an organization-level analysis of voluntary and … · 2006-02-01 · g. douglas jenkins, jr., is...

® Academy of Management Journal1998, Vol. 41, No. 5, 511-525.

AN ORGANIZATION-LEVEL ANALYSISOF VOLUNTARY AND INVOLUNTARY TURNOVER

JASON D. SHAWDrexel University

JOHN E. DELERYG. DOUGLAS JENKINS, JR.

NINA GUPTAUniversity of Arkansas

Although there are many individual-level models of turnover, little research hasexamined the effects of human resource management practices on quit rates anddischarge rates at the organizational level. This study used organization-level datafrom 227 organizations in the trucking industry to explore this issue. Results show thathuman resource management practices predict quit rates and discharge rates hut thatthe determinants of each are quite different. Implications are derived and directionsfor future research suggested.

Turnover is the subject of much research in theorganizational sciences and economics. It is criticalfrom individual, organizational, and industry per-spectives. Yet most of the over 1,500 studies in theorganizational sciences (Muchinsky & Morrow,1980) on this topic focus on individual-level pre-dictors of turnover. In contrast, economists typi-cally examine turnover from an industry perspec-tive (Camphell, 1993), using predictors such asunemployment levels (Hulin, 1979) and lahor forcecomposition (Wachter & Kim, 1979), This researchleaves a critical gap—the determinants of turnoverat the organizational level. The importance of orga-nization-level turnover studies has been implicitlyor explicitly recognized in the organizational (e.g.,Roberts, Hulin, & Rousseau, 1978) and strategichuman resource management (e.g,, Huselid, 1995)literatures but has rarely been addressed specifi-cally. Also lacking in the literatures is differentia-tion between "voluntary" and "involuntary" turn-over. Individual researchers have acknowledgedthis distinction, but the two types of turnover arecollapsed in nearly all organizational studies (e.g.,Alexander, Bloom, & Nuchols, 1994; Bennett,Blum, Long, & Roman, 1993). We examined the

This research was supported by a grant from the Mack-Blackwell Rural Transportation Research Center in Fay-etteville, Arkansas, which was created by and is spon-sored by the United States Department of Transportation.We wish to thank the anonymous reviewers for the sub-stantial improvements in our manuscript resulting fromtheir suggestions.

G. Douglas Jenkins, Jr., is now deceased.

relationships of human resource management(HRM) practices to voluntary and involuntary turn-over, focusing on a single industry, trucking, tocontrol for external factors beyond an individualorganization's influence. Turnover is especially se-vere in the trucking industry, with rates rangingfrom 38 to 200 percent (Corsi & Fanara, 1988),making the industry particularly well suited fororganization-level examinations of turnover.

GENERAL APPROACH

An instance of voluntary turnover, or a quit, re-flects an employee's decision to leave an organiza-tion, whereas an instance of involuntary turnover,or a discharge, reflects an employer's decision toterminate the employment relationship.^ Thecauses and consequences of these distinct deci-sions are likely to be quite different. As notedabove, this distinction is well established in themanagement literature on individuals (e.g., Gupta& Jenkins, 1991; Hulin, Roznowski, & Hachiya,1985; Mobley, 1977,1982); organizational research-ers, in contrast, have elided the issue, lumpingturnover from all causes together (e.g., Alexander etal., 1994; Bennett et al., 1993; Huselid, 1995). But totreat quits, discharges, and total turnover as synon-ymous ignores the markedly different etiologiesand effects of these phenomena. In an organizationwith high quit rates, for various reasons employees

^ Involuntary turnover due to retirement, death, and soforth is essentially uncontrollable and is therefore a lessfertile topic for research.

511

512 Academy of Management Journal October

find it more attractive to leave than to stay. In anorganization with high discharge rates, however,presumably incorrect hiring decisions are reme-died through termination. These phenomena arenot just superficially, but fundamentally, different,and they must be treated differently. This is whatwe propose to do. In the following paragraphs, weoutline a framework for understanding how HRMpractices affect quits and discharges at the organi-zational level. In developing this framework, wechose not to identify and measure systems of prac-tices predicting turnover, as Arthur (1994) andHuselid (1995) did. Rather, we focused on individ-ual HRM practices and sets of HRM practices thathave the greatest impact on voluntary and involun-tary turnover. This focus enabled assessment of thedifferential and combined effects of specific HRMpractices and the development of more precisepractical guidelines.

HRM PRACTICES ASSOCIATED WITHVOLUNTARY TURNOVER

Researchers studying individuals consider vol-untary turnover to be affected by two primary fac-tors; the attractiveness of a current job and theavailability of alternatives (e.g., Hulin et al., 1985).To follow this approach, a comprehensive organi-zational framework of voluntary turnover must alsoincorporate both attractions and alternatives. Wefocus primarily on attractions here for three rea-sons: the availability of alternatives is almost con-stant in our single-industry context; trucking com-panies typically recruit drivers nationally ratherthan locally; and the trucking industry experiencesdriver shortages regularly (Perser, 1994), a fact sug-gesting that alternative job opportunities are easilyavailable to drivers regardless of their current em-ployers. It is thus more useful to focus on the HRMfactors that make drivers' current jobs more or lessappealing to them than alternatives.

Economics research indicates that investmentssuch as pay and benefits in the human capital of anorganization reduce voluntary turnover (Osterman,1987). Strategic HRM research also suggests thatcommitment-enhancing HRM systems reduce turn-over (Arthur, 1994). Common in this research is thenotion that employees maximize their own inter-ests and their own financial and psychological out-comes and that they remain with organizationswhen overall self-interest is maximized by staying.Thus, "Where the exchange is less favorable to theemployee than to the employer, the employee ismost likely to leave the firm as soon as alternativeemployment options are available" (Tsui, Pearce,Porter, & Tripoli, 1997; 1096).

Tsui and colleagues (1997) distinguished amongHRM practices on the basis of "(1) employer-expected employee contributions and (2) induce-ments offered by the employer to the employees"(Tsui et al., 1997; 1101), Inducements and invest-ments in employees increase employees' expectedoutcomes, making a job more attractive. In contrast,employers' expectations increase employees' ex-pected contributions, decreasing the attractivenessof a job. In other words, inducements and invest-ments lower quit rates; expected contributions in-crease quit rates. In this discussion, following Tsuiand colleagues we focus only on the employer as-pect of employer-employee relationships, for tworeasons; (1) because the micro literature examiningthe impact of employees' perceptions on turnoveris indeed massive already and (2) because an orga-nization's investments and expected contributionsare under its direct control, making them moreeasily influenceable in the employer's attempts toreduce turnover.

