Employability Survey (MichiganDepartment of Education, 1989; cited in Ones et al., 1993), which indicated that employers ofentry-level personnel considered integrity and trustworthiness to be the most important qualitiesin such employees.

One concern with the use of self-report measures as both predictors and criteria is thatcorrelations with predictors might be in¯ated due to shared method variance. On the other hand,however, it could be argued that these self-reports would give an estimate of true overallperformance at least as good as that which could be made by supervisors, who presumably wouldnot detect all such instances of employees' delinquent behavior on the job. In any case, to theextent that correlations may be in¯ated, the results of comparisons between broad and narrowpersonality measures should not be a�ected.

Method

Respondents

The respondents for this study were undergraduate students in an introductory psychologycourse who participated in exchange for course credit. There were 131 respondents altogether,of whom 52 were male and 79 were female. Their median age was 19 years.

Criterion measure: Workplace Behavior Questionnaire

The job-related criterion variable used in this study was a self-report measure of workplacedelinquency, entitled the Workplace Behavior Questionnaire, that asked respondents to indicatequantitatively the frequency with which they performed various counterproductive or dishonestbehaviors when working for pay during part-time or summer jobs. The delinquent workplacebehaviors assessed in this study were the following: unnecessary absenteeism, lateness, alcoholuse or in¯uence, safety violations, `goldbricking' (i.e. avoiding work during paid time), theft,giving of `freebies' (i.e. free goods or services) to friends or relatives, and vandalism or sabotage.The items themselves, and their response alternatives, are listed in the Appendix, along withgeneral instructions and two items asking respondents to indicate the approximate number ofhours they had worked for pay during their high school years.

In order to ensure accurate responses, respondents were assured complete anonymity and wereasked not to indicate their names on any of their answer sheets. It was also expected that thequantitative nature of these delinquency items would make subjective interpretations andresponse styles somewhat less prevalent than would be the case if subjects were asked to rate, forexample, their overall job performance.

For each of the delinquency items, responses were scored on a scale of 1 to 6, with highernumbers representing response categories that involve greater delinquency. It was understoodthat the resulting measures would not strictly have interval scale properties, and that theirdistributions might be non-normal, but that these items should nevertheless be fairly amenable tocorrelational analyses. Furthermore, any problems concerning metric properties of the criterion

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

292 M. C. ASHTON

Page 5: Personality and job performance: the importance of narrow traits

variables would not be expected to in¯ate the correlations between those variables and thepredictor scales.

Predictor measures: personality scales

One set of personality measures used in this study was Goldberg's (1992) adjective markers of theBig Five. This inventory comprises ®ve sets of 20 adjectives, and is intended to provide factor-univocal measures of each of the Big Five traits. These adjective markers were developed in thetradition of the lexical hypothesis, from which the Big Five traits were originally discovered. Theyhave shown high reliabilities and have correlated highly with other measures of correspondingBig Five traits (Goldberg, 1992). In this study, scores on each of the Big Five adjective scales werecomputed as the sums of responses to the appropriate 20 adjectives. The other method recom-mended by Goldberg (1992)Ðcomputation of ®ve orthogonal, varimax-rotated factor scoresfrom all 100 adjectivesÐwas not used, because this technique would eliminate any correlationsamong Big Five traits that could produce a general factor.

The second set of personality measures used in this study was an abbreviated version of theJackson Personality Inventory (JPI; Jackson, 1970, 1994). The scales of this questionnaire weredeveloped according to a construct-based strategy, and are designed to cover a wide spectrum ofpersonality traits. High levels of reliability and validity have been reported for the JPI scales(Jackson, 1970, 1994). In addition to the 15 JPI scales, the aggression scale of the PRF was alsoincluded, as a measure of aggressiveness. Eight items from each of the 16 questionnaire scaleswere included, having been selected on the basis of high correlations with their respective scaletotals. Overall, these scales were expected to provide fairly even coverage of a large proportion ofthe personality domain.

Two JPI scales were of particular interest for this study because of their expected relationshipsto the workplace delinquency criteria. One of these scales, responsibility, is conceptually similarto some personality-based integrity tests, and `is identi®ed largely in terms of the degree to whicha person feels an abstract moral obligation to other people and to society at large. A high scorerfeels a sense of obligation to ``do the right thing,'' regardless of possible personal consequences'(Jackson, 1994, p. 23). The other scale, risk taking, measures the extent to which a person seeks tobe exposed to uncertain situations, especially those in which ®nancial gain or loss are involved.

