chapman uggerslev carroll piasentin jones 2005

17
Applicant Attraction to Organizations and Job Choice: A Meta-Analytic Review of the Correlates of Recruiting Outcomes Derek S. Chapman, Krista L. Uggerslev, Sarah A. Carroll, Kelly A. Piasentin, and David A. Jones University of Calgary Attracting high-performing applicants is a critical component of personnel selection and overall orga- nizational success. In this study, the authors meta-analyzed 667 coefficients from 71 studies examining relationships between various predictors with job– organization attraction, job pursuit intentions, accep- tance intentions, and job choice. The moderating effects of applicant gender, race, and applicant versus nonapplicant status were also examined. Results showed that applicant attraction outcomes were predicted by job– organization characteristics, recruiter behaviors, perceptions of the recruiting process, perceived fit, and hiring expectancies, but not recruiter demographics or perceived alternatives. Path analyses showed that applicant attitudes and intentions mediated the predictor–job choice relationships. The authors discuss the implications of these findings for recruiting theory, research, and practice. Keywords: recruiting, job choice, applicant reactions, person– organization fit, meta-analysis The contribution of employee knowledge, skills, and abilities to organizational performance has been increasingly recognized over the past 2 decades (Breaugh & Starke, 2000). Furthermore, learn- ing how to attract the best applicants has become critical for many organizations. Indeed, recruiting qualified applicants may become increasingly difficult over the next 15 years as demographic and economic factors create a “war for talent” (Michaels, Handfield- Jones, & Axelrod, 2001). Hundreds of articles, books, and chapters have been written on recruiting, and several extensive and comprehensive reviews have contributed substantially to the understanding of applicant attrac- tion and job choice (Barber, 1998; Breaugh, 1992; Breaugh & Starke, 2000; Rynes, 1991) as well as applicant reactions to selection procedures (Ryan & Ployhart, 2000). The knowledge gained through research in these related areas has helped to guide human resource practitioners regarding ways to attract and influ- ence the job choices of top applicants. Although narrative reviews (e.g., Rynes, 1991) have furthered our understanding of applicant attraction and job choice processes, a quantitative review of this literature has not been conducted. A meta-analytic review would complement the existing narrative reviews by improving the estimation of the relationships between predictors and outcomes associated with applicant attraction (Schmidt & Hunter, 2001). Accordingly, the first goal of this study was to use meta-analytic techniques to summarize the relationships between traditional predictors and outcomes associated with ap- plicant attraction and job choice processes. These estimates can then be used to assist researchers in building theory and to guide practice. Our second goal was to use meta-analytic techniques to assess whether moderator variables may explain differences in results among primary studies. A third goal of this study was to clarify some of the processes involved in job choice decisions by testing whether the relationship between traditional recruitment predictors and job choice is mediated by attitudes toward the organization and acceptance intentions. Determining the Relevant Outcome and Predictor Variables Four methods were used to determine which variables to include in the meta-analyses. First, we examined several definitions of recruiting to delineate the scope for our study. Second, narrative reviews were consulted to identify key recruiting variables. Next, we used existing recruiting theories to organize and categorize the variables. Finally, we allowed for new variables to emerge based on studies we collected in which the initial variables of interest were examined. Definition of Recruiting Several definitions of recruiting were used to determine the variables that are included in the recruiting rubric. Rynes (1991) defined recruitment as “encompass[ing] all organizational prac- tices and decisions that affect either the number, or types, of individuals that are willing to apply for, or to accept, a given vacancy” (p. 429). Breaugh (1992) provided a similar definition: Derek S. Chapman, Krista L. Uggerslev, Sarah A. Carroll, Kelly A. Piasentin, and David A. Jones, Department of Psychology, University of Calgary, Calgary, Alberta, Canada. Krista L. Uggerslev is now at the I. H. Asper School of Business, University of Manitoba, Winnipeg, Manitoba, Canada. David A. Jones is now at the School of Business Administration, University of Vermont. Funding for this research was provided by the University of Calgary Starter Grants for New Faculty, the Killam Memorial Trust, and the Social Sciences and Humanities Research Council of Canada. We gratefully acknowledge comments provided by Piers Steel and Kibeom Lee on earlier versions of this article, and we are indebted to Jerard Kehoe for his many helpful comments and suggestions. We also thank Jonas Shultz for his research assistance. Correspondence concerning this article should be addressed to Derek S. Chapman, Department of Psychology, University of Calgary, Calgary, Alberta T2N 1N4, Canada. E-mail: [email protected] Journal of Applied Psychology Copyright 2005 by the American Psychological Association 2005, Vol. 90, No. 5, 928 –944 0021-9010/05/$12.00 DOI: 10.1037/0021-9010.90.5.928 928

Upload: nguyen-hoang-minh-quoc

Post on 25-Sep-2015

220 views

Category:

Documents


0 download

DESCRIPTION

Chapman Uggerslev Carroll Piasentin Jones 2005

TRANSCRIPT

  • Applicant Attraction to Organizations and Job Choice: A Meta-AnalyticReview of the Correlates of Recruiting Outcomes

    Derek S. Chapman, Krista L. Uggerslev, Sarah A. Carroll, Kelly A. Piasentin, and David A. JonesUniversity of Calgary

    Attracting high-performing applicants is a critical component of personnel selection and overall orga-nizational success. In this study, the authors meta-analyzed 667 coefficients from 71 studies examiningrelationships between various predictors with joborganization attraction, job pursuit intentions, accep-tance intentions, and job choice. The moderating effects of applicant gender, race, and applicant versusnonapplicant status were also examined. Results showed that applicant attraction outcomes werepredicted by joborganization characteristics, recruiter behaviors, perceptions of the recruiting process,perceived fit, and hiring expectancies, but not recruiter demographics or perceived alternatives. Pathanalyses showed that applicant attitudes and intentions mediated the predictorjob choice relationships.The authors discuss the implications of these findings for recruiting theory, research, and practice.

    Keywords: recruiting, job choice, applicant reactions, personorganization fit, meta-analysis

    The contribution of employee knowledge, skills, and abilities toorganizational performance has been increasingly recognized overthe past 2 decades (Breaugh & Starke, 2000). Furthermore, learn-ing how to attract the best applicants has become critical for manyorganizations. Indeed, recruiting qualified applicants may becomeincreasingly difficult over the next 15 years as demographic andeconomic factors create a war for talent (Michaels, Handfield-Jones, & Axelrod, 2001).

    Hundreds of articles, books, and chapters have been written onrecruiting, and several extensive and comprehensive reviews havecontributed substantially to the understanding of applicant attrac-tion and job choice (Barber, 1998; Breaugh, 1992; Breaugh &Starke, 2000; Rynes, 1991) as well as applicant reactions toselection procedures (Ryan & Ployhart, 2000). The knowledgegained through research in these related areas has helped to guidehuman resource practitioners regarding ways to attract and influ-ence the job choices of top applicants.

    Although narrative reviews (e.g., Rynes, 1991) have furtheredour understanding of applicant attraction and job choice processes,

    a quantitative review of this literature has not been conducted. Ameta-analytic review would complement the existing narrativereviews by improving the estimation of the relationships betweenpredictors and outcomes associated with applicant attraction(Schmidt & Hunter, 2001). Accordingly, the first goal of this studywas to use meta-analytic techniques to summarize the relationshipsbetween traditional predictors and outcomes associated with ap-plicant attraction and job choice processes. These estimates canthen be used to assist researchers in building theory and to guidepractice. Our second goal was to use meta-analytic techniques toassess whether moderator variables may explain differences inresults among primary studies. A third goal of this study was toclarify some of the processes involved in job choice decisions bytesting whether the relationship between traditional recruitmentpredictors and job choice is mediated by attitudes toward theorganization and acceptance intentions.

    Determining the Relevant Outcome and PredictorVariables

    Four methods were used to determine which variables to includein the meta-analyses. First, we examined several definitions ofrecruiting to delineate the scope for our study. Second, narrativereviews were consulted to identify key recruiting variables. Next,we used existing recruiting theories to organize and categorize thevariables. Finally, we allowed for new variables to emerge basedon studies we collected in which the initial variables of interestwere examined.

    Definition of Recruiting

    Several definitions of recruiting were used to determine thevariables that are included in the recruiting rubric. Rynes (1991)defined recruitment as encompass[ing] all organizational prac-tices and decisions that affect either the number, or types, ofindividuals that are willing to apply for, or to accept, a givenvacancy (p. 429). Breaugh (1992) provided a similar definition:

    Derek S. Chapman, Krista L. Uggerslev, Sarah A. Carroll, Kelly A.Piasentin, and David A. Jones, Department of Psychology, University ofCalgary, Calgary, Alberta, Canada.

    Krista L. Uggerslev is now at the I. H. Asper School of Business,University of Manitoba, Winnipeg, Manitoba, Canada.

    David A. Jones is now at the School of Business Administration,University of Vermont.

    Funding for this research was provided by the University of CalgaryStarter Grants for New Faculty, the Killam Memorial Trust, and the SocialSciences and Humanities Research Council of Canada. We gratefullyacknowledge comments provided by Piers Steel and Kibeom Lee on earlierversions of this article, and we are indebted to Jerard Kehoe for his manyhelpful comments and suggestions. We also thank Jonas Shultz for hisresearch assistance.

    Correspondence concerning this article should be addressed to Derek S.Chapman, Department of Psychology, University of Calgary, Calgary,Alberta T2N 1N4, Canada. E-mail: [email protected]

    Journal of Applied Psychology Copyright 2005 by the American Psychological Association2005, Vol. 90, No. 5, 928944 0021-9010/05/$12.00 DOI: 10.1037/0021-9010.90.5.928

    928

  • Employee recruitment involves those organizational activitiesthat (1) influence the number and/or types of applicants who applyfor a position and/or (2) affect whether a job offer is accepted (p.4). Barber (1998) proposed a narrower definition of recruiting thatonly includes purposeful actions taken by the organization; how-ever, this description excludes important unintended influences onapplicant attraction such as applicant reactions to selection proce-dures (e.g., Gilliland, 1993) and organizational image (e.g., Turban& Greening, 1997). We chose to examine a wider range of vari-ables, as suggested by Rynes (1991) and Breaugh (1992).

