Journal of Applied Psychology1996, Vol. 81, No. 3,261-272

Copyright 1996 by the American Psychological Association, Inc.0021-9010/96/$3.00

Effects of Impression Management and Self-Deception on the PredictiveValidity of Personality Constructs

Murray R. Barrick and Michael K. MountUniversity of Iowa

This study tests whether 2 types of response distortion (self-deception and impressionmanagement) affect the predictive validity of 2 of the "Big 5" personality dimensions,conscientiousness and emotional stability, in 2 applicant samples of long-haul semitruckdrivers (n = 147 and n= 139). As hypothesized, conscientiousness ( p = —.26 and —.26)and emotional stability (p - —.23 and —.21) were valid predictors of voluntary turnoverin the 2 samples. Also as hypothesized, conscientiousness was a valid predictor of super-visory ratings of performance (p = .41 and .39) in the 2 samples. Although not hypothe-sized, emotional stability was also significantly related to supervisor ratings of perfor-mance (p = .23 and .27). Results from structural equations modeling indicated thatapplicants did distort their scores on both personality dimensions and the distortion oc-curred both through self-deception and impression management; however, neither typeof distortion attenuated the predictive validities of either personality construct.

The emergence and widespread acceptance of the five-factor framework of personality in combination with re-sults of recent construct-oriented meta-analyses of per-sonality and job performance (Barrick & Mount, 1991;Hough, 1992; Hough, Eaton, Dunnette, Kamp, &McCloy, 1990; Mount & Barrick, 1995a) have led to re-newed interest in personality measures for selectionpurposes. Although a great deal of progress has beenmade in recent years in understanding relationships be-tween personality constructs and job performance indifferent occupations, several important issues have re-ceived relatively little attention.

The primary purpose of this study was to examine theeffect of response distortion on the predictive validity ofpersonality constructs. Our focus was not so much onwhether response distortion influences validity (as mostof the research indicates it does not), but rather whetherthe type of response distortion differentially influencesvalidity. The second purpose was to assess the extent towhich personality measures are useful for predicting vol-

Murray R. Barrick and Michael K. Mount, Department ofManagement and Organizations, University of Iowa.

We thank Laurie Reilly for her help in collecting the data. Anearlier version of this article was presented at the 10th AnnualConference of the Society for Industrial and OrganizationalPsychology, Orlando, Florida, May 1995.

Correspondence concerning this article should be addressedto Murray R. Barrick, Department of Management and Orga-nizations, College of Business Administration, S358 PBAB,University of Iowa, Iowa City, Iowa 52242. Electronic mail maybe sent to mbarrick@uiowa.edu.

untary turnover. Although prior meta-analyses have as-sessed the relationship between personality constructsand withdrawal behaviors, these analyses have not distin-guished between withdrawal initiated by the organization(involuntary turnover) and withdrawal initiated by theindividual (voluntary turnover). This distinction is im-portant because the predictors of involuntary turnoverare likely to differ from those of voluntary turnover(Abelson, 1987; Campion, 1991).

This study uses the five-factor model of personality(FFM), frequently referred to as the Big Five. The FFMprovides a well-accepted taxonomy that enhances under-standing of the relationship between personality con-structs and important organizational criteria. The con-struct labels and representative traits of the FFM are Ex-traversion (sociable, talkative, active, and ambitious);Agreeableness (courteous, trusting, cooperative, andempathic); Conscientiousness (dependable, organized,persistent, and achievement-oriented); Emotional Sta-bility (calm, unemotional, secure, and not angry); andOpenness to Experience (imaginative, cultured, broad-minded, and flexible).

Response Distortion

It is often assumed that applicants differ from currentemployees in their motivation to make themselves "lookbetter." Given the seemingly straightforward or transpar-ent nature of some of the items, it seems likely that someapplicants will try to "beat" the test (i.e., get hired whenthey should not). To illustrate, a few examples of com-monly used personality items are "I am very thorough in

26!

262 BARRICK AND MOUNT

any work I do; "I am very dependable; and "I am verysociable." Evidence is clear that applicants can makethemselves look better on such items if they choose to doso (Braun, 1962; Dunnett, Koun, & Barber, 1981;French, 1958; Hough et al., 1990; White, Nord, Mael, &Young, 1993; Zalinsky & Abrahams, 1979). Further-more, most evidence suggests that there is more responsedistortion among applicants than among job incumbents(Bass, 1957; Dunnette, McCartney, Carlson, & Kirchner,1962; Kirchner, Dunnette, & Mousley, 1960; Michaelis& Eysenck, 1971). A few studies suggest that the inci-dence of response distortion on personality inventories inreal applicant settings is low (McClelland & Rhodes,1969; Orpen, 1971; Schwab & Packard, 1973). However,very few of these studies have distinguished between thetypes of distortion that are possible.

Most studies investigating the effects of response dis-tortion have assumed that it is a unitary construct. How-ever, Paulhus and associates (Paulhus, 1984;1988; 1989;Paulhus & Reid, 1991; Zerbe & Paulhus, 1987) havedemonstrated that response distortion (also referred toas social desirability) consists of two separate constructs:self-deception and impression management. Self-decep-tion is a dispositional tendency to think of oneself in afavorable light, whereas impression management refersto a deliberate attempt to distort one's responses in orderto create a favorable impression with others.

Paulhus and associates have collected compelling evi-dence for the construct validity of these two constructs.In a series of factor analyses, Paulhus (1984; 1988; Paul-hus & Reid, 1991) found strong discriminant validity forthese two scales across many different response distortionscales (e.g., Byrne's Repression-Sensitization scale, Ed-wards' Social Desirability scale, Eysenck's lie scale, Ihile-vich & Gleser's Defense Mechanisms Inventory, the Min-nesota Multiphasic Personality Inventory Lie scale, theMarlowe-Crowne scale, and Wiggins Social Desirabilityscale). Measures of self-deception reflect a positively bi-ased, psychologically well-adjusted self-presentation.High-scoring individuals tend to exhibit more illusion ofcontrol, have excessive confidence in memory judgments,and claim more familiarity with nonexistent products(Paulhus, 1988; Paulhus & Reid, 1991). In contrast,measures of impression management correlate stronglywith traditional lie scales (e.g., Eysenck's lie scale, theMMPI lie scale, and Wiggins SD scale). They appear toreflect a conscious form of self-presentation motivated bya desire to obtain social approval. Paulhus (1984)showed that scores on the impression management con-struct are particularly responsive to social demands. Hereported that under public, as opposed to annonymousadministrations, impression management resulted in sig-nificantly higher scores, whereas scores on the self-decep-tion construct were not significantly higher. Overall, these

studies showed that widely used response distortion mea-sures form separate constructs best characterized as self-deception and impression management.