Several studies have shown that both the responsibility and risk taking scales predict criteriarelated to dishonest behavior. Jackson (1994, pp. 56±57) found that responsibility correlated0.33, and risk taking ÿ0:40, with peer ratings of trustworthiness. In Jackson, Hourany andVidmar (1972), responsibility correlated ÿ0:50, and risk taking 0.44, with willingness to behaveunethically in various hypothetical situational dilemmas. Mikulay and Go�n (1996) found bothof these scales to yield absolute correlations ranging from 0.49 to 0.73 with two subscales of acommercially available integrity test, and also found both responsibility and risk taking toproduce signi®cant correlations with a variety of dishonest behaviors elicited in a laboratoryexperiment, approximating the correlations yielded by the much longer integrity test. On thebasis of all of the above results, the responsibility and risk taking scales were expected to showstrong correlations with the workplace delinquency criteria, exceeding those produced by theother questionnaire scales or by the Big Five adjective scales. These two scales represent thenarrower personality measures whose validities will be compared with those yielded by broadpersonality factors.

Thus, the adjective and questionnaire scales used in this study provided markers of the Big Fivetraits, of narrower constructs expected to correlate with the workplace delinquency criteria, and

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

PERSONALITY AND JOB PERFORMANCE 293

Page 6: Personality and job performance: the importance of narrow traits

of a variety of other personality variables. This array of variables was intended to provide anexcellent opportunity for an integrity-related general factor to emerge, if there is in fact a positivemanifold of correlations among the adjective measures of conscientiousness, agreeableness, andemotional stability, the JPI or PRF scales related to the same Big Five traits (such as organ-ization, (low) aggression, or (low) anxiety), and the integrity-related JPI scales (responsibility andrisk taking). Also, the correlations of the workplace delinquency criteria with this general factor,the Big Five factors, and the rationally-chosen narrow scales (responsibility and risk taking) canbe compared in order to determine whether or not broader personality variables are in fact morevalid predictors of job performance.

Procedure

Respondents ®rst completed the Big Five adjective markers, and then the abbreviated version ofthe JPI. They next performed several mental ability tasks that were unrelated to this study,requiring about 25 minutes total time. Subsequently, they completed the Workplace BehaviorQuestionnaire.

Results

Four respondents, two male and two female, reported that they did not work for pay at any timeduring either the summers or the school terms of their high school years. These respondents wereexcluded from further analysis, leaving a total sample size of 127 (50 male, 77 female). Of these127 respondents, the median category for number of hours worked per week during the schoolyear was 5±9 hours, and the median category for number of hours worked per week during thesummer was 30±39 hours.

Self-reported workplace delinquency

Table 1 shows the number of respondents who chose each of the six response categories for eachof the eight delinquency questions. These results indicate that most of the respondents reportlittle involvement in delinquent behaviors while on the job, but that some respondents reportheavy involvement.

Table 2 gives the correlations among the eight measures of workplace delinquency. The averageinter-item correlation (0.30) corresponds to an internal-consistency reliability (Cronbach's alpha)of 0.77 for a composite delinquency measure calculated as the sum of all eight criteria. Similarly,a principal components analysis of these items produced a general factor (i.e. component),accounting for 40.3 per cent of the total variance, with the questions' loadings ranging from 0.38to 0.82.

The minimum and maximum possible scores on the composite delinquency measure were 8and 48, respectively. The obtained scores ranged from 8 to 36, with a mean of 13.4 and a standarddeviation of 4.9. The mean score for males, 14.3, was less than one-third of a standard deviationabove the mean score for females, 12.8; however, the highest score obtained by any female wasonly 24.

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

294 M. C. ASHTON

Page 7: Personality and job performance: the importance of narrow traits

Personality±delinquency correlations

Table 3 displays the correlations of the eight delinquency variables, plus the composite delin-quency measure, with the 15 JPI scales, the PRF aggression scale, and the ®ve adjective scales.These results indicate that the overall delinquency scale was more strongly associated with thepersonality scales than was any individual delinquency item.

The strongest correlations with the delinquency criteria were those obtained by the abbreviatedJPI responsibility and JPI risk taking scales. When these two scales, which intercorrelate onlyÿ0:17, are aggregated into a unit-weighted dependability composite (i.e. responsibility minusrisk taking), the correlation of this composite with delinquency is ÿ0:45. The only otherstatistically signi®cant correlations with the workplace delinquency composite were obtained bythe adjective agreeableness and conscientiousness scales; interestingly, the adjective emotionalstability scale was uncorrelated with delinquency.