    Recruiting Outcomes

    Although several outcome variables of recruiting have beenexamined, there is considerable variability in the labels affixed tothese outcomes and some confusion about the constructs beingmeasured (Highhouse, Lievens, & Sinar, 2003). By closely exam-ining item content, we identified four recruiting outcome variables:job pursuit intentions, joborganization attraction, acceptance in-tentions, and job choice.

    Job pursuit intentions. Applicant intentions to pursue a job orto remain in the applicant pool are typically measured early in therecruitment process (Rynes, 1991). In this meta-analysis, job pur-suit intentions included all outcome variables that measured apersons desire to submit an application, attend a site visit orsecond interview, or otherwise indicate a willingness to enter orstay in the applicant pool without committing to a job choice.

    Joborganization attraction. One of the most popular outcomemeasures in the recruiting literature involves the applicants over-all evaluation of the attractiveness of the job and/or organization.The measures used to assess attraction reflect three variations thatwe collapsed into one category. First, some items ask the applicantto reflect on the job for which he or she was applying (i.e., jobattraction); for example, How attractive is the job to you? (Saks,Weisner, & Summers, 1994). A second type of item assesses theextent to which an applicant is personally attracted to the prospec-tive organization (e.g., How much would you like to work for thiscompany?; Macan & Dipboye, 1990). A third type of item fo-cuses on the attractiveness of the organization in general, withoutreference to a particular applicants level of attraction (e.g., [Thisorganization] is one of the best employers to work for; Smither,Reilly, Millsap, Pearlman, & Stoffey, 1993).

    Acceptance intentions. Measures of acceptance intentions as-sess the likelihood that an applicant would accept a job offer if onewere forthcoming, and they are frequently assessed when actualjob choice information cannot be obtained. Most studies use asingle item such as How likely are you to accept a job offer fromthis company? or a small number of items to assess acceptanceintentions.

    Job choice. Ultimately, researchers and practitioners are inter-ested in actual job choice. In this meta-analysis, job choice wasdefined as choosing whether to accept a real job offer involving anactual job. When an organization extends a job offer, the applicantmakes a job choice decision that is typically dichotomous in nature(i.e., either to accept or decline the offer). In contrast, the otheroutcome variables reviewed above are attitudinal in nature andtherefore are not dependent on the organization first tendering ajob offer.

    Predictors

    Researchers have examined a wide variety of possible predictorsof applicant attraction over the past 50 years. Several prominenttheories and models in recruiting research (e.g., signal theory;Rynes, Bretz, & Gerhart, 1991) guided our search for relevantpredictors of the outcome variables and helped to organize ourfindings. From these models and the empirical research in whichthe models were tested, we identified six broad factors typicallyexamined as predictors of applicant attraction. Each of these fac-tors is briefly described below.

    Job and organizational characteristics. Objective factor the-ory (Behling, Labovitz, & Gainer, 1968) contends that applicantsbase their job choices largely on their evaluations of the jobattributes or vacancy characteristics of the position being evalu-ated. We categorized these broader attributes into those attributesthat are specific to a job (e.g., pay, benefits, type of work) andthose attributes that are more broadly reflective of the organization(e.g., company image, size, work environment, location, familiar-ity). Thus, joborganization attributes relate to what specific at-tributes applicants seek.

    Recruiter characteristics. Critical contact theory (Behling etal., 1968) suggests that because applicants often have insufficientinformation about job attributes, they have difficulty makingmeaningful comparisons among jobs. Therefore, applicants maybe influenced more by the recruiter than by attributes of the job(Harris & Fink, 1987). Applicants perceptions of a recruitercomprise characteristics of the recruiter (e.g., age, function) andthe recruiters behavior (e.g., friendly, competent), which mayprovide signals about the attractiveness of a given position (Ryneset al., 1991). Thus, recruiter characteristics encompasses both whodoes the recruiting and how the recruiter behaves.

    Perceptions of the recruitment process. Researchers have ex-amined applicants perceptions of the recruitingselection process,typically described as applicant reactions, as focal predictors ofrecruiting outcomes (Ryan & Ployhart, 2000). Research questionsrelating to the recruitment process include whether applicantsperceive they are receiving appropriate interpersonal treatment andtimely information during the recruitment process and whether theselection instruments are perceived to be face valid and procedur-ally fair. Thus, perceptions of the recruitment process reflect howthe recruitmentselection process should be conducted.

    Perceived fit. Thus far, the predictors that we have consideredhave been limited to those that are presumed to have simple, linearrelationships with applicant attraction. A more complex view, withits origins in Schneiders (1987) attraction selection attrition par-adigm and Behling et al.s (1968) subjective factors theory, sug-gests that applicants seek a fit with the organization (personorganization [P-O] fit) or with the type of job being filled (personjob fit; e.g., Cable & Judge, 1996, 1997; Judge & Bretz, 1992;Kristof, 1996; Tom, 1971). Applicants are proposed to interpretcharacteristics of the job, organization, and recruiter in light oftheir own needs and values to determine fit. In other words,applicants perceived fit results from their appraisal of the inter-action between their personal characteristics and needs and joborganizational characteristics and supplies (Kristof, 1996).

    Perceived alternatives. Several researchers (e.g., Bauer,Maertz, Dolen, & Campion, 1998) have examined the extent towhich applicants perceive viable alternative employment opportu-

    929APPLICANT ATTRACTION META-ANALYSIS

  • nities (sometimes referred to as perceived marketability). Moreperceived available opportunities are thought to have a negativeeffect on attraction to any one specific opportunity; however,findings pertaining to this question are mixed (Barber, 1998).

    Hiring expectancies. Expectancy theory (Vroom, 1966) hasplayed an important role in applicant attraction research, althoughfew researchers have examined the model fully by includingmeasures of instrumentality, valence, and expectancy. Nonethe-less, researchers have found supportive results for the role of hiringexpectancies in applicant attraction, which are typically operation-alized as applicants evaluations of the likelihood of being offereda position in an organization. Positive hiring expectancies arepredicted to lead to greater applicant attraction (e.g., Rynes &Lawler, 1983). In this meta-analysis, the hiring expectancy cate-gory included applicants perceptions of the likelihood of receiv-ing a job offer and about their performance during the recruitmentselection process.

    Moderating Effects of Gender, Race, and Applicant Type

    Meta-analysis provides the opportunity to determine whetherbetween-study differences may explain differential results in therelationships between predictors and criteria across primary stud-ies (Schmidt & Hunter, 2001). Three possible moderators that arewidely reported across studies were investigated in this study:gender, race, and applicant type.

    Gender. The benefits of recruiting a diverse workforce toimprove performance and/or to meet legislated goals make itimportant to determine whether specific recruiting activities by anorganization work equally well across men and women. Recruitingresearchers have investigated gender differences in job attributepreferences (e.g., Jurgensen, 1978; Wiersma, 1990), job advertise-ment reactions (e.g., Winter, 1996), recruiter gender (Judge &Cable, 2000), and applicant reactions to selection procedures (e.g.,Chapman & Ployhart, 2001).

    A potential rationale for these gender differences is the influ-ence of role conflict (Wiersma, 1990). For example, job andorganizational attributes that reduce conflict with nonwork roles(e.g., flexible hours, location, family-friendly benefits, on-site day-care) may be more attractive to women than to men, although someresearchers have suggested that these differences are disappearingbecause of changes in societal norms (Barber & Daly, 1996).Nonetheless, female applicants might weigh job and organizationalcharacteristics that have the potential to reduce role conflict moreheavily than might men. Other relationships between recruitment-related predictors and outcomes might also be moderated by gen-der. Several researchers have suggested that individuals fromgroups who have experienced historical discrimination may bemore sensitive to injustice (e.g., Branscombe, Schmitt, & Harvey,1999; Chapman & Ployhart, 2001; Ryan, 2001; Ryan, Sacco,McFarland, & Kriska, 2000). Women, then, may be more sensitivethan men to certain characteristics of selection systems, such astheir perceived fairness. Accordingly, the salience of injustice towomen may result in more negative reactions to selection proce-dures than men, and hence relationships between selection systemfairness and recruitment outcomes may be moderated by gender.Some relationships in the recruitment literature, however, areunlikely to be moderated by gender. Previous research, for exam-ple, has found that men and women do not differ in their reactions

    to recruiter characteristics (such as recruiter gender; e.g., Judge &Cable, 2000).

    Race. Although there is little empirical information on racialdifferences in job attribute preferences, legislative and perfor-mance goals associated with attracting a diverse workforceprompted us to explore race as a moderator of applicant attraction.Racial differences in reactions to recruiter characteristics haveyielded little support (e.g., Judge & Cable, 2000). Nonetheless, weanticipated that salient knowledge of historical discriminationmight result in racial differences regarding the relationship be-tween applicant reactions to selection procedures and recruitmentoutcomes. Specifically, racial minorities may be more vigilantwith respect to justice violations and/or they might react morestrongly to injustices due to the salience of historical racism(Branscombe et al., 1999; Ryan, Sacco, et al., 2000).

    Applicant type. The third moderating variable we exploredwas applicant type. The issue of using laboratory-based designs inrecruiting research is a long-standing debate in the field (Barber,1998; Breaugh, 1992). Specifically, the recruiting context may beparticularly difficult to simulate with nonapplicants because actualjob choice has ramifications that are difficult to model in thelaboratory (e.g., changes in an individuals social group and finan-cial and social status). We were interested in whether studies inwhich participants were asked to role play as job applicants (i.e.,nonapplicants) yielded similar findings to those using actual jobapplicants. Examining applicant type might provide answers toimportant questions about whether and when findings based onnonapplicants generalize to real applicants and, conversely,whether inferences based on applicant samples are consistent withthe possibly more internally valid inferences derived from predom-inantly experimental research using nonapplicant samples. Testingpotential differences between applicants and nonapplicants mightprovide evidence to guide researchers in choosing appropriatemethodologies to study recruiting processes.