Several different methods have been developed formanaging the potential problem of response distortionin personality assessment (for a thorough review of thesemethods, see Paulhus, 1988). The most frequently usedmethod is to include response distortion scales, oftencalled lie scales or social desirability scales, on personal-ity inventories. Scores on these lie scales are then used todetect those who may be attempting to present them-selves in a favorable light. Typically, the effects of re-sponse distortion are then partialed out of responses onthe other personality scales through covariate techniques.

Particularly informative, though rare, are criterion-related validity studies in which the personality-scalescores obtained from applicants have been corrected forresponse distortion. In one study (Christiansen, Goffin,Johnston, & Rothstein, 1994), personality and supervi-sory ratings of performance were obtained from 84 as-sessment center candidates. Response distortion was de-tected with the two response validity scales ("fakinggood" and "faking bad") recommended for use with theSixteen Personality Factor (16PF). Both of these scalescorrespond to the impression management construct. Toassess whether response distortion produced spuriousobserved validities, the two response validity scales werepartialed from the relevant personality scores and ratingsof performance. Results indicated that correction for re-sponse distortion (i.e., impression management) had lit-tle effect on criterion-related validity.

Ones, Viswesvaran, and Reiss (1995) used meta-analytic data to estimate the effect of partialing social de-sirability from the Big Five dimensions in the predictionof job performance. To do this, two meta-analytically cu-mulated correlations were used. The first used the corre-lation between social desirability scale scores and the BigFive dimensions of personality, and the second used thecorrelations between social desirability and job perfor-mance. Applying the formula for a semipartial correla-tion, Ones et al. reported that partialing social desirabil-ity from the Big Five personality dimensions did not at-tentuate the criterion-related validities of the Big Fivevariables. They concluded that controlling for social de-sirability has little effect on the predictive validity of rele-vant personality dimensions.

Similarly, Hough et al. (1990) reported no significantdifferences in the predictive validities of a large group ofsoldiers (N = 5,896-5,997) that scored lower on the so-cial desirability score compared with soldiers (N =2,428-2,480) who might have been trying to look good,as suggested by their higher social desirability scores. Ina second study (Hough et al., 1990), the Assessment ofBackground and Life Experiences (ABLE) personality

RESPONSE DISTORTION 263

inventory was administered twice to 245 newly enlistedsoldiers, once instructing the subjects to respond ashonestly as possible and the second time instructing themto either fake "good" or fake "bad" on the inventory. Re-sults demonstrated that when instructed to do so, soldiersdid distort their responses. Thus, Hough et al. (1990)showed that even though response distortion did not re-duce predictive validities, soldiers were able to distorttheir responses when instructed to fake. Although someresearch shows that response distortion lowers validity(e.g., Dunnette et al., 1962), the preponderance of evi-dence shows that response distortion does not attenuatethe validity of personality measures (e.g., Christiansenet al., 1994; Hough et al., 1990; Ones, Viswesvaran, &Schmidt, 1993).

However, an important question that remains unan-swered is whether the effect of response distortion dependson the type or nature of response distortion studied. It ispossible that the effects of self-deception may be quitedifferent from the effects of impression management. Highscorers on self-deception (Krug, 1978; McCrae & Costa,1983; Paulhus, 1989; Paulhus & Reid, 1991; Zerbe & Paul-hus, 1987) tend to be well-adjusted people who have posi-tively biased self-images and perceive themselves as achieve-ment-oriented. It is possible that self-deception, thoughconceptually distinct, is substantively related to emotionalstability and to a lesser extent conscientiousness. This sug-gests that partialing the variance because of self-deceptionfrom conscientiousness and emotional stability may actu-ally remove some valid variance from these two predictors.Therefore, one question we investigated is whether partial-ing self-deception from personality constructs will reducethe predictive validities.

High scorers on impression management tend to be ap-plicants who are motivated to intentionally raise theirscores in order to "look better." This suggests that scoreson impression management should be substantiallyhigher if applicants do in fact intentionally distort theirresponses when applying for a job because of a strong de-sire to "look better." However, as previously noted, evenif impression management occurs, the literature suggeststhat the predictive validity of personality constructs willnot be attenuated. To examine this issue, we comparedthe effects of partialing impression management on pre-dictor validities with the "unpartialed" validities for bothpersonality constructs by using voluntary turnover as thecriterion. To our knowledge, the effects of response dis-tortion have not been investigated with this criterion.

Voluntary Turnover

Most models of antecedents of voluntary turnoverhave focused on either the ease of employee movement(e.g., number of job alternatives) or the perceived desir-

ability of employee movement (e.g., job satisfaction andjob attitudes) as antecedents of the employee's intentionsto quit (Horn & Griffeth, 1991; Hulin, Roznowski, &Hachiya, 1985; Mobley, 1977; Muchinsky & Morrow,1980; Steers & Mowday, 1981). Because practitioners aregenerally unable to influence whether alternative job op-portunities are available, researchers have devoted con-siderable attention to modifying job content and the workenvironment (e.g., group cohesion, leader-member ex-change, promotions, role clarity, conflict, etc.) to try toinfluence job satisfaction, which in turn is expected tolead to reduced turnover. Taken together, these studieshave attempted to reduce withdrawal cognitions (e.g., in-tent to search, thoughts of quitting, etc.)

Very little research has examined individual differ-ences at the time individuals apply for the job and theinfluence these differences have on decisions regardingvoluntary turnover. The potential value of such a per-spective has recently been illustrated by Arvey, Bou-chard, Segal, and Abraham (1989). They demonstratedthat job satisfaction appears to have a substantial geneticcomponent. Job satisfaction, in turn, has been shown tobe an important antecedent of voluntary turnover(Carsten & Spector, 1987; Cotton & Tuttle, 1986; Horn,Caranakis-Walker, Prussia, & Griffeth, 1992; Steel &Ovalle, 1984; Tett & Meyer, 1993). Therefore, applicantsmay have a dispositional tendency to leave or remain witha firm. Recent research also has demonstrated that indi-viduals who view life negatively (i.e., lower emotionalstability) are more prone to absenteeism, intentions toquit, and voluntary turnover (George, 1989; 1990; Judge,1992). Furthermore, Barrick, Mount, and Strauss(1994) recently illustrated that one dimension of person-ality, conscientiousness, was a valid predictor of involun-tary turnover. Although the antecedents of voluntaryturnover are likely to differ from those of involuntaryturnover, we believe that conscientiousness should also bea valid predictor of voluntary turnover. Because consci-entious employees are more responsible and reliable,they are more likely to be involved in and committed tothe organization; according to the voluntary turnover lit-erature, such individuals are less likely to leave the orga-nization voluntarily (Mathieu & Zajac, 1990). Such re-sults suggest that personality or dispositional measuresplay a key role in reducing voluntary turnover. Thus, thesecond major purpose of this study was to examine thevalidity of specific personality constructs for predictingvoluntary turnover.

Although recent meta-analyses have assessed the rela-tionship between personality constructs and turnover, itshould be noted that there are considerably fewer studiesavailable than those that used more traditional criterionmeasures. Furthermore, these analyses failed to differ-entiate between voluntary and involuntary turnover.