Inducements and Investments

An investment-focused HRM strategy subsumesHRM practices that ensure a high-quality humancapital pool. Primary and certainly the most visibleamong an organization's investments is its compen-sation and benefits package. High pay promotesretention since employees' self-interest is maxi-mized through staying, and it promotes organiza-tional self-interest through attraction and retentionof a superior workforce (Williams & Dreher, 1989).The relationship between pay and voluntary turn-over has been empirically supported at the individ-ual (Gupta & Jenkins, 1980; Park, Ofori-Dankwa, &Bishop, 1994), organizational (Leonard, 1987; Pow-ell, Montgomery, & Cosgrove, 1994; Wilson & Peel,1991), and industry (Parsons, 1977) levels.

The current economic climate highlights the im-portance of benefits, particularly health care bene-fits. Investment in a good benefits package shouldachieve the same ends as does high pay—that is, itshould reduce voluntary turnover. This statementis particularly true given that many benefits im-prove with organizational tenure. The relationshipbetween benefits and voluntary turnover has beenempirically supported at both the individual level(Buchko, 1992; Gupta & Jenkins, 1980; Tsai, Ber-nacki, & Lucas, 1989) and the organizational level(Bennett et al., 1993; Powell et al., 1994).

Compensation and benefits are tangible induce-ments—the rewards against which alternative em-ployment opportunities are directly assessed. Butorganizations also make indirect investments inemployees that affect employee attitudes. Drawing

1998 Shaw, Delery, Jenkins, and Gupta 513

on arguments made by Fishbein and Ajzen (1975)and otbers, psychological and micro-organizationalresearchers have posited that individual attitudeslead to behavioral intentions, which in turn lead tobehaviors. Factors affecting attitudes are the moreindirect manifestations of an organization's invest-ment in human resources.

In an era of corporate downsizing, mergers, andacquisitions, job stability is a critical manifestationof an investment strategy. The salience of tenureamong faculty has risen dramatically in recentyears, for instance. Lack of stability would implythe abrogation of a significant, albeit informal, con-tract by an organization and would diminish em-ployees' sense of attachment and responsibility tothe organization (Ashford, Lee, & Bobko, 1989). Asa critical investment, stability should enhance re-tention of employees. An investment strategy isalso reflected in the training opportunities for em-ployees an organization offers. U.S. companieshave regularly spent billions of dollars on em-ployee training, investing an average of 1.4 percentof payroll (Carnevale & Johnston, 1989) or $300 peremployee (Gordon, 1991). These investmentsshould predict voluntary turnover. Procedural jus-tice is critical for many employee outcomes (Green-berg, 1990; Thibaut & Walker, 1975). It is seldomaddressed in the economics or macro-organiza-tional literature, although Freeman and Medoff(1984) posited voice to be a critical factor in union-ization. Micro-organizational research has indi-cated the centrality of procedural, and not just dis-tributive, fairness. When HRM systems ensureprocedural justice (for example, through grievanceor appeal procedures), the relative attractiveness ofa workplace is enhanced and employees' propen-sity to leave is diminished.

Expected Contributions

HRM practices can also increase the expectedcontributions of employees. Such employer expec-tations and the HRM practices that promote themincrease what employees must give to an organiza-tion. The higher the expectations, the lower thelikelihood that employees' self-interest will bemaximized by their staying with the organization(particularly when it is compared to alternatives),and the higher the quit rate. Expectations-enhanc-ing HRM practices can also influence quit rates byaffecting employees' psychological attachment toan organization.

Employee monitoring is an expectations-enhanc-ing HRM practice. It can take different forms; closesupervision, low span of supervisory control, andelectronic monitoring are a few. Close monitoring

of employees reduces job control and increases jobdemands (Carayon, 1993; Smith, Cohen, & Stam-merjohn, 1981), and low job control and high de-mand in turn influence withdrawal cognitions andresignations (Fisher & Gitelson, 1983; Jackson &Schuler, 1985).

Characteristics of jobs influence employees' ex-pectations of necessary inputs. Micro research hasfocused on working conditions that strengthen orweaken employee-organization links (Mowday,Porter, & Steers, 1982) and the higher propensity ofemployees to quit when working conditions arenoxious (Gupta & Jenkins, 1991; Locke, 1976;Quinn & Mangione, 1973). In the trucking industry,a critical factor affecting expected contributions ishow often drivers are on the road, or time on theroad. A company that enables drivers to be homemore often is placing lower demands on driversthan one that enables their being home less often.The frequency with which drivers are home thusconstitutes another significant expectation.

Thus, we propose the following hypothesis:

Hypothesis 1. Direct HRM investment strate-gies (pay and benefits) and indirect HRM in-vestment strategies (job stability, training, andprocedural justice) will be negatively related tovoluntary turnover at the organizational level,and employer expectations (electronic moni-toring and time on the road) will be positivelyrelated to voluntary turnover at the organiza-tional level.

We have discussed a handful of HRM practices asrepresenting investment strategies and employerexpectations. These practices do not capture theentire construct domain, instead constituting asample of the domain. We expected HRM practicesto affect voluntary turnover additively: the morenumerous the inducements and investments andthe fewer the employer expectations, the lower theturnover.

HRM PRACTICES ASSOCIATED WITHINVOLUNTARY TURNOVER

Involuntary turnover (termination or discharge)has etiological dynamics, consequences, and coststhat are completely different from those of volun-tary turnover. A termination reflects a bad hiringdecision that must be corrected; a quit reflects thelower attractiveness of a current job relative to al-ternatives. Unfortunately, discharges have beendiscussed only briefly in the literatures on microHRM, strategic HRM, and economics, despite thefact that in utility research (e.g., Cascio, 1991;Schmidt, Hunter, Outerbridge, & Trattmer, 1986),


the costs of incorrect hiring decisions have beenestimated to be high. The antecedents of dischargesreside in the hiring and firing systems of an orga-nization. An organization that selects employeescarefully need not discharge as many employees;when selection errors do occur, an organizationwith effective systems for identifying and correct-ing errors should have higher discharge rates thanone that does not.