    A Mediated Model of Job Choice

    Several models of behavioral prediction, such as the theories ofreasoned action (Ajzen & Fishbein, 1977) and planned behavior(Ajzen, 1991), suggest that attitudes relating to a given behaviorlead to behavioral intentions and that intentions predict subsequentbehavior. Implicit in many studies in the recruitment literature isthe belief that applicant attraction to an organization predictsacceptance intentions, which in turn predicts applicant job choice.Additionally, the six categories of predictors outlined above arethought to predict applicant attraction. Thus, we examined whetheracceptance intentions and applicant attraction mediate the relation-ship between the recruiting predictors and job choice. These resultsmay offer evidence regarding the extent to which attitudinal out-come variables serve as reasonable proxies for actual job choicebehavior.

    Method

    Defining the Population of Studies

    A systematic and comprehensive search for studies was conducted infour steps. First, databases in psychology (PsycINFO, January 1967 to July2002), management (ABI Inform), and education (ERIC) were searchedusing 26 recruitment-related terms (e.g., applicant attraction, applicant

    930 CHAPMAN, UGGERSLEV, CARROLL, PIASENTIN, AND JONES

  • reactions, job acceptance, job choice, job applicants, organizational at-tractiveness, recruiter behavior, recruiter characteristics, and recruiting).Second, the reference lists from six recruiting reviews (Anderson, Born, &Cunningham-Snell, 2001; Barber, 1998; Breaugh & Starke, 2000; Ryan &Ployhart, 2000; Rynes, 1991; Rynes & Cable, 2003) were examined. Thesereviews also provided information about studies prior to 1967. Third, wereviewed recent conference programs (1996 to 2002) for the Academy ofManagement and the Society for Industrial and Organizational Psychology.Fourth, 18 prominent researchers in the recruiting field (i.e., those withmultiple and recent publications in this area) were contacted for any workin press, under review, or in progress.

    A total of 298 recruiting-related studies were identified and reviewed.Seventy-one of these studies (comprising 74 independent samples), con-tained data relevant to the applicant attraction variables included in thismeta-analysis. Studies were excluded because they only contained data that(a) were nonempirical (k 35), (b) were from a recruiter as opposed to anapplicant perspective (k 26), (c) were obtained only after a job offer hadbeen made (k 25), (d) involved relationships outside the parametersidentified for this meta-analysis (k 94), and/or (e) did not meet thenecessary statistical assumptions underlying meta-analysis (e.g., they onlyprovided partial or semipartial relationships; Hunter & Schmidt, 1990; k 47).

    Coding the Data

    Examination of the items used in the primary studies revealed that thevariable labels were not always consistent with the scale content. Forexample, the item I intend to continue participation in the applicationprocess (Macan, Avedon, Paese, & Smith, 1994) purportedly measuredacceptance intentions; however, this item was consistent with our defini-tion of job pursuit intentions. Using the original scale items, Krista L.Uggerslev and Kelly A. Piasentin independently coded each outcomevariable into job pursuit intentions, joborganization attraction, acceptanceintentions, or job choice outcome categories. The coders agreed on thecategorizations for 99.3% of the coefficients; 100% agreement was subse-quently reached through discussion.

    The same process was used to categorize the predictor coefficients.Krista L. Uggerslev and Kelly A. Piasentin independently coded each ofthe predictors into six predictor categories (job and organization charac-teristics, recruiter characteristics, perceptions of the recruitment process,perceived fit, perceived alternatives, and hiring expectancies). Subcatego-ries were also inductively identified on the basis of their frequency ofappearance in the literature (e.g., pay, justice perceptions; see the proce-dure used by Kinicki, McKee-Ryan, Schriesheim, & Carson, 2002). Initialagreement was 98.3%, and all discrepancies were resolved throughdiscussion.

    Once the data were compiled into categories, Derek S. Chapman exam-ined the coefficients within each category to detect any anomalies (e.g., alarge negative coefficient where all other coefficients were positive). In oneinstance, contacting the authors of a primary study resulted in a correctionto the original work, and the correct coefficient was used in this meta-analysis. In summary, the coding process resulted in 618 total coefficientsrelevant for 89 predictorcriterion relationships that were to be analyzedand 49 coefficients for the 5 interrelationships among the applicant attrac-tion outcome variables.

    Analyses

    We conducted the meta-analysis at multiple levels of specificity.First, we meta-analyzed the coefficients within each subcategory inrelation to each of the four outcome categories. For example, wemeta-analyzed the coefficients for P-O fit (a subcategory of the per-ceived fit category) with each outcome variable. Second, we meta-analyzed the coefficients from each broader category with the four

    outcome variables. At the category level, all coefficients were includedin the analysis, including those from the subcategories subsumed withinit as well as coefficients that were relevant for a category but that couldnot be classified into a specific subcategory. Building on the earlierexample, at the category level for perceived fit, we incorporated allP-O, personjob, and personrecruiter fit coefficients as well as thosethat were coded to reflect general perceived fit. This procedure allowedus to determine both (a) the effect of a very specific variable (e.g.,recruiter friendliness) and (b) the corrected variability accounted for bya broad category such as recruiter characteristics.

    Nonindependence issues. Many recruiting studies measured multiplepredictors and outcome variables, occasionally at more than one point intime. This introduces the potential for overweighting studies that containnonindependent data that can distort the results of the meta-analysis(Hunter & Schmidt, 1990). Therefore, we used a conservative approach tominimize the problems associated with nonindependent coefficients. Spe-cifically, when correlations were reported for multiple points in time, onlythe wave of data that most closely preceded the outcome variable was usedbecause maximum exposure to the recruitment process had occurred bythis stage. Moreover, more proximal measures of attitudesintentions arethe best predictors of behavior (Ajzen, 1991). Finally, for those casesinvolving more than one coefficient from the same study, the overall studysample size was distributed evenly among the coefficients.

    Computation of population effect sizes. Correlation coefficients werecumulated using the psychometric meta-analytic techniques outlined byHunter and Schmidt (1990). Meta-analysis provides an estimate of the truepopulation effect size for a given predictorcriterion relationship by firstdetermining how much of the variance in effect sizes across studies is dueto various statistical artifacts and then correcting the effect sizes for theeffects of these artifacts (Hunter & Schmidt, 1990; Schmidt & Hunter,2001).

    Sample-size-weighted average correlation coefficients were com-puted between the four criterion categories and each of the predictorcategories and subcategories. Analyses were performed for anypredictor criterion relationship with two or more coefficients. Theseweighted mean correlation coefficients were then corrected for unreli-ability in the predictor and criterion measures. Because several studieswere missing reliability coefficients, artifact distributions were used tocorrect for unreliability of the measures. We created separate artifactdistributions using the available reliabilities for a given analysis. Fol-lowing the procedure to correct for unreliability by Hunter and Schmidt(1990), we first calculated attenuation factors for the predictor andcriterion. The attenuation factors were computed by determining thesquare roots of the reliabilities that were available for coefficients in agiven meta-analysis. The attenuation factors for the predictor andcriterion were then multiplied to create a mean compound artifactattenuation factor, which was used to correct for the effects of unreli-ability in each meta-analysis. This artifact distribution approach al-lowed us to correct the distribution (i.e., mean and variance) of thecorrelation coefficients subsequent to the sampling error correction.Finally, we calculated 95% confidence intervals using the standarderror for the mean sample-size-weighted effect size to assess the accu-racy of the estimated mean of the population parameters (Whitener,1990).

    Moderator detection and estimation. The second stage of the meta-analysis involved determining whether the studies that were cumulated foreach analysis came from heterogeneous populations (i.e., whether moder-ator variables were present) and, if so, whether the moderator variables thatwere specified a priori accounted for a significant amount of the residualvariance. The Q statistic (Hedges & Olkin, 1985) was used as a test ofeffect size homogeneity, with a statistically significant Q indicating heter-ogeneity of effect sizes.

    When a statistically significant Q statistic suggested the presence ofmoderator variables, the significance of race and gender moderator vari-

    931APPLICANT ATTRACTION META-ANALYSIS

  • ables was tested using weighted least squares regression (Hedges & Olkin,1985). Because separate effect sizes for men versus women and Whitesversus minorities were not reported in the primary studies, it was notpossible to conduct subgroup analyses for gender and race moderators. Forthe weighted least squares analyses, the correlation coefficients wereregressed on each moderator variable, and the inverse of the sampling errorwas used as the weighting variable (Hedges & Olkin, 1985). The moderatorvariables were examined alone to determine their individual significanceand as a block to determine the total amount of variance accounted for byall moderators. Race and gender were coded as the percentage of Whitesand the percentage of men in the study samples, respectively.

    To assess the significance of the applicant type moderator, we used thepreferred subgroup method (Schmidt & Hunter, 2001). Because the par-ticipants in the primary studies were either actual job applicants or partic-ipants role playing as applicants, the effect sizes for coefficients computedusing actual job applicants were compared with those computed fornonapplicants.

    Steel and Kammeyer-Mueller (2002) determined that conducting mod-erator analyses when the ratio of variables to studies included in theanalysis was low resulted in biased estimates; therefore, we adopted aconservative approach and only tested for moderator variables when theratio of coefficients to variables was at least 10:1. Moderator tests were notconducted on job choice because of an insufficient number of coefficients.

    Path analyses. To determine whether the relationships between thevarious predictor categories and job choice were mediated by applicantattitudes toward the organization and intentions, we conducted structuralequation modeling (SEM) using the correlation matrices obtained from themeta-analyses. This approach was similar to that used by Tett and Meyer(1993). To use the correlation coefficients produced by our meta-analyseswithin SEM, we assigned each coefficient a mean of zero and a standarddeviation of one (see Tett & Meyer, 1993). We examined whether joborganization attraction (hereafter referred to as attitudes) and acceptanceintentions mediated the relationship between each of the six broad predic-tor categories and job choice.