264 BARRICK AND MOUNT

Nevertheless, the limited meta-analytic results avail-able indicate that conscientiousness is a valid predictorof turnover. Hough et al. (1990) reported observed valid-ities of .24 and .17 for two personality constructs,achievement and dependability (both aspects ofconscientiousness), respectively, for measures of job in-volvement. In their meta-analytic study, Barrick andMount (1991) reported observed validities of approxi-mately the same magnitude for measures of conscien-tiousness when predicting a broad category of criterionmeasures that they labeled withdrawal behaviors.

Recent data also suggests that emotional stability maybe negatively related to voluntary turnover. DeMatteo,White, Teplitzky, and Sachs (1991) reported that the bestpersonality predictor of 1-year attrition in the Army'sProject A data was emotional stability, although consci-entiousness (labeled dependability) was also a significantpredictor. This conclusion was also supported by themeta-analysis of Hough et al. (1990), where emotionalstability (labeled adjustment) was found to be signifi-cantly correlated with turnover (labeled job involve-ment).

Hypothesized Relations Among Variables

In the present study, data were collected from individ-uals applying for the same job (long-haul semitruckdriver) in two separate organizations. In both settings,the firms were very interested in reducing the rate of vol-untary turnover. Therefore, the success of applicants whowere hired was determined by measures of voluntaryturnover, as well as more traditional supervisory ratingsof job performance.

The proposed model that we tested posited relation-ships among two personality constructs, conscientious-ness and emotional stability, and two criteria, supervi-sory ratings of job performance and voluntary turnover.Specifically, we expected that individuals high in consci-entiousness would have higher ratings of job performanceand less voluntary turnover. Furthermore, those high inemotional stability were expected to have less voluntaryturnover. Finally, performance ratings were hypothesizedto be negatively correlated with voluntary turnover. Ameta-analysis by McEvoy and Cascio (1987) providedsupport for the latter hypothesis, as they found that vol-untary turnover had a true score correlation of—.31 withjob performance.

In addition, we examined the effects of the two-response distortion measures, self-deception and impres-sion management, on these predictor-criterion relation-ships. We expected that controlling for self-deceptionwould attenuate the validities between conscientiousnessand emotional stability with voluntary turnover and con-scientiousness with supervisory ratings of job perfor-

mance. Furthermore, although impression managementwas expected to affect responses to both conscientious-ness and emotional stability, these effects were not ex-pected to attenuate the predictive validities of these con-structs. We also examined the validity of extraversion,agreeableness, and openness to experience after control-ling for both types of response distortion; however, be-cause previous research (Barrick & Mount, 1991) sug-gested that these constructs were not valid predictors forthe job and criteria assessed in this study, no hypotheseswere tested.

Method

Participants

The sample consisted of job applicants hired by two trans-portation companies as long-haul semitruck drivers. The totalsample was 166 participants in Sample 1 and 153 participantsin Sample 2. However, 19 applicants in Sample 1 and 14 appli-cants in Sample 2 involuntarily left each firm; consequently, theactual sample size was 147 in Sample 1 and 139 in Sample 2.The typical participant was male, in his late 20s or early 30s,with a high-school education.

Procedure

As part of the hiring process, all applicants completed a per-sonality inventory, the Personal Characteristics Inventory(PCI), and a response distortion questionnaire, the BalancedInventory of Desirable Responding (BIDR), during the finalinterview. Applicants were informed later that they were actu-ally hired on the basis of the brief employment interview and abackground check. Although neither the personality inventorynor the response distortion questionnaire was used for hiring,comments to the test administrators demonstrated that appli-cants thought this information was part of the hiring decision.Applicants not hired (a total of 35 applicants were not hired; n= 21 and 14 in Samples 1 and 2, respectively) were told so atthat time and were excluded from consideration for thepurposes of this study.

Personal Characteristics Inventory (PCI). Form C of thePCI consists of 120 items designed to comprehensively measurethe Big Five personality constructs. Coefficient alpha reliabili-ties are .87, .85, .82, .86, .82, for Conscientiousness, Extraver-sion, Agreeableness, Emotional Stability, and Openness to Ex-perience, respectively. The PCI has been shown to have accept-able evidence of construct validity (for a more thoroughdescription of the item content, development methods, and con-struct validity, see Barrick & Mount, 1993; Barrick, Mount, &Strauss, 1993).

Balanced Inventory of Desirable Responding (BIDR). TheBIDR, Version 6 (Paulhus, 1988) is a 40-item Likert-type mea-sure with two subscales that measure self-deception and impres-sion management. In a series of factor analyses, Paulhus (1984;1988) demonstrated that measures of self-deception and im-pression management form two separate factors. Sample itemsfrom the 20-item self-deception scale include "I never regret my

RESPONSE DISTORTION 265

decisions" and "My first impressions of people usually turn outto be right." Sample items from the 20-item impression man-agement scale include "I always obey laws, even if I'm unlikelyto get caught" and "I never cover up my mistakes." Higherscores indicate greater self-deception and impression manage-ment. A 5-point Likert scale was used, but scores were con-verted according to the standard scoring procedure (Paulhus,1988). Thus, only extreme responses are counted; 1 point wasgiven for each 4 or 5 on both scales. Thus, total scale scorescould range from 0 to 20 for each scale. Paulhus (1988) re-ported that coefficient alphas range between .74 and .86 for thetwo subscales (alpha was .77 and .81 for self-deception and .83and .79 for impression management in the two samples,respectively).

Turnover. In both firms, turnover data were collected 6months after the applicant was hired. There were two turnovercategories of interest: stayers (n = 57 and 45) and voluntaryleavers (n = 90 and 94). The latter category refers to turnoverthat reflected the individual's choice to leave for Samples 1 and2, respectively. It should be noted that the turnover rate re-ported in these two organizations were quite similar to that tra-ditionally reported in the industry. Employment records wereused to identify the reason for turnover. In both organizations,turnover reasons were categorized as employee-initiated rea-sons (quitting) or organization-initiated reasons (being fired).Reasons for leaving were also obtained from the director of hu-man resources and the traffic dispatchers. Across all turnovercases, there was more than 90% agreement between the person-nel file and the human resources director and the dispatchers.Disagreements were resolved by obtaining additional informa-tion from the supervisors and coworkers about the reason forturnover.

Job performance ratings. In both organizations, the em-ployees' supervisors rated their job performance after a 30-dayprobationary period. Supervisors rode with the drivers through-out this period; consequently they were very familiar with thedriver's performance. Drivers were evaluated on nine-dimen-sions in the two organizations. These included quality of work,quantity of work, suitability for the position, personal appear-ance, attendance, dependability, driving skills, and oral andwritten communication skills. Performance was evaluated ona 5-point Likert scale from definitely unsatisfactory tooutstanding. Overall performance was the mean of the ratingsacross all dimensions. The alpha coefficicents were .75 and .83,respectively, in the two organizations.