Staffing Systems

The attention an organization devotes to its em-ployee selection (staffing) systems affects the qual-ity of its hires (Schmidt & Hunter, 1983). Twoelements of staffing systems are particularly impor-tant; how selective the organization can be, or theselection ratio (Taylor & Russell, 1939), and thevalidity of the organization's selection process(Hunter & Hunter, 1984).

The selection ratio addresses how choosy an or-ganization is in hiring employees. Choosy organi-zations presumably select the best of the crop. Incontrast, the more an organization is compelled tohire all who apply, the more bad hires it is likely tohave and the higher the subsequent discharge rateis likely to be. The use of valid selection proceduresshould lead to fewer bad hires (Hunter & Schmidt,1983; 473). Huselid (1995) showed that selectivestaffing practices, used as part of a system of HRMpractices, are related to total turnover. Thus, orga-nizations using valid selection procedures shouldhave lower discharge rates than others.

The interaction between the selection ratio andthe use of valid selection procedures is likely tohave an even more potent impact. When the selec-tion ratio is high, the use of valid selection tech-niques is probably not fruitful—it doesn't matterhow good they are if most, if not all, applicants arehired. But when an organization can be choosy, andit chooses carefully, utility is enhanced. The pres-ence of botb a low selection ratio and valid selec-tion procedures would be effective, but the pres-ence of only one or the other brings costs that arenot necessarily counterbalanced by benefits. Forthis reason, the developers of most utility models(e.g., Cascio, 1989; Naylor & Shine, 1965; Schmidt& Hunter, 1983; Taylor & Russell, 1939) have incor-porated both the validity coefficient and the selec-tion ratio in their calculations, and their interactionshould have greater predictive power than either inisolation. The third element in utility calculationsis the base rate (Taylor & Russell, 1939), a measureof the proportion of applicants who are likely tosucceed on the job. Unfortunately, we could notassess its impact because of data constraints.

An organization may be able to correct selectionproblems by training employees in requisite skills(e.g., Goldstein, 1991; Wexley & Latham, 1991).Training can also define roles more clearly to em-ployees (Naylor, Pritchard, & Ilgen, 1980). Organi-zations with substantial training opportunitiesshould thus have lower discharge rates.

Monitoring Systems

To the extent that a bad hiring decision is madein the first place, how does an organization identifyand correct it? Presumably, the better tbe organiza-tion is able to monitor its employees, the morelikely it is to have information about the quality ofits hiring decisions.

Performance appraisals are ways for organiza-tions to keep track of the value provided by eachemployee (e.g.. Murphy & Cleveland, 1991). Anorganization should terminate an employee whenits investments in that employee (pay, benefits,training, etc.) exceed the contributions made byhim or her. Especially when the cost of turnover ishigh and human capital is crucial to organizationalsuccess, the organization is likely to focus attentionon terminating employees who lack the necessarycapital or the motivation to use that capital to pro-mote organizational objectives. Consequently, per-formance appraisals should be positively related toinvoluntary turnover. Employee monitoring pro-vides organizations with better information aboutemployees. Poor employee bebaviors should bemore obvious to management, increasing the like-lihood of termination. Thus, we propose the fol-lowing:

Hypothesis 2. Organizational staffing systems(a low selection ratio, valid selection proce-dures, and training) will be negatively relatedto voluntary turnover at the organizationallevel, and organizational monitoring systems(performance appraisal and electronic moni-toring) will be positively related to involuntaryturnover at the organizational level.

Hypothesis 3. The interaction of selection ratioand valid selection techniques is a strongerpredictor of involuntary turnover than the"main effects" of each.

In the foregoing discussion, we did not address theissue of arbitrary employee termination. William-son (1981) argued that when turnover costs arehigh, organizations implement practices that mini-mize arbitrary discharge. Such practices may re-duce arbitrary discharges, but they do not necessar-ily raise or lower the involuntary turnover rate per


se. Although certainly a fruitful area of inquiry,investigating these practices was beyond the scopeof this research.

To summarize, our primary interest was three-fold: we sought (1) to assess the extent to whichspecific HRM practices predict turnover at the or-ganizational level, (2) to examine the relationshipbetween these practices and quit and dischargerates separately, and (3) to conduct this examina-tion in a multivariate rather than a bivariate fash-ion.

METHODS

Sample

The original population for this study consistedof 3,104 trucking organizations that reported infor-mation to the Interstate Commerce Commission(ICC) and were included in the 1993-94 TTS BlueBook of Trucking Companies (henceforth, the BlueBook). Since very small companies are unlikely tohave systematic HRM practices, only companieswith at least 30 employees were included. Somecompanies listed in the Blue Book had gone out ofbusiness in the interval between the book's publi-cation and our data collection, and some had nocompany drivers but only "owner-operators." Ex-clusion of these companies resulted in a sample of1,072 companies. We mailed a 24-page question-naire to the highest-ranked HRM manager in eachcompany and received completed responses from379 of the companies. This level of response repre-sented a 36 percent (379/1,072) response rate. Ourquestions focused on the characteristics and effectsof the HRM systems for the firms' driver work-forces. Turnover rates were provided in 227 com-pleted questionnaires; the analyses are based onthese data.

Main Measures

Dependent variables. Key respondent reportsare typically used in organizational turnover re-search (Alexander et al., 1994; Bennett et al., 1993;Huselid, 1995). We did the same. The respondentsreported measures of driver turnover expressed aspercentages of total firm employment. These re-ports focused on drivers only, excluding other per-manent or temporary employees. Two rates werereported: driver quit rates, or voluntary turnover,and driver discharge rates, or involuntary turnover.Reports were for the previous calendar year (1994).All analyses were conducted separately for the twovariables.