Table 1. Frequency distributions of response categories for workplace delinquency items

Unnecessary absenteeism Lateness Alcohol use/in¯uence Safety violations% of shifts N % of shifts N % of shifts N % of shifts N

0 67 0 30 0 103 0 801±3 50 1±3 65 1±3 18 1±3 314±9 9 4±9 15 4±9 2 4±9 10

10±19 1 10±19 12 10±19 4 10±19 420±34 0 20±34 4 20±34 0 20±34 1> 35 0 > 35 1 > 35 0 > 35 1

`Goldbricking' Theft Unauthorized `freebies' Vandalism/sabotage% of paid time N Total $ value N Total $ value N Total $ value N

55 56 0 67 0 76 0 1095±9 35 1±19 39 1±19 34 1±19 12

10±19 19 20±49 7 20±49 8 20±49 320±29 11 50±99 7 50±99 5 50±99 130±49 4 100±199 0 100±199 2 100±199 0> 50 2 > 200 7 > 200 2 > 200 2

See Appendix for workplace delinquency items. For purposes of calculating correlations and of aggregation, items werescored on a 1±6 scale, with larger numbers representing higher levels of delinquency.

Table 2. Correlations among workplace delinquency items

Delinquency item 1. 2. 3. 4. 5. 6. 7. 8.

1. Absenteeism 1.002. Lateness 0.27� 1.003. Alcohol 0.28� 0.28� 1.004. Safety violations 0.33� 0.17 0.13 1.005. `Goldbricking' 0.15 0.28� 0.37� 0.20{ 1.006. Theft 0.29� 0.21{ 0.43� 0.30� 0.44� 1.007. `Freebies' 0.16 0.08 0.28� 0.39� 0.30� 0.70� 1.008. Vandalism 0.30� 0.08 0.19{ 0.55� 0.39� 0.54� 0.46� 1.00

� p < 0:01; { p < 0:05.

See Appendix for workplace delinquency items.

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

PERSONALITY AND JOB PERFORMANCE 295

Page 8: Personality and job performance: the importance of narrow traits

Factor analyses of personality scales

Principal components analysis was performed at the scale level on all 21 personality scales. Scale-level analyses were performed, instead of item-level analyses, because the aim of this study was todetermine whether the constructs de®ned by the questionnaire and adjective scales bear therelationships to each other and to the criterion variables which Ones and Viswesvaran wouldpredict; the factor loadings of particular adjectives or questionnaire items were not of interest.

Table 4 shows the loadings of the personality scales on the ®rst unrotated factor (whichaccounted for 20.7 per cent of the scales' variance), and the correlation of that factor withworkplace delinquency. Instead of showing high loadings for responsibility, risk taking, con-scientiousness, and agreeablenessÐas would be expected of an integrity factorÐthis factor isprimarily de®ned by such scales as breadth of interest, complexity, self esteem, (low) conformity,extraversion, and emotional stability. The correlation of this ®rst unrotated factor with overallworkplace delinquency is approximately zero. Therefore, this factor clearly cannot be describedas a broad integrity factor.

Although a general integrity factor has failed to emerge, there remains the possibility thatbroad personality factors, such as the Big Five, might outpredict the narrow scales they comprise.Table 5 displays both the varimax-rotated solution following extraction of ®ve factors (whichaccounted for 62.0 per cent of the total scale variance) and also the correlations of these factorswith the workplace delinquency composite.

Table 3. Correlations of workplace delinquency items and overall composite with personality scales

Delinquency item/overall compositeAbsent Late Alcohol Unsafe Goldbrick Theft Freebies Vandal Overall