    The SEM was conducted for each of the six predictor categories. To ruleout the possibility that alternative models fit the data as well as or betterthan a fully mediated model, we also assessed three variants of the fullymediated model for each predictor category (described below). Job pursuitintentions was omitted from these models because there were no coeffi-cients between this outcome variable and job choice. In combination, thefour models were designed to assess whether the relationship between eachpredictor category and job choice was fully mediated by joborganizationattraction attitudes and acceptance intentions or whether direct or partiallymediated models provided a better fit to the data (see Baron & Kenny,1986). Because there were two potentially mediating variables, the numberof alternative models we could test was greater than simply testing a singlemediated versus partially mediated relationship; however, the underlyinglogic for our tests was similar. That is, the alternative models weredesigned to establish whether we could rule out direct effects of thepredictor on job choice and whether partial mediation was occurring. Thisapproach is consistent with conservative SEM procedures wherein plausi-ble alternative models can be discounted (Millsap & Meredith, 1994;Williams, Bozdogan, & Aiman-Smith, 1996). Next, we describe each ofthe four models we tested (see Figure 1).

    The direct model (see Figure 1A) was a fully independent modelestimating direct paths from the predictor, attitudes, and intentions to jobchoice without any mediation. The attitude mediated model (see Figure 1B)was a partially mediated model in which the predictor relationship withintentions and job choice was fully mediated by applicant attraction atti-tudes. A direct path from attraction attitudes to job choice was estimated inaddition to a mediated path through intentions. A third model, the inten-tions mediated model (see Figure 1C), emphasized the role of intentions inpredicting job choice by including a direct path from the predictor tointentions (partially bypassing attitudes) and constraining the direct path

    from attitudes to job choice to zero. In the hypothesized fully mediatedmodel (see Figure 1D), the relationship between the predictor category andjob choice was fully mediated by attitudes and subsequently intentions,with no indirect paths estimated. The overall fit for each model wasassessed using the goodness-of-fit index, the adjusted goodness-of-fit in-dex, the root-mean-square error of approximation, Akaikes informationcriterion, and the comparative fit index. Nested model comparisons wereconducted using a combination of chi-square tests and changes in fitindices in accordance with accepted SEM procedures (e.g., Kline, 1998).Models with equal degrees of freedom were also compared using Akaikesinformation criterion with lower numbers indicating a better fit.

    Results and Discussion

    Table 1 presents the meta-analytic results for the recruitingpredictors on each of the attraction outcomes. In the next section,the results are presented and discussed for each outcome variablein relation to all of the predictors sequentially. Within each out-come variable, we focused our discussion on (a) effect sizes, (b)general patterns of findings, and (c) key findings for theory andpractice. Next, moderator analyses are discussed. Finally, we dis-cuss the results from the relationships among the outcome vari-ables and, in particular, their consistency with a fully mediatedmodel in which attitudes and intentions mediate the relationshipbetween recruiting predictors and job choice.

    It is important to note that because the population of applicantsmay differ as applicants move through Barbers (1998) stages ofthe recruitment process (generating applicants, maintaining appli-cant status, and influencing job choice), range restriction in thepredictor variables may become increasingly common. Applicants,for example, may self-select out of the recruiting process if theyperceive poor fit with the organization. Thus, the population ofapplicants who are present early on in the recruiting process(which is typically examined in studies in which job pursuitintentions is used as an outcome variable) may not be the same asthe population later in the presumed causal sequence, such as whenjob choice decisions are being made. Accordingly, comparisons ofrelative strength among the predictors are made more confidentlywithin each outcome variable, whereas a comparison of thestrength of one predictor category across the recruiting outcomesnecessitates more cautious interpretation.

    Recruiting Predictors and Attraction Outcomes

    To briefly summarize our major findings, it was clear that themost difficult recruiting outcome to predict on the basis of tradi-tional recruiting predictors was job choice in that the direct effectsof most of the predictors were either nonsignificant or had verysmall effect sizes. Additionally, our results suggest that the leastpredictive broad category of predictors was perceived alternatives.Effect sizes found for the remaining relationships between thetraditional predictors and the recruiting outcomes were often mod-erate effects and there was less differentiation among the predic-tors than might be expected. Notable exceptions included thesubstantial role that P-O fit played in job pursuit intentions relativeto most other predictors and the strong effects of job and organi-zational characteristics, particularly type of work and perceivedwork environment, on applicant intentions.

    Job pursuit intentions. Table 1 illustrates that two subcate-gories identified within job organization characteristics were

    932 CHAPMAN, UGGERSLEV, CARROLL, PIASENTIN, AND JONES

  • particularly strong predictors of job pursuit intentions: type ofwork ( .53) and organization image ( .51). Indeed, the95% confidence intervals for these two subcategories did notoverlap with the confidence intervals of any other subcatego-ries, indicating that they are the strongest predictors. It isinteresting to note that the meta-analytic effect sizes of the joband organization characteristics with job pursuit intentions wereconsistent with the direct estimation rankings of job and orga-nization attributes found by Jurgensen (1978) and by Turban,Eyring, and Campion (1993). In these studies, applicants rankedtype of work as the most strongly preferred of 10 attributes,

    with pay and advancement in the middle range. Although directestimation techniques have been criticized in the recruitingliterature (e.g., Barber, 1998), our findings suggest that it ispossible that applicants have some degree of accurate insightinto their decision-making processes. Nonetheless, this insightmay be restricted to the strongest (and therefore most salient)predictors, as the order of the remaining predictors was notreplicated.

    Osborn (1990) investigated whether job choice decisionswere made using compensatory models (which propose thathigh attraction to some variables compensates for low attraction

    Figure 1. Direct effects path model (A), attitudes mediated model (B), intentions mediated model (C), and fullymediated path model (D). JOA joborganization attraction; AI acceptance intentions.