HRM practices. Measures of all HRM practiceswere obtained through the questionnaire. Averagepay was measured as the dollar amount specified inresponse to the question "On average, how muchdoes a typical driver earn per year?" The attractive-ness of the benefits plan was assessed through threeLikert-type agree/disagree items with seven re-sponse options. Items included in this and othermulti-item scales are shown in Table 1. Job stabilitywas measured with two Likert-type agree/disagreeitems with seven response options. Training wasmeasured with one item: "How many hours of for-mal training does a typical driver receive in ayear?" Responses were transformed into logarithmsfor analysis. Procedural justice included threeLikert-type agree/disagree items with seven re-sponse options. Electronic monitoring consisted ofthree items asking respondents the percentage oftrucks in their fleets with "on-board computers,""satellite tracking," and "on-board systems to com-municate with dispatchers." The mean of these per-centages served as the measure. Performance ap-praisal was measured as the number of times peryear the company conducted formal performanceappraisals for drivers. Time on the road was mea-sured as the reverse of the number of times driverswere typically routed home each month. The selec-tion ratio was the number of drivers hired duringthe past year divided by the number of individualswho had applied for driving jobs in the past year.

We used a validity generalization approach inderiving our measure of selection procedures. Aprocedure was included in this measure if it had avalidity coefficient of at least .25 in a meta-analyticor validity generalization study. This resulted inthe inclusion of eight valid selection procedures:structured interviews, mental ability tests, physicalability tests, technical knowledge tests, perfor-mance or job sample tests, personality tests, hon-esty or integrity tests, and biographical informationquestionnaires. For each procedure, respondentcompanies were given a rating of 1 if they placed atleast some weight on the procedure in the selectionprocess and rating of 0 if the procedure was notused. The number of valid selection techniquesused by the company was then counted. This indexrepresented the general validity of an organiza-tion's selection procedures rather than validitywith respect to driving jobs specifically and is anapproximation of how careful a trucking companywas in driver selection.

Control Variables

To enhance generalizability of the results, weincluded several control variables based on the re-


TABLE 1Factor Analysis Results for Multi-item Subjective Scales"

Factor Loading

Scale and Item 1

.85

.83

.79

.06

.01

.24

.04

.00

.08

.04

2.7020.95

.78

2

.12-.02

.20

.86

.81

.68

.05

.06

.11

.04

1.7319.31

.67

3

- .01.02.05

.01

.03

.08

.92

.91

- .01.12

1.4917.08

.83

4

-.04.03.07

.02

.05

.11

.01

.11

.87

.87

1.3415.33

.68

BenefitsOur drivers have a better benefits package than other drivers in the industry.Our drivers could get a better benefits package in another trucking company.Our benefits package helps us attract good drivers.

Procedural justiceWe rule on disputes only after we investigate all sides of the issue thoroughly.Drivers have a chance to answer any complaints made against them.Our company has formal procedures to ensure that drivers are treated fairly.

Job stabilityWe guarantee drivers a certain amount of pay every pay period.We guarantee drivers a certain amount of work every pay period.

Applicant poolWe have many people to choose from when hiring drivers.We have so many applicants that we don't have to recruit actively.

EigenvaluePercentage of variance explainedAlpha

' Factor loadings for the correct scale are shown in bold type.

suits of other organization-level studies of turn-over, macro research in general, or the results ofmotor carrier studies in the transportation litera-ture. Control variahles included both general orga-nizational factors and a factor specific to the motorcarrier industry.

Organizational controls. The general organiza-tional variahles were organizational size and age,union status, and the applicant pool. Organiza-tional size was the logarithm of the total numher ofemployees in a company, as reported hy the re-spondent. Organizational age was the logarithm of1994 minus the founding year. Information onfounding years was obtained from the Blue Book.Union status was the percentage of drivers coveredunder a collective bargaining agreement. This in-formation was reported hy respondents. Finally, weused a perceptual measure of the available appli-cant pool, with two Likert-type agree/disagreeitems that had seven response options. We usedthis perceptual measure because trucking compa-nies often recruit and hire nationally, not region-ally, and hecause they hire only commerciallylicensed drivers; thus, we considered local un-employment rates inappropriate measures of theavailable labor supply.

Industry controls. Organization-level variablesspecific to the trucking industry also affect turn-over and other organizational outcomes (Corsi &Stowers, 1987; Lemay, Taylor, & Turner, 1993).Trucking firms can be categorized into two carrier

types, truckload organizations and less-than-truck-load organizations. In addition, some companieshandle special commodities, such as hazardousmaterials. Truckload organizations serve customershaving sufficient volume to load an entire trailer,whereas less-than-truckload (LTL) organizationsserve customers with smaller shipments. Becauseof differences in the demands placed on these twogroups, their turnover rates are quite different (Gil-roy, 1992; Lemay et al., 1993; Murphy, 1992; Per-ser, 1994). The variahle carrier type was defined asthe primary husiness of a company reported by therespondent. Because LTL carriers show turnoverpatterns that are remarkably different from othercarriers', they were placed in one category (coded0), and specialized commodity and truckload orga-nizations were placed in the other (coded 1).

Analyses

Exploratory factor analysis. To assess the extentto which the a priori multi-item scales held to-gether, we conducted an exploratory factor analy-sis. This analysis included the items in the multi-item scales measuring benefits, procedural justice,job stability, and the applicant pool; the electronicmonitoring items were excluded from this analysissince they pertained to factual information that wasnot predicted to he internally consistent. The re-sults of this principal components analysis with"varimax" rotation, shown in Table 1, suggest that


the scales held together quite well. Table 1 alsoshows the coefficient alpha reliability estimates forthe scales.

Analytic strategy. Zero-order correlations be-tween each predictor and the turnover variableswere computed. Of greater interest were multipleregression analyses that allowed us to look at thepredictors simultaneously. Hierarchical regres-sions (Cohen & Cohen, 1983) enabled us to enterthe variables as blocks and to assess the incremen-tal explanatory power of each block. The first blockcontained the control variables; the second block,the HRM practices; and the third, the interaction ofthe selection ratio and selection procedures. Weused all the HRM variables to predict both quitrates and discharge rates and to thus assess thediscriminant validity of the theoretical framework,since HRM practices predicted to relate to quitsshould be unrelated to discharges, and vice versa.

RESULTS

Response Bias and Measurement Checks

Response bias check. Following the work of Os-terman (1994), we ran a logistic regression analysisto test for differences between respondent and non-respondent companies. The dependent variablewas dummy-coded 1 if a usable questionnaire hadbeen returned and 0 otherwise. The independentvariables for this bias check were obtained from the1994 Blue Book and included number of drivers,total fringe benefits cost, total highway milesdriven, total wages paid, average haul (in miles),total insurance costs, current assets, company age,tons per mile, and average load (in tons). None ofthese variables was significant in the equation, in-dicating that response bias should not have affectedour results. A comparison of respondents who pro-vided turnover data with those who did not re-vealed only one significant difference: the lattergroup came from older companies.