Questionnaire scalesBreadth of interest ÿ0:02 ÿ0:04 ÿ0:07 ÿ0:06 ÿ0:06 ÿ0:05 ÿ0:16 0.01 ÿ0:09Complexity 0.03 0.12 ÿ0:10 0.14 0.06 ÿ0:02 ÿ0:15 0.09 0.04Innovation 0.08 0.07 ÿ0:04 0.10 ÿ0:06 ÿ0:06 ÿ0:12 0.09 ÿ0:01Tolerance ÿ0:13 0.03 ÿ0:03 ÿ0:17 ÿ0:06 ÿ0:04 ÿ0:17 ÿ0:10 ÿ0:13Interpersonal a�ect ÿ0:01 0.14 ÿ0:03 ÿ0:17 ÿ0:05 ÿ0:14 ÿ0:22� ÿ0:16 ÿ0:13Anxiety 0.05 ÿ0:03 0.09 ÿ0:11 ÿ0:02 ÿ0:13 ÿ0:02 ÿ0:15 ÿ0:08Conformity 0.03 0.01 0.04 ÿ0:11 0.02 ÿ0:13 ÿ0:04 ÿ0:13 ÿ0:07Social participation ÿ0:04 ÿ0:02 0.07 0.05 ÿ0:03 0.11 0.09 0.03 0.06Self esteem 0.09 0.04 ÿ0:02 0.22� ÿ0:01 0.07 ÿ0:01 0.09 0.09Energy level ÿ0:00 ÿ0:12 ÿ0:02 0.08 ÿ0:09 ÿ0:03 0.03 0.00 ÿ0:03Social adroitness 0.19� 0.15 0.13 0.14 0.10 0.04 0.05 0.07 0.16Risk taking 0.13 0.06 0.20� 0.24{ 0.24{ 0.25{ 0.16 0.19� 0.30{Organization ÿ0:11 ÿ0:06 ÿ0:02 ÿ0:10 ÿ0:14 ÿ0:08 ÿ0:06 ÿ0:12 ÿ0:14Value orthodoxy 0:02 ÿ0:10 ÿ0:04 0.05 ÿ0:16 ÿ0:23y ÿ0:08 ÿ0:14 ÿ0:16Responsibility ÿ0:23y ÿ0:21� ÿ0:36y ÿ0:20� ÿ0:25y ÿ0:28y ÿ0:30y ÿ0:20� ÿ0:40yAggression ÿ0:00 0.08 0.02 0.04 0.05 0.05 0.10 ÿ0:03 0.07

Adjective scalesExtraversion 0.08 0.05 0.05 0.12 0.05 0.07 ÿ0:03 0.07 0.09Agreeableness ÿ0:14 ÿ0:03 ÿ0:10 ÿ0:17 ÿ0:12 ÿ0:15 ÿ0:26y ÿ0:09 ÿ0:21�Conscientiousness ÿ0:26y ÿ0:19� ÿ0:12 ÿ0:11 ÿ0:17 ÿ0:06 ÿ0:12 ÿ0:13 ÿ0:22�Emotional stability ÿ0:00 ÿ0:14 ÿ0:05 0.06 0.04 0.12 ÿ0:01 0.16 0.04Intellect ÿ0:05 0.03 0.03 0.02 0.03 0.01 ÿ0:17 0.04 ÿ0:01

{ p < 0:01; � p < 0:05.See Appendix for workplace delinquency items.

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

296 M. C. ASHTON

Page 9: Personality and job performance: the importance of narrow traits

The factors shown in Table 5 are clearly interpretable as the Big Five, with each of the ®veadjective scales univocally de®ning one factor. In this solution, JPI responsibility showedsubstantial loadings on the agreeableness and conscientiousness factors, while JPI risk takingproduced important loadings on the conscientiousness and openness/intellect factors.

The correlations of workplace delinquency with the agreeableness and conscientiousnessfactors (not to be confused with the adjective scales of the same names) were ÿ0:25 and ÿ0:30,respectively; no other factor yielded a statistically signi®cant correlation with the delinquencycomposite. The correlation of the unit-weighted sum of these two factors with overall delin-quency is ÿ0:38, which, although substantial, is actually less than the ÿ0:45 correlation obtainedby the unit-weighted combination of two narrow scales, responsibility and risk taking. Because ofthe high correlation between these two composite predictors (0.76), the di�erence between thesecorrelations in favor of the narrow scales is not statistically signi®cant, according to Meng,Rosenthal and Rubin's (1992) Z-test for di�erences between correlated correlations (Z � 1:25,p < 0:21). However, this non-signi®cant superiority of the narrow scales essentially eliminates thepossibility that the broad factors might have shown appreciably superior validity in a largersample, because the 95 per cent con®dence interval for the di�erence between these correlations(using the formula from Meng et al., 1992) does not contain a di�erence equal in size butopposite in sign to the one found in this study.