Analyses. Two methods of analysis were used to assess thepotential effect of response distortion on the predictive validityof the personality measures. First, the personality scale was re-gressed on the response distortion score. The residual was thenused as a corrected score free of contamination from responsedistortion. This procedure results in a semipartial correlationthat corrected the unadjusted personality scores by an amountcommensurate with the contamination because of response dis-tortion, thereby making it possible to assess the effect of re-sponse distortion on the predictive validity of the personalityconstructs. In the second method, latent-variable modeling wasused to illustrate the relationship of response distortion to thevalidity of personality measures. Two recent articles illustratedthe advantages of using LISREL analysis (Schmitt, Nason,

Whitney, & Pulakos, 1995; Williams & Anderson, 1994). Sim-ilar to the procedure described in these two articles, we testeda series of nested models to examine whether either responsedistortion construct affected the predictors or criteria, or moreimportantly, whether distortion affected the correlation be-tween the predictors and criteria.

To maximize the power of the test for the effect of either re-sponse distortion construct, the data from both firms weremerged into one large data set (N = 286). In this analysis, eachlatent variable had three indicator variables, except for volun-tary turnover, which had a single indicator. Three indicator vari-ables were developed by randomly combining an equal numberof items from the supervisory ratings of job performance mea-sure described above. Three other indicator variables were sim-ilarly formed for both self-deception and impression manage-ment. Three subscales have previously been formed for consci-entiousness (Mount & Barrick, 1995b). They are dependability(being responsible, careful, and reliable), efficiency (having theability to plan, and being neat, orderly, punctual, anddisciplined), and industriousness (being hardworking, persis-tent, energetic, and achievement-striving). Thus, the sum of theitems for each subscale provided the three indicator variablesfor conscientiousness. The emotional stability measure con-sisted of two subscales; steadiness (patient, not jealous, even-tempered, and steady) and security (not nervous, secure, notworrisome, and not vulnerable). An equal number of itemswere randomly selected from each subscale to form the thirdindicator variable for emotional stability.

Finally, it should be noted that the point biserial correlationsfor the turnover constructs were corrected to biserial corre-lations (Cohen, 1984; Hunter & Schmidt, 1990) to reflect con-tinuous constructs. That is, even though the turnover classifi-cations were dichotomous (voluntary leavers vs. stayers) be-cause decisions to leave are a function of both the organizationand the employee, these measures may be better viewed as con-tinuous theoretical constructs (Campion, 1991). For example,an employee voluntarily deciding to leave may be better con-ceived as a continuum ranging from completely voluntary turn-over (e.g., the employee quits because of better pay elsewhere)through mutual agreement (e.g., the employee agrees to quitbecause of a disagreement with management) to involuntaryturnover (dismissed).

Results

Descriptive statistics and correlations between vari-ables uncorrected for statistical artifacts are presented inTable 1 for Samples 1 and 2. As shown in this table, thedirection and magnitude of the relations reported in bothsamples are comparable. As hypothesized, the resultsshowed that voluntary turnover was predicted by bothconscientiousness (r = —.23 and —.23; p = —.26 and—.26, respectively) and emotional stability (r = -.20 and—.18; p = -.23 and —.21, respectively) in both samples(p represents the true score correlation after correctingfor unreliability in the predictor and criterion). Also ashypothesized, supervisory ratings of job performancewere predicted by conscientiousness (r = .27 and .26; p

266 BARRICK AND MOUNT

Table 1Correlation Matrix for Applicants in Both Samples

Variable M SD 1

1 . Conscientious2. Emotional

Stability3. Agreeable4. Extraversion5. Openness6. Self-Deception7. Impression

Management8. Turnover9. Performance

2.67

2.432.522.132.315.23

10.170.644.58

0.31

0.390.350.310.354.30

3.890.710.84

(.87)

.55*

.52*

.24*

.42*

.24*

.39*-.23*

.26*

.52*

(.86).46*.13.39*.35*

.44*-.18*

.18*

* .51**

.47**(.82).34**.40**.17*

.42**-.13*

.00

.24**

.13

.09(.85)

32**.22**

.12-.12-.04

.47**

.41**

.39**

.37**(.82).17*

.39**

.10-.05

.29**

.54**

.24**

.25**

.32**(.77)

.51**-.10

.15*

.26**

.43**

.23**

.01

.25**

.43**

(.81)-.10-.07

.23**

-.20**-.13-.04-.05-.11

-.15*(.90)

-.23*

.27**

.15*

.06

.02

.07

.09

.17*-.20**(.50)

Note. Correlations reported above the diagonal are from Sample 1 (n = 147); correlations below the diagonal are from Sample 2 (« = 139). Valuesin parentheses represent coefficient alpha except for the criterion measures, which were assumed to be .90 for voluntary turnover and .50 forsupervisory ratings of job performance. The average interrater reliability of a single supervisor's rating of .50 was based on the average mean estimatereported by Rothstein (1990). For voluntary turnover, voluntary leavers = 1 and stayers = 0. Means and standard deviations represent average valuesacross Samples 1 and 2.* p < .05, one-tailed. **p < .01, one-tailed.

= .41 and .39, respectively) in both samples. Emotionalstability was also found to be a valid predictor of ratingsof performance (r = .15 and .18; p = .23 and .27,respectively). Furthermore, self-deception was signifi-cantly correlated with ratings of job performance in onesample (r = .15, p = .24), and impression managementwas significantly correlated with both voluntary turnoverand ratings of performance in one of the samples (r =-.15 and .17; p = -.18 and .27, respectively). Finally,ratings of performance were significantly correlated withvoluntary turnover in both samples (r = —.20 and —.23;p = -.30 and —.34, respectively).

An important purpose of this study was to establishwhether the validity of these two personality constructsfor predicting voluntary turnover differed after adjusting(partialing) for the effect of two facets of response distor-tion with the regression procedures described earlier.These adjusted validities are compared with the unad-justed validities in Tables 2 and 3. Table 2 shows the ad-justed validities for self-deception for conscientiousnessand emotional stability. As shown, they were smaller, butcontrary to our hypotheses, were not substantiallydifferent from the unadjusted correlations. Table 3 showsthat the adjusted validities for impression managementwere quite similiar to those for self-deception in that theadjusted validities are slightly smaller. As expected, noneof the differences between adjusted and unadjusted (forimpression management) correlations for conscientious-ness and emotional stability was substantially different.Finally, for the other personality constructs, none of thedifferences between unadjusted and adjusted validities inTables 2 and 3 was substantially different, although theadjusted validities generally were slightly smaller.

Structural equation modeling (LISREL; Joreskog &

Sorbom, 1989) was also used to investigate the role thatresponse distortion might play in over- or underestimat-ing the predictor-criterion relationships. These analysesare based on similar procedures proposed recently byothers (Schmittetal., 1995; Williams & Anderson, 1994)with an all-Y model specification. Table 4 provides themeans, standard deviations, and intercorrelations of theindicator variables used in the structural analysis.