Measurement checks. Most of our measureswere based on respondent reports, and the validityof such reports is open to question. Information onthree control variables—size, carrier type, and av-erage haul—was available both in the questionnaireand in the Blue Book for 1993-95. The correlationsbetween the questionnaire and Blue Book measuresranged from .84 to .91 for size, from .72 to .87 forcarrier type, and from .41 to .66 for average haul.The Blue Book also contained potential proxies fortwo HRM practices, pay and benefits. It reportedtotal wage costs for drivers and helpers, whereaswe measured average driver pay. We constructed atotal pay measure from the questionnaire (average

pay multiplied by the number of drivers) to parallelthe Blue Book measure; we also constructed anaverage pay measure from the Blue Book (totalwage cost divided by the number of drivers andhelpers) to parallel the questionnaire measure. Forthe year 1994 (the focus of the questionnaire), thecorrelation between the two average pay measureswas .33, and the correlation between the two totalpay measures was .88. The average pay measurefrom the Blue Book had a correlation of -.16 withquit rates, and the average pay measure from thequestionnaire had a correlation of -.19 with quitrates. The alternative measures were thus signifi-cantly and moderately related and showed a simi-lar pattern of relationships with turnover. We con-sidered our average pay measure to be superior tothe Blue Book measure for several reasons: (1) ourmeasure focused only on drivers, (2) the Blue Bookmean pay measure did not necessarily reflect whatan average driver made, and (3) the latter measurecontained some obvious errors (for instance, onefirm's total annual wage cost was listed as less than$5,000).

The Blue Book also contains information on thetotal costs of fringe benefits for entire organiza-tions, a measure that included benefits costs forexecutives, office staff, and other employee groups.This measure was substantively different from ourvariable assessing the attractiveness of fringe ben-efits for drivers. We nonetheless correlated ourmeasure with the Blue Book measure, which wascalculated as total benefits divided by total employ-ees, after eliminating obvious errors in the book'sdata (e.g., negative benefits costs). This correlationwas .26 for 1994. Again, the questionnaire measureprovided a better estimate of the inducement qual-ity of benefits than did the Blue Book measure.

These measurement checks indicated that the re-spondent reports used in this research had reason-able convergence with information from publishedsources and that the more similar the wording andfocus of the two measures, the higher the correla-tions. To preserve sample size (1994 Blue Bookinformation was available for only a subset of re-spondents), and because the questionnaire pro-vided better information on the variables of inter-est, we used only the questionnaire measures ofHRM practices in our further analyses.

Predictors of Voluntary and InvoluntaryTurnover

Table 2 contains the means and standard devia-tions for, and the correlations among, all the vari-ables in the study. We generated the correlationmatrix using a "pairwise" deletion procedure. The


positive correlation (r = .38, p < .01) between thequit rate and the discharge rate suggests that thetwo, although correlated, are distinct phenomena,validating our decision to examine them sepa-rately.

Tahle 3 contains the results of the hierarchicalregression analyses for both dependent variahles.The columns headed "AR^" show the incrementalexplanatory power of each step, and the beta col-umns show the unique explanatory power of eachvariable in the final equation. The table shows thatthe significant predictors of the discharge and quitrates were quite different.

The control variables accounted for about 27 per-cent of the variance in quit rates. Of greater interestwere the variables measuring inducements and in-vestments and employer expectations. The resultsgenerally confirmed our expectations. Average payand time on the road had strong relationships withquit rates (/3 = - . 3 1 , p < .01, and ^ = .25, p < .01,respectively), and benefits and electronic monitor-ing had moderate relationships with quit rates ()3 =-.16, p < .05, and )3 = .16, p < .05, respectively).The hypothesized predictors of discharge ratesonly (selection ratio, selection procedures, and per-formance appraisal) were unrelated to quit rates.The step 3 interaction was neither hypothesizednor empirically found to be related to quit rates. In

all, these results provide partial support forHypothesis 1.

The control variables accounted for 26 percent ofthe variance in discharge rates. In the HRM prac-tices block, training and the selection ratio weresignificant predictors of discharges (j3 = .15, p <.05, and /3 = .26, p < .01, respectively). Contrary toHypothesis 2, training was positively related to dis-charge rates. Hypothesis 2 was supported for theselection ratio only. The results also confirmed ourprediction of an interaction hetween the selectionratio and selection procedures (/3 = .32, p < .01),supporting Hypothesis 3. Figure 1 shows a plot ofthis interaction. The figure was plotted using val-ues one standard deviation above and one standarddeviation below the mean for the selection ratioand selection procedures. The figure shows thatwhen the selection ratio was low and the use ofvalid selection procedures was high, dischargerates were low. When valid selection procedureswere not used, however, the selection ratio wasirrelevant to discharge rates. It is noteworthy thatdischarge rates were highest when valid selectionprocedures were used, even when the selectionratio was high.

To assess the extent to which the models for thequit and the discharge rates were different, we con-structed 95 percent confidence intervals around the

TABLE 2Descriptive Statistics and Correlations for All Variables'*

Variable

Turnover1. Quit rates2. Discharge rates

Controls3. Size4. Age5. Union status6. Applicant

pool7. Carrier type

Mean

25.555 5.98

4.843.52

20.922.68

0.84Human resource practices

8. Average pay9. Benefits

10. Job stability11. Training12. Procedural

justice13. Electronic

monitoring14. Performance

appraisal15. Time on

the road16. Selection

ratio17. Selection

procedures

34,9124.862.952.385.81

15.66

0.68

16.70

0.31

3.43

s.d. 1

32.398.60 .38**

1.22 .26**0.93 -.24**

38.95 -.29**1.50 -.26**

0.36 .23**

6,289 - .21**1.28 -.23**1.70 - .21**1.12 -.010.94 -.23**

25.77 .25**

1.34 -.03

11.35 .39**

0.20 .21**

2.42 -.02

2

.35**-.20*-.19*-.10

.04

.12-.12

.03

.19*-.09

.12

.01

.11

.36**

.11

3

-.04-.07

.01

- .27* '