It could be noted that another method of determining whether there existed a general integrityfactor would have been to use an oblique rotation of the ®ve factors. Evidence of such a factor

Table 4. Loadings of questionnaire and adjective scales on ®rstunrotated factor, and correlation of that factor with overalldelinquency

Factor

Questionnaire scalesBreadth of interest 0.69Complexity 0.64Innovation 0.58Tolerance 0.56Interpersonal a�ect 0.17Anxiety ÿ0:40Conformity ÿ0:61Social participation 0.26Self esteem 0.62Energy level 0.38Social adroitness ÿ0:39Risk taking 0.45Organization ÿ0:18Value orthodoxy ÿ0:09Responsibility 0.36Aggression ÿ0:42

Adjective scalesExtraversion 0.55Agreeableness 0.34Conscientiousness ÿ0:07Emotional stability 0.58Intellect 0.45

Overall delinquency ÿ0:01

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

PERSONALITY AND JOB PERFORMANCE 297

Page 10: Personality and job performance: the importance of narrow traits

would be provided by strong correlations among the conscientiousness, agreeableness, andneuroticism factors. However, an oblimin rotation of the ®ve factors showed essentially zerocorrelations among these three factors: ÿ0:06 between agreeableness and neuroticism, 0.02between agreeableness and conscientiousness, and ÿ0:04 between conscientiousness andneuroticism.

Discussion

The results of this study showed that there was no integrity-related general factor of personality,and that broad personality measures were slightly less correlated with workplace delinquencythan were narrow measures, despite the brevity of the latter. Several possible concerns about thegenerality of these ®ndings are discussed below, along with recommendations based on these®ndings.

Concerns might be raised about the modest sample size of the present study, in which therewere 127 respondents. Although it is true that the factor loadings or criterion validities obtainedon a sample of this size will di�er somewhat from their population values, it is also true that asample size of 127 is su�cient to obtain a fairly close approximation to the true factor structure

Table 5. Loadings of questionnaire and adjective scales on ®ve varimax-rotated factors, and correlation ofthose factors with overall delinquency

Factors1 2 3 4 5

Questionnaire scalesBreadth of interest 0.42 ÿ0:23 0.56 ÿ0:09 0.11Complexity 0.38 ÿ0:10 0.64 ÿ0:03 0.11Innovation 0.00 ÿ0:02 0.85 ÿ0:11 0.15Tolerance 0.72 ÿ0:19 0.12 ÿ0:18 0.11Interpersonal a�ect 0.66 0.47 0.10 ÿ0:05 0.07Anxiety ÿ0:09 0.77 ÿ0:03 0.14 0.10Conformity 0.04 0.71 ÿ0:20 0.03 ÿ0:35Social participation 0.34 0.25 ÿ0:15 ÿ0:19 0.63Self esteem 0.01 ÿ0:22 0.25 0.08 0.81Energy level ÿ0:15 ÿ0:23 0.35 0.26 0.35Social adroitness ÿ0:36 0.43 0.15 ÿ0:16 ÿ0:25Risk taking ÿ0:20 ÿ0:26 0.41 ÿ0:44 0.29Organization ÿ0:11 0.04 ÿ0:08 0.81 0.03Value orthodoxy 0.12 0.19 0.05 0.64 ÿ0:05Responsibility 0.56 ÿ0:13 0.16 0.44 ÿ0:02Aggression ÿ0:70 0.30 0.03 0.17 0.11

Adjective scalesExtraversion ÿ0:02 ÿ0:09 0.22 ÿ0:03 0.85Agreeableness 0.72 0.04 0.09 0.24 ÿ0:00Conscientiousness ÿ0:07 ÿ0:07 0.00 0.83 0.03Emotional stability 0.18 ÿ0:80 0.12 ÿ0:07 0.07Intellect 0.02 0.05 0.79 0.14 0.04

Overall delinquency ÿ0:25� ÿ0:03 0.02 ÿ0:30� 0.10

� p < 0:01.

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

298 M. C. ASHTON

Page 11: Personality and job performance: the importance of narrow traits

and the true pattern of correlations with criteria. If an integrity-related general factor ofpersonality did in fact exist, a sample size of 127 would be large enough to detect it, just asthe general intelligence factor obtained from mental ability tests is routinely obtained frommodest samples, even when the range of intelligence is restricted. Similarly, if the Big Five factorswere appreciably more valid than the narrow dependability composite, then a sample size of127 would be very unlikely to produce results suggesting modest superiority for the narrowscales.

The fact that the narrow dependability composite produced a validity somewhat (though non-signi®cantly) superior to that of the Big Five factors is perhaps surprising in light of the numbersof items involved. While the abbreviated JPI responsibility and JPI risk taking scales used in thisstudy contained only 16 items altogether, the two Big Five factors were de®ned by scales thataccounted for approximately two-®fths of the total pool of 228 items. An employer interested inpredicting workplace delinquency (and hence, job performance) in a sample of entry-level jobapplicants would be better advised to use the full-length, 20-item versions of the responsibilityand risk taking scales than to use the combinations of scales (comprising about 80±90 items intotal) that de®ne the conscientiousness and agreeableness factors. Not only would the increasedlength of the narrow scales improve their reliability and increase their validity further beyond thatof the Big Five factors, but the total number of items (40) would still be only half the numbercontained by those factors. This would allow the additional inclusion of, for example, scales todetect faking or to measure personality traits related to certain other aspects of job performancefor the particular job in question.