    933APPLICANT ATTRACTION META-ANALYSIS

  • Tab

    le1

    Met

    a-A

    naly

    ses

    ofR

    ecru

    itm

    ent

    Pre

    dict

    ors

    and

    App

    lica

    ntA

    ttra

    ctio

    nO

    utco

    mes

    Pred

    icto

    r

    Job

    purs

    uit

    inte

    ntio

    nsJo

    bor

    gani

    zatio

    nat

    trac

    tion

    Acc

    epta

    nce

    inte

    ntio

    nsJo

    bch

    oice

    kn

    r xy

    95

    %C

    Ik

    nr x

    y

    95%

    CI

    kn

    r xy

    95

    %C

    Ik

    nr x

    y

    95%

    CI

    Job

    and

    orga

    niza

    tiona

    lch

    arac

    teri

    stic

    s60

    7,17

    1.3

    2.3

    8.3

    5.4

    098

    6,58

    9.3

    2.3

    9.3

    6.4

    268

    9,58

    9.4

    7.5

    7.5

    5.5

    914

    748

    .09

    .09

    .02

    .17

    Job

    char

    acte

    rist

    ics

    181,

    921

    .27

    .30a

    .26

    .35

    281,

    986

    .24

    .30a

    .25

    .35

    232,

    633

    .39

    .45a

    .42

    .49

    366

    1.1

    0

    .03

    .18

    Com

    pens

    atio

    nan

    dad

    vanc

    emen

    t14

    1,07

    7.1

    3.1

    4.0

    8.2

    016

    1,72

    8.2

    2.2

    7.2

    2.3

    37

    906

    .30

    .42

    .34

    .51

    258

    4.1

    2.0

    4.2

    0Pa

    yb7

    934

    .14

    .15

    .09

    .22

    81,

    362

    .22

    .27

    .21

    .33

    376

    1.2

    8.2

    8.2

    1.3

    52

    584

    .12

    .0

    4.2

    0T

    ype

    ofw

    ork

    370

    5.4

    6.5

    3.4

    6.6

    09

    1,08

    4.2

    9.3

    7.3

    0.4

    49

    1,48

    1.4

    4.5

    2.4

    7.5

    6

    Org

    aniz

    atio

    nal

    char

    acte

    rist

    ics

    405,

    647

    .31

    .36a

    .33

    .39

    656,

    072

    .31

    .37a

    .34

    .40

    315,

    692

    .31

    .36a

    .34

    .39

    670

    2.0

    7

    .0

    1.1

    4W

    ork

    envi

    ronm

    ent

    648

    9.1

    9.2

    3.1

    3.3

    48

    860

    .47

    .60

    .53

    .66

    61,

    104

    .45

    .53

    .48

    59

    Org

    aniz

    atio

    nim

    age

    183,

    718

    .43

    .51

    .47

    .54

    273,

    121

    .40

    .48

    .44

    .51

    174,

    547

    .34

    .41

    .38

    .44

    L

    ocat

    ion

    3

    703

    .24

    .32

    .22

    .41

    31,

    123

    .31

    .35

    .28

    .40

    362

    5.0

    6

    .0

    2.1

    3Si

    ze

    41,

    079

    .11

    .12

    .06

    .19

    231

    2.0

    3.0

    3

    .09

    .14

    Fa

    mili

    arity

    83,

    172

    .17

    .21

    .17

    .25

    62,

    197

    .26

    .31

    .27

    .36

    42,

    768

    .24

    .32

    .27

    .36

    H

    ours

    3

    554

    .14

    .20

    .08

    .31

    Rec

    ruite

    rch

    arac

    teri

    stic

    s8

    1,52

    9.3

    3.3

    7.3

    4.4

    454

    1,56

    1.2

    3.2

    9.2

    3.3

    641

    2,88

    0.2

    5.3

    2.2

    8.3

    64

    599

    .10

    .11

    .02

    .19

    Dem

    ogra

    phic

    s

    832

    7.0

    3.0

    3

    .09

    .15

    356

    7

    .04

    .0

    5

    .14

    .05

    G

    ende

    r

    332

    7.0

    4.0

    4

    .08

    .16

    256

    7

    .04

    .0

    5

    .14

    .05

    Fu

    nctio

    n

    332

    7

    .01

    .0

    1

    .12

    .10

    Rec

    ruite

    rbe

    havi

    ors

    81,

    529

    .33

    .37a

    .32

    .42

    381,

    268

    .23

    .29

    .23

    .36

    312,

    348

    .22

    .29a

    .24

    .34

    461

    4.0

    9.1

    0.0

    2.1

    9Pe

    rson

    able

    ness

    31,

    123

    .50

    .4

    6.5

    59

    896

    .34

    .42

    .35

    .49

    91,

    836

    .24

    .30

    .25

    .36

    359

    9.1

    0.1

    1.0

    3.2

    0C

    ompe

    tenc

    e

    854

    5.2

    4.2

    9.1

    9.3

    98

    1,42

    5.1

    9.2

    4.1

    8.3

    0

    Info

    rmat

    iven

    ess

    8

    717

    .24

    .31

    .22

    .40

    783

    9.0

    6.0

    9

    .01

    .18

    T

    rust

    wor

    thin

    ess

    4

    130

    .21

    .26

    .06

    .47

    319

    0.1

    8.2

    3.0

    5.4

    0

    Perc

    eptio

    nsof

    recr

    uite

    r

    524

    7.2

    1.2

    5.1

    1.3

    94

    306

    .42

    .53a

    .42

    .65

    Pe

    rcep

    tions

    ofre

    crui

    tmen

    tpr

    oces

    s35

    2,85

    9.2

    3.2

    7.2

    3.3

    162

    9,06

    9.3

    4.4

    2.4

    0.4

    440

    7,23

    1.3

    3.4

    2.3

    9.4

    53

    674

    .07

    .09

    .00

    .18

    Just

    ice

    perc

    eptio

    ns12

    1,39

    4.2

    1.2

    5.1

    9.3

    143

    8,13

    5.3

    3.4

    0a.3

    7.4

    220

    5,44

    5.3

    0.4

    0a.3

    7.4

    33

    674

    .07

    .09

    .00

    .18

    Proc

    edur

    alju

    stic

    e11

    1,39

    4.2

    0.2

    5.1

    9.3

    139

    8,13

    0.3

    2.3

    9.3

    7.4

    119

    5,44

    5.3

    0.4

    0.3

    7.4

    33

    674

    .07

    .09

    .00

    .18

    Tim

    ely

    resp

    onse

    2

    375

    .41

    .46

    .36

    .55

    Opp

    ortu

    nity

    tope

    rfor

    m

    358

    2.3

    6.4

    6.3

    7.5

    52

    329

    .38

    .53

    .40

    .66

    Jo

    bre

    late

    d

    123,

    462

    .34

    .42

    .38

    .45

    585

    1.1

    6.2

    1.1

    3.2

    9

    Tre

    atm

    ent

    5

    2,71

    4.3

    5.4

    5.4

    1.4

    9

    C

    onsi

    sten

    cy2

    512

    .18

    .21

    .11

    .30

    246

    0.2

    5.3

    2.2

    1.4

    32

    466

    .22

    .29

    .17

    .41

    Pe

    rcei

    ved

    fit

    81,

    093

    .44

    .55

    .49

    .61

    103,

    191

    .30

    .45

    .40

    .50

    752

    7.3

    2.3

    7.2

    8.4

    76

    118

    .06

    .07

    .1

    3.2

    6Pe

    rson

    org

    aniz

    atio

    nfi

    t3

    745

    .50

    .62a

    .55

    .68

    444

    8.4

    0.4

    6.3

    7.5

    6

    311

    8.1

    7.1

    8.0

    0.3

    7Pe

    rson

    job

    fit

    241

    4.3

    7.4

    5.3

    4.5

    52

    118

    .0

    6

    .06

    .2

    4.1

    3Pe

    rson

    rec

    ruite

    rfi

    t

    448

    9.2

    8.3

    4.2

    4.4

    33

    386

    .03

    .04

    .0

    8.1

    6

    Perc

    eive

    dal

    tern

    ativ

    es

    92,

    501

    .12

    .16a

    .11

    .21

    81,

    668

    .0

    5

    .06a

    .1

    3.0

    05

    816

    .06

    .07

    .0

    1.1

    5H

    irin

    gex

    pect

    anci

    es12

    1,50

    8.2

    6.3

    3.2

    7.3

    931

    5,38

    7.2

    8.3

    3.3

    0.3

    628

    5,94

    0.2

    3.3

    0.2

    7.3

    46

    720

    .16

    .17

    .09

    .24

    Perc

    eive

    dhi

    ring

    expe

    ctan

    cies

    71,

    438

    .28

    .35

    .29

    .41

    91,

    089

    .22

    .26a

    .19

    .32

    132,

    096

    .28

    .36a

    .31

    .41

    672

    0.1

    6.1

    7.0

    9.2

    4Pe

    rcep

    tions

    abou

    tpe

    rfor

    man

    ce3

    519

    .22

    .25

    .15

    .34

    214,

    480

    .30

    .36

    .32

    .39

    144,

    465

    .22

    .30a

    .26

    .34

    Not

    e.W

    hen

    r xy

    (mea

    nw

    eigh

    ted

    coef

    fici

    ent)

    and

    (c

    oeff

    icie

    ntco

    rrec

    ted

    for

    the

    unre

    liabi

    lity

    ofpr

    edic

    tor

    and

    crite

    rion

    )ar

    eid

    entic

    al,i

    tis

    beca

    use

    ofro

    undi

    ng.A

    dash

    inth

    e

    colu

    mn

    indi

    cate

    sth

    atno

    relia

    bilit

    yin

    form

    atio

    nw

    asav

    aila

    ble

    toco

    rrec

    tth

    ew

    eigh

    ted

    mea

    nco

    effi

    cien

    tfo

    rth

    isar

    tifac

    t.C

    I

    conf

    iden

    cein

    terv

    al.

    aT

    his

    rela

    tions

    hip

    has

    asi

    gnif

    ican

    tQ

    stat

    istic

    .b

    Pay

    isal

    soin

    clud

    edin

    the

    subc

    ateg

    ory

    ofco

    mpe

    nsat

    ion.

    934 CHAPMAN, UGGERSLEV, CARROLL, PIASENTIN, AND JONES

  • to other variables) or noncompensatory models (which proposethat minimal levels of attractiveness of certain variables mustbe met before the position is competitive). Osborn found thatpay level, benefits offered, and advancement opportunities werepotential deal breakers (i.e., noncompensatory factors). Never-theless, we found that pay and compensation and advancementpredicted job pursuit intentions to a much lesser extent thanmost other job and organization characteristics. Why might pay,compensation, and benefits be relatively weak predictors? Themajority of studies that have investigated pay were field studieswhere pay varied naturally; thus, there may have been a restric-tion in the range of pay. As Rynes, Schwab, and Heneman(1983) highlighted, attributes that do not vary much acrossoffers may receive less weight than attributes that have greatervariability. In short, noncompensatory strategies may be usedearly in the job choice process to reduce the number of potentialemployers (Osborn, 1990) and, therefore, applicants may nothave applied for a position in the first place when the pay wasdeemed unacceptable. Thus, applicants may have used rationaleconomic models during the job search process and only ap-plied to those jobs meeting their reservation wage (Andersonet al., 2001). Concerning the overall job organizational char-acteristics predictor category, the weak effects for pay ( .15)and for the more global compensation and advancement sub-category ( .14) may have diluted the stronger effects fromtype of work and organization image, resulting in a medium-sized overall category validity coefficient ( .38; Cohen,1988).

    Among the recruiter characteristics, recruiter personablenesswas a particularly strong predictor of job pursuit intentions( .50). Nonetheless, this large validity coefficient must beinterpreted with caution because it was only based on threestudies (N 1,123). Additional research on other recruiterbehaviors is needed to ascertain whether recruiter personable-ness is significantly more predictive of job pursuit intentionsthan other recruiter behaviors (e.g., competence,informativeness).

    Joborganization attraction. As with job pursuit intentions,Table 1 reveals that the perceived work environment has thestrongest relationship with joborganization attraction (r .60).Medium effect sizes were observed for a large number of tradi-tional recruiting predictors, such as P-O fit ( .46), organiza-tional image ( .48), and justice perceptions ( .40). Amongthe organizational characteristics predictor category, work envi-ronment ( .60) and organizational image ( .48) were thestrongest predictors of joborganization attraction.

    The recruiter demographic variables were not significantpredictors of job organization attraction. Despite contrary find-ings reported in primary studies (cf. Taylor & Bergmann, 1987;Turban & Dougherty, 1992), these null meta-analytic coeffi-cients are consistent with the weak overall effects summarizedby Barber (1998). In contrast, the recruiter behaviors all havemedium effect sizes, with rhos ranging from .26 to .42. Thisfinding has implications for the signal model of recruiting(Rynes et al., 1991) in that applicants may experience recruiterbehaviors as stronger signals than recruiter demographics whenforming impressions of the organization. Furthermore, becauserecruiter characteristics ( .29) were not found to haverelationships with job organizational attraction as strong as

    job organization characteristics ( .39), perceptions of therecruitment process ( .42), or perceived fit ( .45),applicants may rely less on signals from recruiters as moreinformation about job and organizational characteristics be-comes available. These findings are consistent with Behling etal.s (1968) critical contact theory.

    Acceptance intentions. Table 1 shows that the broad job andorganizational characteristics category was a stronger predictorof acceptance intentions ( .57) than were recruiter charac-teristics ( .32), perceptions of the recruitment process ( .42), perceived fit ( .37), perceived alternatives ( .06),and hiring expectancies ( .30). Among the strongest narrowpredictors of acceptance intentions were perceived work envi-ronment ( .53), perceptions of the recruiter ( .53),opportunity to perform ( .53), and type of work ( .52).This suggests that applicants place a lot of weight on what theyimagine their future job environment will be like when formingtheir acceptance intentions. Thus, later in the recruiting processemployers may want to focus their efforts on providing detailedinformation on actual working conditions in the organization. Itis also apparent that there is some spillover from the applicantmaintenance stage in that applicants who perceive that theorganizations selection practices are not providing them withan opportunity to demonstrate their abilities will be less likelyto form intentions of accepting an offer. This finding highlightsthe importance of ensuring the face validity of selection pro-cedures and/or providing detailed explanations of why selectionprocedures are chosen.