.29**

.13*

.01

.15**

.09

.28**

.20**

.24**

.10

.08

4

.15**

.04

- .13*

.00

.03

.08-.03

.12*

-.07

-.01

-.09

-.01

-.01

5

.11

-.28**

.19**

.14*

.11-.00

.11

-.02

-.18**

-.26**

-.09

.06

6

-.23**

.18**

.15**

.15**

.04

.18**

-.05

.00

-.15**

-.21**

.02

7

-.24**-.08-.06

.01- .13*

.03

.02

.25**

.04

-.10

8

.23**

.07

.13*

.18**

.17**

.08

.04

-.02

.13*

9

.09

.12*

.28**

.13*

.11*

-.05

-.22**

.05

10

.09

.07

-.06

.01

-.13**

.01

.12*

11

.25**

.07

.13*

-.00

-.05

.23**

12

.02

.11

-.14*

-.11

.10

13

.15**

.19**

-.14*

.00

14 15 16

.05

.07 -.07

.19** .04 -.03

"Range of Nis 148-311.* p < .05

** p < .01


TABLE 3Results of Hierarchical Regression Analysis for Quit Rates and Discharge Rates

step

12

3

Independent Variable

ControlsHuman resource practices

Average payBenefitsJob stabilityTrainingProcedural justiceElectronic monitoringPerformance appraisalTime on the roadSelection ratioSelection procedures

InteractionSelection ratio by selection procedures

HypothesesTested

111

1, 21

1, 221223

Quit Rates

Adjusted R' AR'

.24** .27**

.41** .20**

.40** .00

Dependent

P

- .31**-.16*-.09-.04-.02

. 1 6 * •

- . 1 1.25**.02.01

-.03

Variable"

Discharge Rates

Adjusted R'

.23**

.28**

.38**

.26**

.10*

.09**

P

.06

.05

.00

.15*-.07

.02- .11

.01

.26**

.08

.32**

° Adjusted R^ values are reported for each step. Beta values are reported for the final step,iV = 141 for quit rates and Af = 148 for discharge rates.

* p < .05** p < .01

FIGURE 1Interaction of Selection Procedures and the Selection Ratio in Predicting Discharge Rates

DischargeRates

20-,

18-

16-

14-

12-

10-

8-

6-

4-

2-

0-

—•— High Use of SelectionProcedures

- - • - - Low Use of SelectionProcedures

11.66

5.81

4.78

1.68

Favorable (Low)' Unfavorable (High)

Selection Ratio

regression coefficient for each predictor and exam-ined the overlap between the confidence intervals.There was no overlap for average pay and the in-teraction between the selection ratio and selection

procedures, and there was an extremely small over-lap for benefits, electronic monitoring, training,time on the road, and the selection ratio (theseconstituted the significant predictors in the two


models). In all but one case (electronic monitoring),the significant coefficients for quit rates were out-side the confidence intervals for discharge ratesand vice versa. It was reasonable to conclude thatthe two models were indeed different and that thepredictors of quit rates were different from those ofdischarge rates,

DISCUSSION

The results of this study are encouraging withrespect to the relationship between organization-ally controllable human resource management fac-tors and turnover. They show that (1) it is impor-tant to examine organization-level quit rates anddischarge rates separately, (2) the specific HRMpractices that predict quit rates are quite differentfrom those that predict discharge rates, and (3) theinteraction of selection procedures and the selec-tion ratio is a potent predictor of discharge rates.

Differentiating between Voluntary andInvoluntary Turnover

The data confirm the usefulness of differentiatingtypes of turnover in organization-level research.We analyzed (but did not report) the predictors ofvoluntary, involuntary, and total turnover. Thethree sets of analyses, taken together, indicate notonly that voluntary and involuntary turnover havedifferent etiological dynamics, but that examina-tion of total turnover may be misleading. With rareexceptions (e.g., Powell et al,, 1994), researchersconducting strategic HRM and economics studieshave tended to treat these constructs as synony-mous. Since total turnover incorporates elements ofboth quits and discharges, however, any observedpredictive effects may be contaminated. Quit dy-namics may cloud discharge dynamics.

The levels and predictors of voluntary and invol-untary turnover were quite different. The quit ratesobserved were much higher (x = 25.55%) than thedischarge rates (x =5,98%). This finding is consis-tent with industry reports that driver quit rates, notdriver discharge rates, are a significant problem(Lemay et al,, 1993). Control variables explainedabout 27 percent of the variance in both the quitand discharge rates, but HRM practices explainedmore variance in quit rates (20%) than in dischargerates (10%), and the selection procedures/selectionratio interaction was significant only for dischargerates (9 percent of the variance).

In terms of the distinctions between voluntaryand involuntary turnover, then, our results indicatethat (1) the two phenomena are related but are notsynonymous, (2) the two phenomena have different

levels, (3) investments and inducements and em-ployer expectations predict quit rates but not dis-charge rates, and (4) staffing practices predict dis-charge rates but not quit rates. We included allpredictors in equations predicting both quits anddischarges to assess the validity of our argumentsthat different HRM practices would be predictive ofdifferent types of turnover. Essentially, we were"stacking the deck" against chance findings. Theresults demonstrate convincingly the merit of ex-amining voluntary and involuntary turnover sepa-rately at the organizational level, just as these phe-nomena have been treated separately at theindividual level (e.g., Mobley, 1982; Price, 1977),Our measures of quit and discharge rates werebased on respondent reports and were probablyaffected by difficulties and interorganizational dif-ferences in classification schemes, yet the differ-ences in results were marked. If organizationalturnover data more precise than the data we em-ployed could be obtained, the observed differencesmight be magnified.

Predictors of Voluntary Turnover

The proposed framework for examining volun-tary turnover was substantively validated. Two ofthe five measures of inducements and investments(pay and benefits) and both measures of employerexpectations (electronic monitoring and time onthe road) were related to quit rates in the predicteddirection. When Tables 2 and 3 are compared, fur-ther insights are obtained. Job stability and proce-dural justice had significant zero-order relation-ships with the quit rate (Table 2), but thesecorrelations faded when other predictors were in-troduced (Table 3), The distributive justice mea-sures, the expectation of monitoring, and a high,level of time on the road apparently absorbed muchunique variance, leaving little explanatory powerfor procedural justice and job stability. The exis-tence of bivariate relationships, combined with theabsence of multivariate relationships, highlightsthe value of multivariate examinations of turnover.