It has been noted in previous studies (Ashton et al., 1995; Goldberg, 1993) that comparisons ofthe validities of personality factors and their constituent scales must take into account the factthat the narrower measures may produce, by virtue of their greater number, spuriously highmultiple correlations which are unlikely to survive cross-validation. This problem has been dealtwith in this study by concentrating only on the two narrow scalesÐresponsibility and risktakingÐthat had the clearest substantive relationships to the criteria, and by using a unit-weighted sum of these variables, rather than a multiple regression equation, in prediction.Therefore, because the narrow measures did not outnumber the factors, and because the weightsapplied to these measures were not sample-speci®c, the validity of the narrow predictors wouldnot be expected to shrink, any more than the validity of the factors might shrink, in any newsituation involving the same criteria.

Concerns might also be raised about the skewed distributions of respondents' scores on thecriterion variables (an almost inevitable result when dealing with delinquent behavior), and theinterpretation of correlations involving these variables. In order to ensure that correlations werenot in¯ated by the small numbers of respondents in the higher categories of delinquency, theanalyses reported above were performed again, this time collapsing the four highest categories oflateness and of goldbricking (thereby producing less-skewed, trichotomous variables), andcollapsing the ®ve highest categories of the other six delinquency criteria (thereby producing less-skewed, dichotomous variables). The new overall delinquency composite correlated ÿ0:48 withthe dependability composite (responsibility minus risk taking), and ÿ0:44 with the sum of theconscientiousness and agreeableness factors. Thus, the broader measures still failed to outpredictthe much shorter narrow measures, and the validities of the predictors were actually somewhatincreased. However, these new correlations may slightly overestimate the validity of the person-ality predictors, which should be expected to di�erentiate people with very high levels ofdelinquency (i.e. the people who incur the heaviest costs on an employer, and who thereforeclearly should not be hired) from those with only moderate levels, and not merely to di�erentiatethose with moderate or high levels from those with low levels.

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

PERSONALITY AND JOB PERFORMANCE 299

Page 12: Personality and job performance: the importance of narrow traits

It is somewhat unfortunate both that the sample of respondents that was used in this study waslimited to university students, and that no independent, non-self-report measure of workplacedelinquency was available. However, it is important not to overestimate these limitations.Although the respondents were ®rst-year university students, they were responding to questionsregarding their paid employment, which was, for most respondents, full-time during the summersand part-time during the other months of their high school years. Although the criterion measureof job performance was a self-report questionnaire regarding workplace delinquency, thecircumstances of the respondents' participation were such that there was no incentive to giveinaccurate responses. It is di�cult to imagine how the relative validities of broad versus narrowpersonality measures would have di�ered, had the respondents been drawn from some othersample of entry-level personnel, or had there been some independent assessment of workplacedelinquency (the key determinant of job performance in many entry-level jobs). It is even moredi�cult to imagine why a general integrity factor of personality would be any more likely toemerge in a di�erent sample of respondents. Nevertheless, it will be interesting to see the resultsof future studies addressing the same questions, but using supervisor-rating criteria for samplesof current entry-level employees.

One might wonder whether the predictor±criterion relationships found in this study were duein considerable part to response styles, such as social desirability, common to the predictors andthe criteria. For the two best predictors of workplace delinquencyÐJPI responsibility and JPIrisk takingÐthis seems unlikely, for several reasons. First, as mentioned in the Introduction,previous studies have found signi®cant relationships between these variables and various non-self-report measures, such as peer ratings of trustworthiness, behavior in hypothetical ethicaldilemmas, and dishonest behaviors in laboratory experiments. Second, risk taking is onlyweakly negatively correlated with social desirability (Jackson, 1994), and several scales(e.g. anxiety, self esteem, aggression) that have moderate social desirability saturations, similarto that of responsibility (Jackson, 1989, 1994), were virtually uncorrelated with workplacedelinquency.