An important advantage of using LISREL is that al-lows a series of hierarchically nested model comparisonsto be tested to establish the relative fit associated withspecific changes in parameter estimates. The differencein model chi-square values were used to test a series of

Table 2Difference Between the Unadjusted Validities and the AdjustedValidities for Voluntary Turnover and Job Performance AfterPartialing Self-Deception

VariableUnadjusted Adjusted Difference

rvl. rxr in rxv

Voluntary turnover and jobperformance with

1.

2.

3.

4.

5.

Conscientiousness

Emotional Stability

Agreeableness

Extraversion

Openness toExperience

—.23**.27**

-.19*.17*

-.13*.03

-.08-.01

.07

.01

-.19*.22**

-.18*.09

-.12.00

-.07-.01

.03

.05

-.04-.05-.01-.08-.01-.03-.01

.00-.04

+.04

Note. All values represent average values across Samples 1 and 2;TV = 286.* p < .05, one-tailed. **p < .01, one-tailed.

RESPONSE DISTORTION 267

Table 3Difference Between Unadjusted Validities and AdjustedValidities for Voluntary Turnover and Job Performance AfterPartialing Impression Management

Unadjusted Adjusted DifferenceVariable in rxy

Voluntary turnover and jobperformance with

1.

2.

3.

4.

5.

Conscientiousness

Emotional Stability

Agreeableness

Extraversion

Openness to Experience

-.23**.27**

-.19*.17*

-.13*.03

-.08-.01

.07

.01

-.17*.25**

-.13*.13*

-.09.02

-.09.03.07.09

-.06-.02-.06-.04-.04-.01+.01+.04

.00+.08

Note. All values represent average values across Samples 1 and 2;N = 286.* p < .05, one-tailed. **p < .01, one-tailed.

nested structural models. The first of these nested modelsconsisted of the structural model specified in the hypoth-esis section, hereafter referred to as Model 1. Model 1also included structural parameters for the methodeffects of both response distortion constructs on the indi-cators of the substantive predictor and criterion con-structs. The second nested model, Model 2, is the samestructural model as Model 1 except that the methodeffects for either response distortion construct on the sub-stantive constructs were eliminated by fixing these pa-rameters to zero. Comparing these two models enabledus to test whether there were response distortion effectsin these data.

The chi-square values, associated degrees of freedom,and fit statistics from Model 1 (i.e., the hypothesized dis-tortion model) are reported in the first six columns inTable 5. As shown, this model fit the data acceptably,X2(79,A r= 286) = 205.98, p<.0l; goodness of fit index(GFI) = .924, normed fit index (NFI) = .926, root meansquare residual (RMSR) = .021. These fit statistics werewithin the range normally considered to indicate an ac-ceptable fit. Furthermore, as shown in the sevenththrough ninth columns in Table 5, a comparison of thefit statistics for a structural model including methodeffects for both response distortion constructs (Model 1)fit significantly better than a structural model without themethod effects of either response distortion construct(Model 2): AX

2(20, TV = 286) = 130.50, p < .01. Thisevidence indicates that there was support at this stage forretaining the effects (factor loadings) of the two responsedistortion constructs on the indicators of the substantivepredictor and criterion constructs.

Although not shown in the table, for Model 1 (i.e., the

hypothesized distortion model), 6 of the 10 factor load-ings for each response distortion construct on the indica-tors of the substantive constructs were significant (ts >1.96). All three indicators for the two predictors(conscientiousness and emotional stability) had signifi-cant loadings from both self-deception and impressionmanagement. This means that both response distortionconstructs significantly affect scores on both personalityconstructs. In contrast, none of the factor loadings be-cause of self-deception or impression managment wassignificant for any of the criterion indicators (supervisoryratings of job performance or voluntary turnover) dem-onstrating that response distortion constructs did notaffect scores on the criterion indicators.

To examine the nature of the method effects of eachresponse distortion construct individually on the predic-tor and criterion measures, we conducted additionalmodel comparisons. In Model 3, the method effects asso-ciated with self-deception were estimated, whereas thestructural parameters for the method effects of impres-sion management were fixed at zero. In Model 4, themethod effects associated with impression managementwere estimated, whereas the structural parameters fromself-deception were fixed at zero. These models permittedexamination of the relative effects of measures for eachresponse distortion construct. Comparisons of these twomodels with Model 2 (i.e., the model with no responsedistortion) provided a test whether method effects for ei-ther of the response distortion constructs would better fitthe data than a model that did not account for themethod effects of response distortion. Furthermore, acomparison of Model 3 (i.e., the model with self-decep-tion only) and Model 4 (i.e., the model with impressionmanagement only) with Model 1 (i.e., the hypothesizeddistortion model) established the method effect of justone of the response distortion measures relative to themethod effects of both response distortion measures.

As reported in Table 5, structural models that includedthe effects of either response distortion construct (Model3 or 4) resulted in significantly better fit than a structuralmodel without the method effects; Model 2 vs. Model 3:AX

2( 10, N= 286) = 97.07, p<. 01; or Model 2 vs. Model4, AX2(10, N = 286) = 91.09, p < .01, respectively. How-ever, the model that included the effects of both responsedistortion constructs (Model 1) fit the data significantlybetter than either of the models (Model 3 or 4), whichonly included the effect of one response distortion con-struct; Model 1 vs. Model 3: AX

2( 10, N = 286) = 33.43,p < .01; or Model 1 vs. Model 4, AX

2( 10, TV = 286) =39.41 , p < .01, respectively. Thus, at this stage, we con-cluded that the method effects of both response distortionconstructs were significant and that both self-deceptionand impression management had important effects onthe manifest indicators for the personality constructs.

268 BARRICK AND MOUNT

I.

~5S

^

•§

§to

TtI)

?J

i O\ ON £

§ £0 o•5 #

Q <N -^ fN — ON •— t3

1 8i OO u~i — rn O O r^ O

<p cd

I — — CN */") r~- cc soOoo 3 eI ^ *± fN-^rn r*^r^!r^ *"* O

ra H"E y

i c^ — Q f*-] r*-, — r«-j> tni

• — ONOorr o —• r- roi• r~~- ^" r*^ ^~ c-J (N r*"i r*i <

'•3 3 .;•s a 3

^_ (L) C3 w

I <N iN ° <N - - P - 2 2q2 M g ^

oo 2 E- - ^~ ^" I— O ^O t— (N fN t—• C r/}

|(N — (N — — — Q — p O '>,—CTJ K)

S i soo so " O " J" oo f*} v^ — ro *o ^? QrscNCNi -H CNJ —< o —• O — OO « W "35

X) -) OO OO i~~l O w~i ON \o r*p OO —•* t~~- —-< fN OO O O ^

o o o o ooo ooo —• — —• _; — _." c d w * gO H ' "