In our framework, we posited that inducementsand investments and employer expectations wererelated to voluntary turnover (quits). Overall, thiswas the case. Pay was clearly the strongest corre-late. That pay and quits are related has been sub-stantiated in micro HRM, macro HRM, and eco-nomics research (e,g., Gupta & Jenkins, 1980;Parsons, 1977; Powell et al., 1994). What our resultsadd to this literature is a framework within whichpay dynamics can be seen to emerge. We arguedthat employees will remain with an organization aslong as it serves their self-interest to do so better

1998 Shaw, Delery, Jenkins, and Giipta 521

than the alternatives available to them. Pay clearlyoffers one such inducement, but it is not the onlyone. Other HRM practices, particularly benefits,also induce employees, bonding them to their em-ploying organization. Much time away from homeand electronic monitoring, on the other hand,weaken this bond. The emergence of strong predic-tors within both the inducements and investmentsand employer expectations groups validates ourproposed framework.

The variables in our model explained only 47 per-cent of the variance in voluntary turnover (R^ = .47,adjusted R^ = .41). Forty-seven percent is a substan-tial figure, but there is room for improvement. Thesampling of variables we picked is just that, a sam-pling. All inducements in oxn study concerned HRMpractices designed to enhance extrinsic inducements.But many studies have also stressed the relevance ofinfrinsic rewards to employee attitudes and behaviors(e.g., Hackman & Lawler, 1971; Hackman & Oldham,1980). Our data precluded an empirical examinationof intrinsic rewards as inducements, but the issueholds potential.

Two variables in the employer expectationsgroup (electronic monitoring and time on the road)received validation as predictors of quit rates, but ahost of other HRM practices can be subsumedwithin this group. For example, Arthur's (1994)commitment system includes supervision, partici-pation, decentralization, and social relationships.Other HRM systems (Huselid, 1995; MacDuffie,1995) are similar. A more comprehensive enuncia-tion and validation of these practices within theemployer expectations group is the next logicalstep.

Predictors of Involuntary Turnover

Involuntary turnover was hypothesized to relateto staffing and monitoring practices, and the empir-ical results partly supported this hypothesis. Staff-ing practices predicted discharge rates. We alsoposited that monitoring would give employers bet-ter information on which to base discharges, butthis effect was not observed. Perhaps monitoringprovides valid employee information, raising notdischarge rates per se, but rather, the proportion ofright discharges. That is, monitoring may increasethe validity, not the absolute level, of dischargedecisions.

By far the most interesting finding with respect todischarge rates concerns the interaction of selec-tion procedures and the selection ratio. This effect,although long recognized (Taylor & Russell, 1939),has seldom been empirically demonstrated. Theresults suggest that the use of valid selection tech-

niques, combined with a favorable (low) selectionratio, reduces discharge rates. But when the selec-tion ratio was high, using valid selection proce-dures was associated with higher discharge rates.This is puzzling. Maybe companies with valid se-lection techniques have well-developed HRM sys-tems in other areas, providing them with bettermechanisms for identifying and discharging poordrivers. Alternatively, companies with higher dis-charge rates may be more motivated to improvetheir selection practices. Base rate could also affectthis dynamic, an explanation we could not address.This is a fruitful subject for future research.

Another intriguing finding was the positive rela-tionship between training and the discharge rate, afinding counter to our prediction of a negative re-lationship. Many explanations are possible here.Perhaps companies that provide more training aremore clearly focused on employee skills and per-formance and thus have a greater number of dis-charges. Perhaps companies that experience a highdischarge rate initiate training programs because ofthe poor quality of their labor pools.

In general, discharge rates are not studied much,and our proposed framework for studying theserates was more nebulous than that for quit rates.The results substantiate the relevance and utility ofparts of the proposed framework. Construction of amore comprehensive framework, perhaps with bet-ter enunciation of the effect of monitoring on dis-charges, is a critical next step.

An HRM Systems versus an HRM PracticesApproach

A study of the predictors of turnover can go dif-ferent ways. By far the most common approach instrategic HRM research has been to identify sys-tems, or bundles, of HRM practices (Arthur, 1994;Delery & Doty, 1996; Huselid, 1995; MacDuffie,1995); this method is similar to the configurationalapproach found in strategic management research(Meyer, Tsui, & Hinings, 1993). We chose to focuson specific HRM practices instead. The commonassumption in the bundles approach is that HRMpractices act synergistically, making a bundlegreater than the sum of its parts. The influence of aspecific practice could be markedly different de-pending on the presence or absence of other prac-tices. In essence, complex interactions may existamong interdependent HRM practices. But Huselid(1995) used principal components analysis to dis-cover underlying groups of HRM practices andcomputed components scores simply by summingpractices that loaded on a component. This analyticapproach implies that HRM practices are equiva-


lent and substitutable—an organization is consid-ered better if it has either more of one HRM practiceor more HRM practices overall. Some researchershave considered it valuable to sum scores on spe-cific HRM practices to develop a system measure(e.g., Becker & Huselid, 1998), but this proceduredoes not dovetail with the dominant conceptualapproach in strategic HRM. Additive empiricaltests are not appropriate for assessing theorizedmultiplicative effects. Becker and Gerhart (1996)recommended the use of cluster analysis as an al-ternative. Arthur (1994) used cluster analysis toclassify organizations into one of two types of HRMsystems on the basis of their use of various HRMpractices. Cluster analysis is popular in the config-urational literature also. The assumptions here arethat only a finite number of coherent groups areviable in the real world and that cluster analysis isuniquely suited to detect these groups: "Just a frac-tion of the theoretically conceivable configurationsare viable and apt to be observed empirically"(Meyer et al., 1993:1176). But cluster analysis over-looks the magnitude of within-cluster variations,particularly in cross-sectional examinations of or-ganizations that may be in transition. There are alsoother concerns about the use of cluster analysis inthis context (Delery, in press). Moreover, althoughboth Arthur (1994) and Huselid (1995) uncoveredsignificant relationships between HRM practicesbundles and different measures of effectiveness,their studies shed little light on the specific HRMpractices that are most effective. Perhaps only oneHRM practice explains all the effects found in theirresearch, or perhaps several effects are interdepen-dent or synergistic. The actual effects of HRM prac-tices may also be over- or underestimated usingthese approaches.