The fact that the responsibility and risk taking scales slightly outpredicted Big Five factors thatwere de®ned, in part, by those scales themselves suggests that the Big Five personality factors arenot likely to provide maximal validity in certain predictive contexts, even though Big Five traitshave been found repeatedly in lexical analyses involving several languages and in a wide range ofpersonality questionnaires. In this regard the Big Five traits do not compare favorably with the gof mental ability tests, even though the proportion of personality scale variance accounted for bythe Big Five exceeds the proportion of mental ability test variance accounted for by g. Thepredictive validity shown by mental ability tests with respect to many academic, job performance,and other behavioral criteria is often almost entirely due to the variance that is common to thosetests. By contrast, optimal prediction of an important and fairly broad criterion such as work-place delinquency (and hence, in this sample, overall job performance) requires variance speci®cto certain narrow personality traits in addition to variance attributable to the Big Five factors.Perhaps personality traits such as responsibility and risk taking should be considered asconstructs analogous to, but predictively more important than, the `group factors' (e.g. verbal,spatial, perceptual, memory) of mental ability tests.

The results of the present study suggest that integrity-related scales, such as JPI responsibilityand JPI risk taking, are likely to be good predictors of job performance in entry-level jobs. Inmany jobs, however, especially non-entry-level jobs, performance involves much more thanmere non-delinquency, and will likely be predicted by other personality traits. Analyses of thedemands of a given job should provide insights into the particular personality variables that willrelate to performance (Robertson, 1993; Tett, Jackson and Rothstein, 1991). It is likely that for

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

300 M. C. ASHTON

Page 13: Personality and job performance: the importance of narrow traits

some jobs, performance will be predicted optimally by broad measures, such as the Big Fivefactors. But the results of this study indicate that this is unlikely to be the case for manyother jobs, and that the use of narrow personality scales will be necessary to achieve maximumvalidity. Exclusive reliance upon broad personality measures is likely to limit validity in manyemployment contexts.

Conclusions

The results of this study suggest that integrity-related personality traits do not de®ne a generalfactor of personality, and that these narrow traits are better predictors of job performance thanare the factors that subsume them. Optimal prediction of job performance will sometimes requirethe use of narrow personality measures as predictors.

Acknowledgements

The author thanks Richard D. Go�n, Douglas N. Jackson, and Paul F. Tremblay for theircomments on an earlier version of this paper. Errors remain the responsibility of the authoralone.

References

Ashton, M. C., Jackson, D. N., Paunonen, S. V., Helmes, E. and Rothstein, M. G. (1995). `The criterionvalidity of broad factor scales versus speci®c facet scales', Journal of Research in Personality, 29,432±442.

Barrick, M. R. and Mount, M. K. (1991). `The big ®ve personality dimensions and job performance:A meta-analysis', Personnel Psychology, 44, 1±26.

Barrick, M. R. and Mount, M. K. (1994, April). `Do speci®c components of Conscientiousness predictbetter than the overall construct?' In: Page, R. (Chair) Personality and Job Performance: Big Five versusspeci®c traits, Symposium conducted at the annual meeting of the Society for Industrial andOrganizational Psychology, Nashville, TN.

Goldberg, L. R. (1977). `What if we administered the `wrong' inventory? The prediction of scores onPersonality Research Form scales from those on the California Psychological Inventory, and vice versa',Applied Psychological Measurement, 1, 339±354.

Goldberg, L. R. (1992). `The development of markers for the Big-Five factor structure', PsychologicalAssessment, 4, 26±42.

Goldberg, L. R. (1993). `The structure of personality traits: vertical and horizontal aspects. In: Funder,D. C., Parke, R. D., Tomlinson-Keasey, C. and Widaman, K. (Eds) Studying Lives Through Time:Personality and Development, American Psychological Association, Washington, DC.

Gough, H. G. (1957). California Psychological Inventory, Consulting Psychologists Press, Palo Alto, CA.Jackson, D. N. (1970). Jackson Personality Inventory, Sigma Assessment Systems, Port Huron, MI.Jackson, D. N. (1989). Personality Research Form Manual, 3rd edn, Sigma Assessment Systems, PortHuron, MI.

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

PERSONALITY AND JOB PERFORMANCE 301

Page 14: Personality and job performance: the importance of narrow traits

Jackson, D. N. (1994). Jackson Personality InventoryÐRevised Manual, Sigma Assessment Systems, PortHuron, MI.

Jackson, D. N., Hourany, L. and Vidmar, N. J. (1972). `A four-dimensional interpretation of risk taking',Journal of Personality, 40, 483±501.

Meng, X.-L., Rosenthal, R. and Rubin, D. B. (1992). `Comparing correlated correlation coe�cients',Psychological Bulletin, 111, 172±175.

Michigan Department of Education (1989). The Michigan Employability Survey, Author, Ann Arbor, MI.Mikulay, S. M. and Go�n, R. D. (1996). `Discriminant and criterion-related validity of an integritymeasure', manuscript submitted for publication.