1 Op ^" ^" SO f^i OO O"> OJ ON OO \O ON >_ - .

• O rsir-ir-i r^rsr^i r-i —• —* r^irN

^ ill*^ o rt

,c * t 3 - i§c 5 5 cd c 42 u UJ *5.„ 3 •*-* O 2 0 ™ *^a1.2 ~ S. c g c

i_M«a|-M«|-M«g-M m | -M ro E||

Z £ • • • c2 Hfe 5 - g»^°" ~^^ - ^ - ^vo

0. U W

Is

II

RESPONSE DISTORTION 269

Table 5Fit Statistics and Estimates of Predictor-Criterion Relationships for the Various Latent- Variable Models Evaluated

Structural model no.and type

1 , hypothesized distortion"2, no response distortion3, self-deception only

4, impression managementonly

5, self-deception and

x2

(N = 286)

205.98336.48239.41

245.39

206.12

df

799989

89

84

GFI

.924

.884

.911

.909

.924

AGF1

.870

.840

.865

.861

.876

NFI

.926

.878

.914

.911

.926

RMSR

.021

.063

.028

.028

.025

Model comparison test

Comparison

1 vs. 22 vs. 3h

1 vs. y2 vs. 4d

1 vs. 41 vs. 5e

Ax2

130.50**97.07**33.43**91.09**

39.41**0.14

fdf

20101010

105

Conscientiousness

'erformanceP

.34**

.35**

.34**

.34**

.35*

TurnoverP

-.14*-.16*-.14*

-.15*

-.16*

EmotionalStabilityturnover

-.08-.11-.08

-.09

-.11impression managementdirect effect only

6, null 2,768.63 120 .350 .263 — .998

Note. GFI = goodness of fit index; AGFI = adjusted goodness of fit; NFI = normed fit index (any goodness of fit > .9 is an indication of acceptableoverall model fit); RMSR = root mean square residual (any RMSR < . 10 is an indicator of acceptable fit). All NFIs are based on chi-square values.a With response distortion effects. bWithout response distortion effects. cWith self-deception effect only. "With impression management effectonly. "This comparison tests for response distortion effects on estimates of the predictor-criterion parameters.*p<.05. **p<.01.

A further comparison was conducted with Model 5.Model 5 was similar to Model 1, in that it accounted forthe method effects of both response distortion constructs.However, in Model 5, the structural coefficients relatingthe predictors to the criteria were constrained to equalthe estimates obtained from Model 2, which did not in-clude method effects for either response distortion con-struct. Therefore, Model 5 fixed the relationship betweenthe predictors and criteria to the estimates derived froma model that did not account for response distortion, yetincluded the method effects of both response distortionconstructs on the indicators of the two predictors and thetwo criteria. A comparison between Model 1 (i.e., the hy-pothesized distortion model) and Model 5 (i.e., themodel with self-deception and impression managementdirect effects only) provided a test of the effects the re-sponse distortion constructs had on the predictive validi-ties of interest, which is the most important model com-parison when conducting a validation study.

The nonsignificant chi-square difference betweenModel 1 (i.e., the hypothesized distortion model) andModel 5 (i.e., the model with self-deception and impres-sion management direct effects only, Ax2(5, N = 286) =0.14, ns, indicated there was no significant influence ofresponse distortion on the key parameter estimates rep-resenting the relationships among the predictor-criterionconstructs. The magnitude of the changes in the esti-mates of the predictor-criterion relationships across thevarious structural models is reported in the last three col-umns of Table 5; they vary only slightly regardless ofwhether the effects of response distortion are adjusted for.Therefore, even though the initial model comparison in-dicated that both response distortion constructs affect

scores on conscientiousness and emotional stability, sucheffects did not affect the magnitude of the validities re-ported for either personality predictor.

The last of the hierarchical nested models consisted ofthe absolute null latent model (Model 6), which proposesthat the relationships among the latent variables are con-strained to zero. The chi-square value from this modelwas used for calculating the NFI for Models 1-5.

Discussion

The results of the present study confirm previous find-ings showing that respondents can and do distort theirresponses on personality inventories and that this distor-tion does not influence the predictive validity of person-ality constructs. Our results also extend these findings byshowing that the effects are the same for two distinct typesof response distortion, self-deception and impressionmanagement. Furthermore, our findings extend previousresearch by showing that personality measures are usefulfor predicting voluntary turnover, a criterion measurethat has not been investigated frequently.

The major finding of this study with regard to responsedistortion is that although both types of response distor-tion influenced scores on personality constructs, neithertype attenuated the predictive validity of the personalitymeasures. In two separate samples, the unadjusted crite-rion-related validities for conscientiousness and emo-tional stability across two separate criteria (supervisoryratings of performance and voluntary turnover) wereslightly larger than the adjusted criterion-related validi-ties (either adjusted for unintentional self-deception orintentional impression management). Thus, two types of

270 BARRICK AND MOUNT

response distortion in two separate applicant settings didnot reduce the predictive validity of relevant personalityconstructs for either criterion.

Contrary to our expectations, the type of response dis-tortion did not differentially affect these findings. Partial-ing either self-deception or impression management fromconscientiousness and emotional stability consistently re-duced (albeit only slightly) their predictive validity.These findings are particularly interesting because appli-cants did in fact distort their responses in a more sociallydesirable manner. The results of the structural equationsanalyses were useful in demonstrating that both self-deception and impression management constructs sig-nificantly influenced responses on all indicators of con-scientiousness and emotional stability. This means ap-plicants with higher scores on the response distortionmeasures also had higher scores on both of these person-ality constructs. Such results can be interpreted as evi-dence that applicant scores on the personality measuresare inflated by trying to look good in order to get hired.It should also be noted that the correlations among theBig Five constructs are higher in these two applicant sam-ples than in nonapplicant settings with the same instru-ment (e.g., Barrick & Mount, 1993; Mount, Barrick, &Strauss, 1994). The existence of such high correlationsbetween constructs is consistent with the findings ofSchmit and Ryan (1993) that the factor structure of theBig Five may differ in applicant settings. Taken together,these findings suggest that applicants respond in a moresocially desirable manner than do nonapplicants.

The existence of inflated personality scores from re-sponse distortion effects does have some important im-plications for practitioners and researchers. First, thefinding that personality scores are likely to be higher onaverage in selection settings than applicant settings sug-gests that test administrators should not apply cut-offscores derived from a sample of present employees to jobapplicants because the applicant scores are likely to beinflated. Second, Christiansen et al. (1994) demon-strated that personality scores adjusted for response dis-tortion may result in different hiring decisions thanwould have occurred with unadjusted scores. Therefore,it may prove quite difficult to defend the practice of hiringapplicants on the basis of personality scores adjusted forresponse distortion, given that adjusting for response dis-tortion does not improve criterion-related validities.