    Later in the recruiting process the applicant may have a betterunderstanding about other characteristics of the job and orga-nization and, therefore, have more complete information onwhich to base their acceptance intentions. In addition, eventsoccurring closer to the job choice decision may be more salientand, thus, have greater influence than earlier events (Barber,1998). More evidence was found for the importance of theperceived fit between applicant and job characteristics overrecruiter fit in the later stages of recruitment. Specifically,personjob fit appeared to be a stronger predictor of acceptanceintentions ( .45) than personrecruiter fit ( .04), giventhat the confidence intervals did not overlap. This finding isalso consistent with that of Judge and Cable (1997) who foundthat perceived fit significantly predicted acceptance intentionsincremental to the attractiveness of job attributes. Thus, futureresearch should assess unique effects and whether a greaterproportion of variability in acceptance intentions can be ac-counted for by considering multiple recruiting variables simul-taneously. Also, a somewhat surprising finding was that theperceived alternatives category was not a significant predictorof acceptance intentions. We speculate that perhaps the qualityof the competing alternatives is more important than the numberof competing offers. Given the small number of studies thathave examined this issue, continued research is necessary inorder to fully understand whether the perceived alternativescategory is a useful predictor of acceptance intentions.

    Job choice. All predictors of job choice had either smalleffects or were not significant. Indeed, the largest significantvalidity coefficient was only .17 (hiring expectancy),

    935APPLICANT ATTRACTION META-ANALYSIS

  • accounting for just over 3% of the variability in job choice.1

    There are at least four possible explanations for why the effectsizes for predicting job choice decisions are low. First, becausejob choice decisions are typically dichotomous in nature, point-biserial correlations were used to calculate the effect sizes.Point-biserial correlations are limited by the distributions ofboth the dichotomous and continuous variables and in the mosttypical cases will have ceilings well below .80,2 and, thus,comparing the effect sizes for job choice and other outcomevariables should be done with caution. Second, job choicerequires both the applicant and the employer to commit to oneanother. It is likely that some employers are highly attractive tomost applicants; however, if only a few positions are offered,the relationship between attraction and job choice is necessarilyattenuated. Studies that measure job choice regardless ofwhether an applicant receives an offer from a specific employertypically find that the applicants simply reciprocate the rejec-tion from the target employer and choose another positionregardless of their initial attraction to the target organization(e.g., Chapman & Webster, in press). This reduces the observedeffect sizes for predictors of job choice when rejected appli-cants are included in the sample.

    Third, Judge and Cable (1997) offered a possible explanation forthe nonsignificant relationship between P-O fit and job choice.These authors suggested that the attenuated effect is due to the factthat only applicants who pursued and then received jobs areincluded at this late stage of the recruitment process. In otherwords, applicants may be offered positions because of a perceivedfit with the organization (Chatman, 1989, 1991); however, theymay self-select out of the recruitment process prior to the joboffering stage if they perceive a lack of fit with the organization(Rynes, 1991). Thus, there may be range restriction in the level ofP-O fit of applicants in the pool when job choice decisions aremade (Judge & Cable, 1997). Indeed, Rynes et al. (1991) foundthat nearly half of all applicants withdrew after an initial interview,and Barber, Hollenbeck, Tower, and Phillips (1994) reported thatone third of applicants withdrew prior to the job choice stage.Thus, our meta-analytic evidence is consistent with the contentionthat applicant withdrawal during the recruitment process mayresult in restricted ranges for many of the recruiting predictors, andit may attenuate the effects of these predictors on job choice(Rynes et al., 1983).

    Finally, it is highly likely that the relationship between thesepredictors and job choice is not direct (Judge & Cable, 1997).Specifically, processes such as attraction and intentions may me-diate these relationships as might other variables such as perceivedalternatives (Cable & Judge, 1996) or knowledge of the organiza-tion (Chapman, Uggerslev, & Webster, 2003). We present theresults of our tests of these mediated models after the results of themoderator analyses.

    In all analyses, only a few studies were included in which jobchoice was measured. Recruiting scholars (e.g., Barber, 1998;Ryan & Ployhart, 2000) have urged researchers to move beyond anexamination of intentions to accept an offer or pursue a particularjob by examining actual job acceptance behaviors. More primaryresearch is needed to assess whether job choice can be reliablypredicted.

    Moderator Analyses

    The Q statistic indicated the presence of moderator variables for17 relationships between the predictor subcategories and the atti-tudinal outcomes. Where sufficient data were available, theweighted least squares regression analyses found little or no evi-dence supporting moderator effects for gender or race. Gender wasa significant moderator for only 2 of 11 opportunities. Specifically,women placed more weight on job characteristics ( .51, p .001) and less weight on fairness perceptions ( .46, p .001)than did men in determining the attractiveness of the joborganization. Race was not a significant moderator for any of theeight opportunities where sufficient data were available to test thismoderator.

    Table 2 shows the results of the subgroup analyses that wereconducted to estimate the relationships between predictors and theapplicant attraction outcomes for applicants and nonapplicantswhen the Q statistic detected the presence of moderator variables.Significant subgroup differences are indicated by confidence in-tervals that do not overlap. The subgroup analyses show that jobapplicants were likely to weigh job characteristics more stronglythan nonapplicants in evaluating joborganization attractiveness.Real applicants also weighed justice perceptions more stronglythan nonapplicants in evaluating the attractiveness of jobs andorganizations.

    The type of applicant also moderated two predictoracceptanceintentions relationships. Table 2 shows that organizational charac-teristics were a more important consideration to applicants than tononapplicants. Likewise, applicant justice perceptions were alsomore strongly related to acceptance intentions for real applicantsthan they were for nonapplicants.

    Relationships Among the Outcome Variables

    The meta-analytic results for the intercorrelations among thefour applicant attraction variables are presented in Table 3. Thecorrelation coefficients among the applicant attraction outcomeswere substantial except for the coefficients with job choice behav-ior. All of the perceptual variables were measured prior to an offerbeing extended (and, therefore, prior to a job choice decision).Higher coefficients between these perceptions and job choice mayhave been obtained if the perceptions were measured after the jobchoice decision was made (Lawler, Kuleck, Rhode, & Sorensen,1975). Specifically, applicants may adjust their perceptions tomatch their behaviors after the fact (Soelberg, 1967), or there maybe perceptual distortion after a job choice decision (Vroom, 1966).These mechanisms, as well as the others mentioned earlier, mayaccount for the relatively low correlations between job choice andthe perception measures.

    Testing a Fully Mediated Model in the Job ChoiceProcess

    We tested the four models of job choice processes illustrated inFigure 1 using the meta-analytically derived matrices in Tables 1

    1 Although P-O fit has the largest rho, the confidence interval for thispredictor includes zero and, thus, P-O fit is not a significant predictor of jobchoice.

    2 We thank an anonymous reviewer for pointing out this explanation.

    936 CHAPMAN, UGGERSLEV, CARROLL, PIASENTIN, AND JONES

  • and 3 for each predictor category. Specifically, the fit of themeta-analytic data to the direct effects, attitude mediated, inten-tions mediated, and fully mediated models was compared sepa-rately with each of the six predictor categories. Neither the fullymediated nor the direct effects model provided the best fit to thedata for any of the six predictor categories.3 Instead, for allpredictor categories, one of the two partially mediated models (i.e.,the attitude mediated or intention mediated model) provided thebest fit to the data (see Tables 4 and 5 for the path coefficients andfit indices for each predictor category with its best fitting model).4

    These two models are similar in that they both permit partiallymediated mechanisms involving attitudes and intentions but theypreclude a direct relationship from the predictor category to jobchoice. The direct effects model was the poorest fitting for all ofthe predictors, and fit indices suggest that this was an unacceptablefit in all circumstances. The fully mediated model provided anacceptable fit for each of the predictor categories except for joband organizational characteristics. Otherwise, the models testedresulted in an acceptable fit for the predictors, and the differencesamong them, although small, were often significant.

    Recruiter characteristics, hiring expectancy, and perceived fitwere all found to predict job choice through the attitudes mediatedmodel (see Table 4). Recruiter characteristics may be providinginformation about the attractiveness of the organization through asignal mechanism (Rynes et al., 1991), and the resultant attitudesinfluence on job choice is then partially mediated by intentions.Total effects of recruiter characteristics were .37, .30, and .07, forattitudes, intentions, and job choice, respectively. The finding thathiring expectancies also predicted job choice through a positiverelationship with attitudes is consistent with Janis and Manns(1977) bolstering theory of decision making. Initially, individuals

    tend to elevate choices that are more likely to occur by inflating thepositive aspects of that option and playing down the negativeaspects (see Chapman & Webster, in press; the total effects wereas follows: attitudes .33, intentions .26, and job choice .06). The finding that the relationship between perceived fit andjob choice was also best described by the attitude mediated modelis somewhat surprising in that a more direct route was expectedbetween the appraisal of fit and intentions. The total effects ofperceived fit were found to be .45, .35, and .09 for attitudes,intentions, and job choice, respectively.

    The intentions mediated model provided the best fit for thejoborganizational characteristics, perceptions of the recruitingprocess, and perceived alternatives predictor categories. Table 5reveals that joborganizational characteristics played a prominentrole in forming acceptance intentions and consequently a strongerrole in job choice than the predictor categories that were bestdescribed by the attitude mediated model (total effects were asfollows: attitudes .39, intentions .57, and job choice .19).Similarly, perceptions of the recruiting process also played a majorrole in job choice as a result of the more immediate effects onapplicant intentions (total effects were as follows: attitudes .42,

    3 The direct model was also tested with the parameter estimated betweenjoborganization attraction and acceptance intentions. Although this im-proved the overall fit of the direct model, it did not improve it sufficientlyto make it the best fitting model in any circumstances.

    4 For the sake of parsimony, path coefficients and fit indices are notpresented for each of the six predictor categories with each of the fouralternative path models. For the complete path analysis results, pleasecontact Derek S. Chapman.