Nunnally, J, C. (1978). Psychometric Theory, McGraw-Hill, New York.Ones, D. S., Schmidt, F. L. and Viswesvaran, C. (1994, April). `Do broader personality variables predictjob performance with higher validity?' In: Page, R. (Chair) Personality and Job Performance: Big Fiveversus speci®c traits, Symposium conducted at the annual meeting of the Society for Industrial andOrganizational Psychology, Nashville, TN.

Ones, D. S. and Viswesvaran, C. (1996). `Bandwidth±®delity dilemma in personality measurement forpersonnel selection', Journal of Organizational Behavior, 17, 609±626.

Ones, D. S., Viswesvaran, C. and Schmidt, F. L. (1993). `Meta-analysis of integrity test validities', Journalof Applied Psychology, 78, 679±703.

Paunonen, S. V. (1993). `Sense, nonsense, and the Big Five factors of personality'. Invited address,Division 5 of the American Psychological Association, Toronto.

Reynolds, C. H. and Nichols, R. C. (1977). `Factor scales of the CPI: Do they capture the valid variance?'Educational and Psychological Measurement, 37, 907±915.

Robertson, I. T. (1993). `Personality assessment and personnel selection. Special issue: the validity andutility of personality assessment in occupational psychology', European Journal of Applied Psychology,43, 187±194.

Tett, R. P., Jackson, D. N. and Rothstein, M. G. (1991). `Personality measures as predictors of jobperformance: A meta-analytic review', Personnel Psychology, 44, 703±742.

Appendix: Workplace Behavior Questionnaire

Please answer the following questions as accurately as possible, by circling the appropriatealternative. Remember that your responses are completely anonymous; there is no need to try tomake a good (or bad) impression.

During high school, how many hours per week did you work for pay, on average, during theACADEMIC YEAR? (Only include semesters during which you were enrolled as a student).

0 h 1±4 h 5±9 h 10±14 h 15±19 h 20±29 h > 30 h

During high school, how many hours per week did you work for pay, on average, during theSUMMER? (Also include semesters during which you were not enrolled in school).

0 h 1±9 h 10±19 h 20±29 h 30±39 h 40±49 h > 50 h

On what percentage of your scheduled work shifts have you called in sick and/or with afamily crisis, when you actually were not currently sick and did not have an immediate familycrisis?

0 per cent 1±3 per cent 4±9 per cent 10±19 per cent 20±34 per cent > 35 per cent

On what percentage of your scheduled work shifts did you arrive at work late?

0 per cent 1±3 per cent 4±9 per cent 10±19 per cent 20±34 per cent > 35 per cent

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

302 M. C. ASHTON

Page 15: Personality and job performance: the importance of narrow traits

On what percentage of your scheduled work shifts have you either (a) consumed alcohol, or(b) been at least somewhat under the in¯uence of alcohol? (Do not include instances in whichyour supervisor speci®cally allowed alcohol consumption).

0 per cent 1±3 per cent 4±9 per cent 10±19 per cent 20±34 per cent > 35 per cent

On what percentage of your shifts did you deliberately violate your workplace's safetystandards (e.g. by driving too fast, handling dangerous materials or machinery without properequipment, etc.)?

0 per cent 1±3 per cent 4±9 per cent 10±19 per cent 20±34 per cent > 35 per cent

What percentage of the time on an average work shift did you spend NOT doing what you weresupposed to be doing, but instead relaxing, socializing, hiding, etc.? (Do not include breaks thatwere speci®cally allowed by your supervisor, except where you took breaks longer than thesupervisor speci®ed).

55 per cent 5±9 per cent 10±19 per cent 20±29 per cent 30±49 per cent > 50 per cent

What is the total dollar value of cash, merchandise, and equipment that you have stolen fromyour workplaces, or used or consumed at your workplaces without authorization and withoutpaying?

$0 $1±$19 $20±$49 $50±$99 $100±$199 >$200

What is the total dollar value of goods or services that you have given to your friends orrelatives for free from your workplaces? (Do not include instances in which your supervisorspeci®cally allowed this).

$0 $1±$19 $20±$49 $50±$99 $100±$199 >$200

What is the total dollar value of damage or lost production you have caused at yourworkplaces through deliberate vandalism or sabotage?

$0 $1±$19 $20±$49 $50±$99 $100±$199 >$200

# 1998 John Wiley & Sons, Ltd. J. Organiz. Behav. 19: 289±303 (1998)

PERSONALITY AND JOB PERFORMANCE 303