The second major finding in this study is that two mea-sures of personality, conscientiousness and emotional sta-bility, were found to predict voluntary turnover, a crite-rion that had not been the focus of much personality-related research. Also as hypothesized, conscientiousnesswas significantly related to supervisory ratings of job per-formance. Finally, emotional stability was also a validpredictor of supervisor ratings of job performance. These

results contribute to a growing body of evidence that sug-gests relevant personality constructs are important pre-dictors of job proficiency.

As pointed out above, the point biserial correlationsfor turnover constructs were corrected to biserial corre-lations to reflect continuous constructs. Although someadvocate this procedure (e.g., Campion, 1991; Hunter &Schmidt, 1990), others (Williams, 1990) have suggestedthat such corrections are inappropriate or unneccessary.Had we not corrected for dichotimization in the presentstudy, the magnitude of the correlations between the pre-dictor variables and the turnover construct would havebeen approximately one third smaller.

The finding that conscientiousness and emotional sta-bility were significant predictors of voluntary employeeturnover is important because much of the research onindividual choices regarding voluntary employee turn-over has focused on understanding the intermediate link-ages between the employees' satisfaction with the job anddecisions to voluntarily leave, and only rarely considersthe dispositional tendencies applicants bring when join-ing the organization. Our results demonstrate that an in-dividual's personality characteristics may provide valu-able insight into an applicant's propensity to withdraw ingeneral or, as shown in this study, to voluntarily leave theorganization. An advantage of this approach is informa-tion about one's propensity to withdraw can be assessedbefore the individual joins the organization. This suggestsa quite different approach to understanding turnoverthan that traditionally taken in the voluntary employee-turnover literature and should provide relatively inexpen-sive, but quite useful information to organizations inter-ested in reducing the rate and cost of voluntary turnover.

In conclusion, the most important implication of ourresults is that even though response distortion does occurin applicant settings, it does not reduce the predictive va-lidity of relevant personality constructs. This conclusioncorresponds to that reported by Hough et al. (1990),Ones et al. (1995), and Christiansen et al. (1994). Moreimportantly, the results contribute to the literature by ex-amining two separate constructs of response distortionand showing that neither influences the validity of per-sonality constructs. Additionally, an important strengthof the present study is that it used actual job applicantsrather than newly hired employees and included two rel-atively large, independent samples (N = 147 and 139)from two companies. Furthermore, the use of structuralequations modeling better enhances our understandingof the nature of the method effects attributed to the tworesponse-distortion measures assessed. On the basis ofpresent findings, we have concluded that neither type ofresponse distortion attenuates correlations between per-sonality constructs and job proficiency.

RESPONSE DISTORTION 271

References

Abelson, M. A. (1987). Examination of avoidable and un-avoidable turnover. Journal of Applied Psychology, 72, 382-386.

Arvey, R. D., Bouchard, N. L., Segal, N. L., & Abraham, L. M.(1989). Job satisfaction: Environmental and genetic compo-nents. Journal of Applied Psychology, 74, 187-192.

Barrick, M. R., & Mount, M. K. (1991). The Big Five person-ality dimensions and job performance: A meta-analysis. Per-sonnel Psychology, 44, 1-26.

Barrick, M. R., & Mount, M. K. (1993). Autonomy as a mod-erator of the relationships between the Big Five PersonalityDimensions and Job Performance. Journal of Applied Psy-chology, 78, 111-118.

Barrick, M. R., Mount, M. K., & Strauss, J. P. (1993). Consci-entiousness and performance of sales representatives: Test ofthe mediating eifects of goal setting. Journal of Applied Psy-chology, 75,715-722.

Barrick, M. R., Mount, M. K., & Strauss, J. P. (1994). Ante-cedents of involuntary turnover due to a reduction in force.Personnel Psychology, 47, 515-536.

Bass, B. M. (1957). Faking by sales applicants of a forcedchoice personality inventory. Journal of Applied Psychology,41, 403-404.

Braun, J. R. (1962). Effects of a top management faking seton the Gordon Personal Inventory. Psychological Reports, 10,611-614.

Campion, M. A. (1991). Meaning and measurement of turn-over: Comparison of alternative measures and recommenda-tions for research. Journal of Applied Psychology, 76, 199-212.

Carsten, J. M., & Spector, P. E. (1987). Unemployment, jobsatisfaction, and employee turnover: A meta-analytic test ofthe Muchinsky model. Journal of Applied Psychology, 72,374-381.

Christiansen, N. D., Coffin, R. D., Johnston, N. G., &Rothstein, M. G. (1994). Correcting the 16PF for faking:Effects on criterion-related validity and individual hiring de-cisions. Personnel Psychology, 47, 847-860.

Cohen, J. (1984). The cost of dichotomization. Applied Psy-chological Measurement, 7, 249-253.

Cotton, J. L., & Tuttle, J. M. (1986). Employee turnover: Ameta-analysis and review with implications for research.Academy of Management Review, 11, 55-70.

DeMatteo, J. S., White, L. A., Teplitzky, M. L., & Sachs, S. A.(1991, October). Relationship between temperament con-structs and selection for special forces training. Paper pre-sented at the 33rd Annual Conference of the Military TestingAssociation, San Antonio, TX.

Dunnett, S., Koun, S., & Barber, P. J. (1981). Social desirabilityin the Eysenck Personality Inventory. British Journal of Psy-chology, 72, 19-26.

Dunnette, M. D., McCartney, J., Carlson, H. C., & Kirchner,W. K. (1962). A study of faking behavior on a forced-choiceself-description checklist. Personnel Psychology, 15, 13-24.

French, E. G. (1958). A note on the Edwards Personal Prefer-ence Schedule for use with basic airmen. Educational andPsychological Measurement, 48, 109-115.

George, J. M. (1989). Mood and absence. Journal of AppliedPsychology, 74, 317-324.

George, J. M. (1990). Personality, affect, and behavior ingroups. Journal of Applied Psychology, 75, 107-116.

Horn, P. W., Caranikas-Walker, E, Prussia, G. E., & Griffeth,R. W. (1992). A meta-analytic structural equations analysisof a model of employee turnover. Journal oj'Applied Psychol-ogy, 77, 890-909.

Horn, P. W, & Griffeth, R. W. (1991). Structural equationsmodeling test of a turnover theory: Cross-sectional and longi-tudinal analyses. Journal of Applied Psychology, 76, 350-366.

Hough, L. M., (1992). The "Big-Five" personality variables—Construct confusion: Description versus prediction. HumanPerformance, 5, 139-155.

Hough, L. M., Eaton, N. K., Dunnette, M. D., Ramp, J. D., &McCloy, R. A. (1990). Criterion-related validities of person-ality constructs and the effect of response distortion on thosevalidities. Journal of Applied Psychology Monographs, 75,581-595.

Hulin, C. L., Roznowski, M., & Hachiya, D. (1985). Alterna-tive opportunities and withdrawal decisions: Empirical andtheoretical discrepancies and an integration. PsychologicalBulletin, 97, 233-250.