    Table 2Subgroup Meta-Analyses for Applicant Type Moderator Variable

    Predictor

    Job pursuit intentions Joborganization attraction Acceptance intentions

    k n rxy 95% CI k n rxy 95% CI k n rxy 95% CI

    Job characteristics 28 1,986 .24 .30 .12 .25.35 23 2,633 .39 .45 .09 .42.49Applicants 5 286 .41 .53 .08 .41.65 11 2,261 .40 .47 .10 .43.51Nonapplicants 23 1,700 .21 .27 .09 .21.32 12 372 .32 .35 .00 .25.45

    Organizational characteristics 40 5,647 .31 .36 .18 .33.39 65 6,072 .31 .37 .18 .34.40 31 5,692 .31 .36 .12 .34.39Applicants 4 1,658 .34 .41 .13 .34.47 10 2,540 .29 .35 .19 .31.39 18 3,651 .35 .42 .07 .38.45Nonapplicants 36 3,989 .29 .34 .19 .31.38 55 3,532 .32 .39 .16 .35.42 13 2,041 .23 .27 .14 .22.31

    Justice perceptions 43 8,135 .33 .40 .06 .37.42 20 5,445 .30 .40 .07 .37.43Applicants 34 7,139 .34 .42 .05 .39.44 13 4,880 .31 .42 .07 .39.46Nonapplicants 9 996 .27 .30 .00 .23.36 7 565 .21 .28 .00 .17.38

    Recruiter behaviors 31 2,348 .22 .29 .08 .24.34Applicants 24 2,157 .23 .29 .09 .24.34Nonapplicants 7 191 .17 .21 .00 .03.38

    Perceived alternatives 8 1,668 .05 .06 .15 .13.00Applicants 5 1,435 .04 .06 .14 .13.01Nonapplicants 3 168 .04 .04 .13 .19.12

    Perceived hiring expectancies 13 2,096 .28 .36 .08 .31.41Applicants 9 1,519 .26 .33 .07 .27.39Nonapplicants 4 577 .33 .42 .06 .32.51

    Perceptions about performance 14 4,465 .22 .30 .15 .26.34Applicants 5 3,922 .21 .30 .15 .25.34Nonapplicants 9 543 .30 .38 .11 .28.48

    Note. Dashes indicate that there were no data available to estimate these relationships. rxy mean weighted coefficient; coefficient corrected for theunreliability of predictor and criterion; corrected standard deviation; CI confidence interval.

    937APPLICANT ATTRACTION META-ANALYSIS

  • intentions .42, and job choice .14). This finding underscoresthe need to pay attention to recruiting and selection practices thathave the potential to be perceived negatively by applicants. Forexample, delays in responding to applicants or using selectionprocedures that produce negative reactions may have unintendedramifications for job choice. Perceived alternatives were alsofound to predict job choice through the intentions mediated model.Overall the effect sizes for perceived alternatives were marginal:.16, .06, and .02 for attitudes, intentions, and job choice,respectively.

    General Discussion

    Our first goal for this study was to provide a comprehensive,meta-analytically derived examination of the relationships be-tween commonly used predictors of applicant attraction and otherimportant outcomes of recruitment. Three important patternsemerged from these results that warrant further elaboration. First,these meta-analyses underscore that what is being offered by the

    organization is related to applicant attraction. Characteristics ofboth the job and organization (i.e., objective factors; Behling et al.,1968) were important determinants of recruiting outcomes.

    Second, it is clear that how the recruiting is conducted (i.e.,critical contact; Behling et al., 1968) is also important; however,who does the recruiting appears not to be important. Therefore,recruiters who are selected for desired qualities (e.g., personable-ness) and trained to provide the correct information in a mannerthat is consistent and fair will likely be more successful, regardlessof their organizational function, gender, or race. This does notpreclude the possibility that there may be other individual differ-ences that could predict recruiter effectiveness; however, research-ers should examine factors other than demographics. For example,cognitive ability, personality, recruiting experience, physical at-tractiveness, and training should be explored to determine whetherindividual differences in recruiter effectiveness exist. Furthermore,recruiter behaviors and organizational characteristics that enhanceapplicants expectations of receiving an offer were also related toapplicant attraction.

    Third, and perhaps not surprisingly, perceptions of fit (i.e.,subjective factors) proved to be one of the strongest predictors ofthe attitudinal applicant attraction outcomes. Nevertheless, despitemany instances in which fit perceptions were significantly strongerpredictors than recruiter characteristics, joborganizational char-acteristics, and so forth, the extent of the improvement was oftensmall. This is important because it takes considerably more orga-nizational resources to target individual applicant needs than itdoes to provide broad-based recruiting practices that are attractiveto the vast majority of applicants. For certain key positions or forpositions that are difficult to fill, it may still be beneficial to engagein highly targeted recruitment processes to maximize fit, but thegains of such practices may be smaller for positions that havenumerous vacancies or are filled frequently. Future researchshould examine the relative use of customized recruitment prac-tices versus those that have broader appeal.

    The second goal of the present research was to identify the role,if any, that moderators might play in the relationships among thepredictors and outcomes. Our analyses suggested that several ofthese relationships might contain moderators that deserve futureresearch attention and further exploration into why these differ-ences exist. Some of the heterogeneity observed in these results,however, may have been a consequence of combining predictorand outcome variables into larger categories. Nonetheless, we

    Table 3Meta-Analysis for Coefficients Between Applicant Attraction Outcomes

    Outcome

    Job pursuit intentions Joborganization attraction Acceptance intentions

    k n rxy k n rxy k n rxy

    Job pursuit intentions Joborganization attraction 7 2,371 .56 .67 Acceptance intentions 8 2,826 .61 .74 26 7,470 .61 .78 Job choice 3 656 .18 .19 5 752 .29 .33

    Note. Dashes in the job choice row indicate that there were insufficient data to estimate these relationships. rxy mean weighted coefficient; coefficient corrected for the unreliability of predictor and criterion.

    Table 4Path Estimates and Fit Indices for Recruiter Characteristics,Hiring Expectancy, and Perceived Fit Using the AttitudeMediated Model

    Estimated pathand model statistic

    Recruitercharacteristics

    Hiringexpectancy

    Perceivedfit

    Estimated pathPredictorJOA .37 .33 .45JOAAI .78 .78 .78JOAJC .17 .17 .17AIJC .46 .46 .46

    Model statisticsRJC

    2 .12 .12 .122 2.82 11.33 1.31GFI .998 .993 .999AGFI .991 .963 .995AIC 26.47 27.33 25.14RMSEA .075 .079 .074CFI .991 .990 .992

    Note. This is the best fitting of the models tested. To obtain the results forthe poorer fitting models, please contact Derek S. Chapman. JOA joborganization attraction; AI acceptance intentions; JC job choice;RJC

    2 variance accounted for in job choice; GFI goodness-of-fit index;AGFI adjusted goodness-of-fit index; AIC Akaikes informationcriterion; RMSEA root-mean-square error of approximation; CFI comparative fit index.

    938 CHAPMAN, UGGERSLEV, CARROLL, PIASENTIN, AND JONES

  • found evidence that two of the three moderators we identified apriori may explain some of this heterogeneity.

    Our moderator analyses provided interesting results with respectto gender differences in the recruiting and job choice processes. Itis interesting to note that the differences between these men andwomen were evident for how the recruiting process was conductedas well as for what was being offered in the job. Women usedinformation about job characteristics (e.g., location, pay) morethan men in determining the attractiveness of the position. Thisfinding is consistent with role theories (e.g., Wiersma, 1990) inthat women may be more likely than men to seek out positions thatoffer a location or benefits that minimize conflicts with other liferoles (e.g., spouse, parent). We also found in one instance thatperceptions of fairness had weaker effects on joborganizationalattraction among women than among men. This finding is in theopposite direction of what we expected on the basis of the notionthat women might be particularly sensitive to perceived unfairnessdue to historical discrimination. One possible explanation for thismoderator effect is that despite persistent gender inequalities in thelabor force (e.g., Gibelman, 2003), women tend to report beingsatisfied with their situation, which produces a paradoxical con-tentment effect (Major, 1994). Thus, some women may come toaccept historical discrimination as the way things are and, hence,may weigh fairness information less heavily in the recruitmentprocess than men. We caution against overinterpreting this effector the explanation we offered, however, because the moderatoreffect for fairness was found in only one instance and our specu-lative explanation has not been tested in the recruiting literature.

    A long-standing debate is the relative merits of laboratoryversus field research for understanding workplace phenomena.Concerns about the generalizability of laboratory results and theinternal validity of field designs are common. Our results provide

    some justification for both positions. In support of laboratoryresearch, most relationships did not vary considerably because ofapplicant type. This finding is likely due, in part, to the manycreative laboratory studies that have used a variety of techniques tomake the process approximate real job choices as closely aspossible. Nevertheless, our results showed several areas in whichlaboratory researchers may need to be cautious. For instance,applicants weighed job characteristics more heavily than nonap-plicants when determining job pursuit intentions and joborganizational attractiveness. In addition, applicants paid moreattention to organizational characteristics when determining accep-tance intentions. We speculate that it is possible that considerablyless effort is required for laboratory participants to pursue a va-cancy. Applicants may consequently screen out job opportunitiesmore readily on the basis of job characteristics such as compen-sation or type of work to avoid expending limited resources forless attractive jobs. Later in the process, applicants may considerorganizational criteria more heavily because the consequences ofpicking a poorly fitting organization are more salient. Justiceperceptions were also weighed more heavily in determining joborganizational attraction and acceptance intentions for real appli-cants versus nonapplicants. Simulating injustice in a recruitmentcontext may be difficult in a laboratory setting where participantsmay feel less personally involved in the process.