Hunter, J. E., & Schmidt, F. L. (1990). Dichotomization ofcontinuous variables: The implications for meta-analysis.Journal of Applied Psychology, 75, 334-349.

Joreskog, K. G., & Sorbom, D. (1989). LISREL 7. Chicago:SPSS.

Judge, T. A. (1992). The dispositional perspective in humanresources research. In G. R. Ferris & K. M. Rowland (Eds.),Research in personnel and human resources management(Vol. 10, pp. 31-72). Greenwich, CT: JAI Press.

Kirchner, W. K., Dunnette, M. D., & Mousley, N. (1960). Useof the Edwards Personal Preference Schedule in the selectionof salesmen. Personnel Psychology, 13, 421-424.

Krug, S. E. (1978). Further evidence on 16PF distortion scales.Journal of Personality Assessment, 42, 513-518.

Mathieu, J. E., & Zajac, D. M. (1990). A review and meta-analysis of the antecents, correlates, and consequences of or-ganizational commitment. Psychological Bulletin, 108, 171-194.

McClelland, J. N., & Rhodes, F. (1969). Prediction of job suc-cess for hospital aides and orderlies from MMPI scores andpersonal history data. Journal of Applied Psychology, 53, 49-54.

McCrae, R. R., & Costa, P. T., Jr. (1983). Social desirabilityscales: More substance than style. Journal of Consulting &Clinical Psychology, 51, 882-888.

McEvoy, G. M., & Cascio, W. F. (1987). Do good or poor per-formers leave? A meta-analysis of the relationship betweenperformance and turnover. Academy of Management Jour-nal, 30, 744-762.

Michaelis, W, & Eysenck, H. J. (1971). The determination ofpersonality inventory patterns and intercorrelations bychanges in real-life motivation. Journal of Genetic Psychol-ogy, 118, 223-234.

Mobley, W. H. (1977). Intermediate linkages in the relation-

272 BARRICK AND MOUNT

ship between job satisfaction and employee turnover. Journalof Applied Psychology, 62, 237-240.

Mount, M. K., & Barrick, M. R. (1995a). The Big Five person-ality dimensions: Implications for research and practice inhuman resource management. Research in Personnel andHuman Resources Management, 13, 153-200.

Mount, M. K., & Barrick, M. R. (1995b). The Personal Char-acteristics Inventory manual. Unpublished manuscript. Uni-versity of Iowa, Iowa City.

Mount, M. K., Barrick, M. R., & Strauss, J. P. (1994). Validityof observer ratings of the Big Five personality factors. Journalof Applied Psychology, 79, 272-280.

Muchinsky, P. M., & Morrow, P. C. (1980). A multidisciplinarymodel of voluntary employee turnover. Journal of VocationalBehavior, 77,263-290.

Ones, D. S., Viswesvaran, C., & Reiss, A. D. (1995). Responsedistortion and social desirability in personality testing andpersonnel selection. Symposium conducted at the 10th An-nual Meeting of the Society of Industrial and OrganizationalPsychology, Orlando, FL.

Ones, D. S., Viswesvaran, C., & Schmidt, F. L. (1993). Com-prehensive meta-analysis of integrity test validities: Findingsand implications for personnel selection and theories of jobperformance. Journal oj'Applied Psychology Monograph, 78,679-703.

Orpen, C. (1971). The fakability of the Edwards Personal Pref-erence Schedule in personnel selection. Personnel Psychology,24, 1-4.

Paulhus, D. L. (1984). Two-component models of socially de-sirable responding. Journal of Personality & Social Psychol-ogy, 46, 598-609.

Paulhus, D. L. (1988). Measurement and control of responsebias. In J. P. Robinson, P. Shaver, & L. Wrightsman (Eds.),Measures of personality and social psychological attitudes(Vol. 1, pp. 17-60). New York: Academic Press.

Paulhus, D. L. (1989). Socially desirable responding: Somenew solutions to old problems. In D. M. Buss & N. Cantor(Eds.), Personality psychology: Recent trends and emergingdirections (pp. 201-209). New York: Springer-Verlag.

Paulhus, D. L., & Reid, D. B. (1991). Enhancement and denialin socially desirable responding. Journal of Personality andSocial Psychology, 60, 307-371.

Rothstein, H. R. (1990). Interrater reliability of job perfor-mance ratings: Growth to asymptote level with increasing op-portunity to observe. Journal of Applied Psychology, 75, 322-327.

Schmit, M. J., & Ryan, A. M. (1992). Test-taking dispositions:A missing link. Journal of Applied Psychology, 77, 629-637.

Schmit, M. J., & Ryan, A. M. (1993). The Big Five in personnelselection: Factor structure in applicant and nonapplicantpopulations. Journal of Applied Psychology, 78, 966-974.

Schmitt, N., Nason, E., Whitney, D. J., & Pulakos, E. D.(1995). The impact of method effects on structural parame-ters in validation research. Journal of Management, 21, 159-174.

Schwab, D. P., & Packard, G. L. (1973). Response distortionon the Gordon Personal Inventory and the Gordon PersonalProfile in a selection context: Some implications for predict-ing employee tenure. Journal of Applied Psychology 58, 372-374.

Steel, R. P., & Ovalle, N. K. (1984). A review and meta-analysisof research on the relationship between behavioral intentionsand employee turnover. Journal of Applied Psvchology, 69,673-686.

Steers, R. M.,&Mowday, R. T. (1981) . Employee turnover andpostdecision accommodation processes. In L. Cummings &B. Staw (Eds.), Research in organizational behavior (Vol. 3,pp. 235-281). Greenwich, CT: JAI Press.

Tett, R. P., & Meyer, J. P. (1993). Job satisfaction, organiza-tional commitment, turnover intention, and turnover: Pathanalyses based on meta-analytic findings. Personnel Psychol-ogy, 46, 259-293.

White, L. A., Nord, R. D., Mael, F. A., & Young, M. C. (1993).The assessment of background and life experiences (ABLE).In T. H. Trent & J. H. Laurence (Eds.), Adaptability screen-ing for the Armed Forces (pp. 101-161). Washington, DC:Office of the Assistant Secretary of Defense.

Williams, C. R. (1990). Deciding when, how, and if to correctturnover correlations. Journal of Applied Psychology, 75,732-737.

Williams, L. J., & Anderson, S. E. (1994). An alternative ap-proach to method effects by using latent-variable models:Applications in organizational behavior research. Journal ofApplied Psychology, 79, 323-331.

Zalinsky, J. S., & Abrahams, N. M. (1979). The effects of itemcontext in faking personnel selection inventories. PersonnelPsychology, 32, 161-166.

Zerbe, W. J., & Paulhus, D. L. (1987). Socially desirable re-sponding in organizational behavior: A reconception. Acad-emy of Management Review, 12, 250-264.

Received June 2, 1995Revision received December 19, 1995

Accepted December 21, 1995 •