    A greater number of significant Q statistics (see Table 1) sug-gest that there was more heterogeneity of effect sizes and moresubgroup differences for acceptance intentions than for job pursuitintentions. Thus, researchers may do a better job using nonappli-cants to model early recruitment processes than those more prox-imal to job choice decisions. One challenge for laboratory researchis that it is difficult to simulate the job choice context because theconsequences of that choice are not immediate. Furthermore, un-like nonapplicants, real applicants may progressively self-selectout of the applicant pool on the basis of perceived fit or otherfactors at each stage of the recruitment process creating differencesin range restriction. In fact, the reduction of range restriction inlaboratory settings may facilitate modeling job choice decisions ifrange restriction was the origin of the observed differences.5

    Nevertheless, until it is ruled out that the mind-set of laboratoryparticipants is different from that of real applicants in the finaldecision stages, researchers should be cautioned that using nonap-plicants to model later stages of the recruiting process may gen-erate inaccurate effect sizes and distorted models. In summary, wefeel strongly that laboratory-based research has a substantial roleto play in examining recruiting processes, particularly for earlierstages of employee attraction. We would also encourage recruitingresearchers to design more sophisticated longitudinal researchusing computer simulations to more closely approximate the re-cruiting and job choice process. For example, computer simula-tions that require the participant to expend considerable effort toapply for a job, or that generate more tangible consequences formaking simulated job choices, are more likely to generalize to realworld conditions. Realistic computer simulations also may providethe opportunity to test mechanisms of job choice in real timesomething that is nearly impossible to capture in a field setting.

    5 We thank an anonymous reviewer for pointing out this potential benefitof laboratory research.

    Table 5Path Estimates and Fit Indices for JobOrganizationalCharacteristics, Perceptions of the Recruiting Process, andPerceived Alternatives Using the Intentions Mediated Model

    Estimated pathand model statistic

    Joborganizationalcharacteristics

    Perceptions ofthe recruiting

    processPerceived

    alternatives

    Estimated pathPredictorJOA .39 .42 .16PredictorAI .31 .11 .19JOAAI .66 .73 .81AIJC .33 .33 .33

    Model statisticsRJC

    2 .11 .11 .112 24.77 10.98 27.27GFI .984 .993 .984AGFI .920 .964 .919AIC 40.77 26.98 43.27RMSEA .123 .077 .125CFI .980 .991 .974

    Note. This is the best fitting of the models tested. To obtain the results forthe poorer fitting models, please contact Derek S. Chapman. JOA joborganization attraction; AI acceptance intentions; JC job choice;RJC

    2 variance accounted for in job choice; GFI goodness-of-fit index;AGFI adjusted goodness-of-fit index; AIC Akaikes informationcriterion; RMSEA root-mean-square error of approximation; CFI comparative fit index.

    939APPLICANT ATTRACTION META-ANALYSIS

  • As noted previously, the Q statistic indicated the likely presenceof moderators for 17 of the relationships in Table 1. Althoughgender and applicant type accounted for a significant proportion ofthe residual variance for some of these relationships, unexplainedvariance for other relationships suggests the occurrence of addi-tional moderators.6 One potential moderator that was not directlytested in this meta-analysis was the recruiting stage in which therelationship was measured. Although we have suggested a greaterpresence of moderators for the acceptance than job pursuit inten-tions and that acceptance intentions may typically be measuredlater in the recruitment process, future research should directlyexamine the stage at which the outcome variables are measured totest this postulate explicitly. Another potential moderator of re-cruiting effects that has received recent theoretical attention andempirical support is the degree to which applicants carefullyconsider recruiting messages (see Chapman & Jones, 2002; Shultz,Jones, & Chapman, 2004). Additionally, there may be strongereffect sizes for career as opposed to position jobs. Further researchconducted using more established nonstudent applicants for actualpositions will also afford a refinement of the real applicantssubgroup (into student and nonstudent subgroups). Finally, moreresearch using longitudinal versus cross-sectional designs willallow for a test of whether research design might moderate thepredictoroutcome relationships.

    The third goal of our research was to examine the strengths andpatterns of the relationships among the recruiting outcome vari-ables. Notably, there was considerable overlap among the out-comes given that some of the corrected correlations approached.80. Thus, more work is needed in the recruiting area to developcommon definitions and operationalizations for the constructs be-ing measured. In conducting this meta-analysis, we had to relyheavily on item content rather than the labels affixed to scales.Factor analytic work (e.g., Highhouse et al., 2003) may be partic-ularly valuable to (a) determine orthogonality among the outcomevariables and to (b) construct valid items that provide sufficientdiscriminant validity among the recruiting outcome constructs.

    The results from our path analyses suggest that the relationshipsbetween the recruiting predictors and job choice are neither directnor predicted by a fully mediated model in which a given predictorrelates to attraction attitudes, which leads to acceptance intentionsand, in turn, job choice. Instead, we found that two models involv-ing a partial mediation by attitudes and intentions of the relation-ships between predictors and job choice provided the best descrip-tion. These results may explain why it is difficult to predict jobchoice directly. Recruiting variables, such as pay, type of work,and organizational image, are therefore important to the extent thatthey predict individuals attitudes toward the job and organizationand/or intentions to accept an offer. Because intentions were foundto be the most proximal mediator of recruiting predictors and jobchoice, those recruiting factors that have a more direct influence onintentions rather than attitudes are likely to be most effective. Forexample, joborganizational characteristics, such as the type ofwork and perceptions of the work environment, were found to havea more direct influence on intentions and therefore had the stron-gest role in determining job choice. Perceptions of the recruitingprocess were also found to predict job choice through the inten-tions mediated model and therefore were somewhat more predic-tive of job choice than other predictors. Although still important,recruiter behaviors such as personableness had a weaker effect on

    job choice because of the fact that they influenced attitudes towardthe organization more than intentions. It would be interesting todetermine whether recruiting techniques could be developedwhereby recruiter behaviors have a direct positive influence onintentions.

    Our results suggest that measuring acceptance intentions is thebest available proxy variable for actual job choice, in that it wasfound to mediate much of the variance between traditional predic-tors and job choice. Nevertheless, researchers should be cautiousnot to become overly reliant on intentions alone as considerableunexplained variance in job choice existed in our models, and, forsome predictors, the relationship between joborganizational at-traction and job choice was only partially mediated by intentions.

    An important practical question is, What should employers do tomaximize the effects of their recruiting efforts with the fewestresources? Our results suggest several answers to this question.Early in the recruiting process, recruiters demonstrating personablebehaviors may entice applicants to pursue the position. Thus,selecting recruiters for personableness or training them to bepersonable would be worthwhile. Emphasizing positive character-istics associated with the work environment and organizationalimage may also enhance attraction to the position. Fair and con-siderate treatment throughout the recruiting process appears to beimportant with respect to acceptance intentions. Training recruitersto enhance perceptions of fairness by providing explanations forselection procedures, keeping applicants informed, and avoidingundue delays in responses are all recommended to improve re-cruiting effectiveness. Although it is not a marked effect, a re-cruiter may entice a desired applicant into accepting a job offer byletting the applicant know that a job offer is likely forthcoming inan effort to raise expectations about being hired. At a minimum,recruiters using difficult selection procedures should attempt tomitigate the negative consequences of reduced expectations ofbeing hired by informing candidates that the selection task is verydifficult and that many successful applicants find it challenging.Recruiters should also be aware that if time and resources areavailable, additional gains may be had by focusing on the valuesand needs that seem most in line with the values and needs of theapplicant (i.e., enhancing the applicants perceived fit with theorganization). Next, we will discuss the limitations associated withthis research and suggest some issues that we believe researchersshould focus on in the future.

    All meta-analyses share the same underlying weakness in thatthe results obtained are only as meaningful as the primary studiesfrom which they were derived. For example, many of the studiesexamining attraction and job choice processes involved graduatingstudents seeking their first jobs upon graduation. Although this isa legitimate population to study, researchers should be encouragedto examine these processes in a wider variety of applicants toensure that processes generalize to other situations.

    On the basis of this meta-analysis, we are unable to determinewhether the recruiting variables can predict the applicant attractionoutcomes incrementally to one another. Wanous and Collela(1989) described a contest among the recruiting variables to

    6 Some of the residual variance may be explained by artifacts notcorrected for in this meta-analysis (e.g., dichotomization, departures fromperfect construct validity of the predictors or criterion).

    940 CHAPMAN, UGGERSLEV, CARROLL, PIASENTIN, AND JONES

  • determine whether job or organization attributes or recruitmentpractices had a greater impact on job choice. They maintained thatit is difficult to set up fair comparisons among the recruitingvariables (Cooper & Richardson, 1986). Deciding which of Beh-ling et al.s (1968) theories is most important may be supercededby deciding whether objective, subjective, or critical contactsindependently affect job choice and the extent to which they can bemodified in a recruitment context (Barber, 1998).

    Additionally, many of the predictors in recruiting may be inter-dependent. Rottenberg (1956) suggested that an applicant mightevaluate one piece of information about a position in light of otheravailable information. For example, a candidate may consider theamount of pay relative to the remoteness of the location whenmaking a decision to accept or reject a position. The applicant mayreject the position if the pay is not sufficiently high or accept it ifthe pay is deemed high enough to offset the remoteness of thelocation. Thus, the relative strength of predictors across studiesmay vary somewhat depending on what other factors are beingconsidered by the applicant or measured by the researcher in eachstudy. Although meta-analysis is designed, in part, to minimize thecontextual effects created by each study, future studies shouldstrive to measure as many variables as is practical to determine therelative effects of each variable and to provide as much detail aspossible about the sample and occupation being evaluated.

    Another limitation of these meta-analyses arises from the ne-cessity of collapsing many narrow predictors into larger categories.Although we found high interrater agreement for the categoriza-tion, it remains possible that the broader categories are less homo-geneous than desirable. For example, the variance accounted forby recruiter characteristics may be underestimated because of thenonsignificant effects from recruiter demographics; other recruiterbehaviors may indeed be very important. To help address thisissue, we also presented the narrowest predictor level warranted bythe available data in addition to the broader categories.

    As with all meta-analyses using large numbers of coefficients,some caution should be used with respect to interpreting whethermoderators exist for some of these relationships. Although the Qstatistic is more prone to Type II error than Type I bias (Hunter &Schmidt, 1990), the large number of moderator analyses conductedincreased the chances of detecting moderators