For these reasons, we did not use a typical bun-dles approach; rather, we focused on specific HRMpractices and specific additive or interactive ef-fects. We hypothesized and tested additive effectswith respect to voluntary turnover and also hypoth-esized and tested interactive effects with respect toinvoluntary turnover. The interaction essentiallyreflects the bundling of specific HRM practices(valid selection procedures and the selection ratio).Our analytic approach was thus consistent with ourtheoretical approach. This analytic strategy notonly enabled us to verify the existence of predicted,additive and multiplicative effects, but also en-abled us to determine specific HRM practices thatheld explanatory power. Superior theoretical andpractical clarity were the result. Parenthetically,given our markedly different results for quit anddischarge rates, the study also suggests that re-searchers making predictions based on bundles

should address specific outcomes differently aswell, rather than considering all outcomes (finan-cial performance, turnover, etc.) as predictively in-terchangeable.

Limitations

Our results should be viewed in the light of thedata's limitations. Turnover measures were ob-tained from key respondent reports. Of course, allturnover data are reported by individuals. Archivalturnover sources are compiled from company re-ports of turnover levels, and the turnover calcula-tions used by some researchers are based oncompany reports of employment levels. The re-spondents in our study were generally high-rank-ing members of the organizations' HRM personnelstaffs or members of top management and shouldhave been quite familiar with the turnover rates oftheir organizations; thus, response error shouldhave been low. Differences in calculations of quitand discharge rates or in company policies (forexample, is a driver who does not show up for workfor a week considered as having quit or as havingbeen discharged?) may also introduce error into themeasures. Also, a substantial number of the re-turned questionnaires did not contain turnoverrates. Perhaps those who were unsure of their turn-over rates simply chose not to report them, suggest-ing that when the information was provided, it wasvalid. Nonresponse may also have been random.Certainly, it is important to design ways to ensurethat this information is obtained from all respon-dents.

The study was conducted in one industry, anindustry in which employees are often away fromhome for weeks at a time. This fact may raise ques-tions about the applicability of our results to otherindustries. Controlling for strong industry-specificfactors should have reduced this problem, so thatthe results regarding HRM practices can be gener-alized across industries.

The impact of all our predictors over time cannotbe addressed because the study was cross-sectional,and we cannot confirm the causal nature of thehypothesized relationships. A challenge for futureresearch is to determine the influence of HRM prac-tices on organizational outcomes over time.

Practical Implications

What do our results say to HRM managers?Within the trucking industry, of course, they em-phasize the importance of good pay, good benefits,low use of electronic monitoring, and low time onthe road if the objective is to reduce quit rates. They


also suggest the importance of bundling valid se-lection techniques with a favorable (low) selectionratio for controlling discharge rates. Beyond thetrucking industry, our results underscore the needfor careful examination of the investments made in,and the expectations held of, an organization's hu-man capital pool. These HRM systems are withinan organization's control, and a manager whowants to reduce turnover should carefully attend tomaximizing employees' financial and psychologi-cal interests. In contrast, when an organization hasan unacceptably high discharge rate, corrective ac-tion should be directed toward its staffing systems.Employees should be selected carefully when, andonly when, the selection ratio is low. The problemof bad hires and the need for termination is thenreduced.

We did not address the practical utility of humanresource management practices. Practices such ashigh pay and good staffing systems may reduceturnover, but they may not necessarily be cost-effective. An HRM practice that reduces turnovermay simultaneously reduce financial performance,an issue we did not address in this study. Futureresearch must examine these differential effects ofHRM practices on organizational outcomes.

Conclusion

This study highlights the value of differentiatingHRM practices that may influence quits from thosethat influence discharges. Although limited to oneindustry, the study nevertheless provides general-izable insights about the role of HRM practices inaffecting voluntary and involuntary turnover.

REFERENCES

Alexander, J. A., Bloom, J. R., & Nuchols, B. A. 1994.Nursing turnover and hospital efficiency: An organi-zation-level analysis. Industrial Relations, 33: 505-520.

Arthur, J. B. 1994. Effects of human resource systems onmanufacturing performance and turnover. Academyof Management Journal, 37: 670-687.

Ashford, S. J., Lee, C, & Bobko, P. 1989. Content, causes,and consequences of job insecurity: A theory-basedmeasure and substantive test. Academy of Manage-ment Journal, 32: 803-829.

Becker, B., & Gerhart, B. 1996. The impact of humanresource management on organizational perfor-mance: Progress and prospects. Academy of Man-agement Journal, 39: 779-801.

Becker, B., & Huselid, M. A. 1998. High performancework systems and firm performance: A synthesis ofresearch and managerial implications. In G. R. Ferris

(Ed.), Research in personnel and human resourcemanagement, vol. 16: 53-101. Greenwich, CT: JAIPress.

Bennett, N., Blum, T. G., Long, R. G., & Roman, P. M.1993. A firm-level analysis of employee attrition.Group and Organization Management, 18: 482-499.

Buchko, A. A. 1992. Employee ownership, attitudes, andturnover: An empirical assessment. Human Rela-tions, 45: 711-733.

Gampbell, G. M., III. 1993. Do firms pay efficiency wages?Evidence with data at the firm level. Journal ofLabor Economics, 11: 44-470.

Garayon, P. 1993. Effect of electronic performance mon-itoring on job design and worker stress: Review ofthe literature and conceptual model. Human Fac-tors, 35: 385-395.

Garnevale, A. P., & Johnston, J. W. 1989. Training Amer-ica: Strategies for the nation. Alexandria, VA:American Society for Training and Development.

Gascio, W. F. 1989. Using utility analysis to assess train-ing outcomes. In I. L. Goldstein (Ed.), Training anddevelopment in organizations: 63-88. San Fran-cisco: Jossey-Bass.

Gascio, W. F. 1991. Costing human resources: The fi-nancial impact of behavior in organizations. Bos-ton: Kent.

Gohen, J., & Gohen, P. 1983. Applied multiple regres-sion/correlation analysis for the behavioral sci-ences. Hillsdale, NJ: Erlbaum.

Gorsi, T. M., & Fanara, P. 1988. Driver management pol-icies and motor carrier safety. Logistics and Trans-portation Review, 24: 153-164.

Gorsi, T. M., & Stowers, J. R. 1987. Effects of a deregu-lated environment on motor carriers: A systematic,multi-segment analysis. Transportation Journal,3(2): 4-28.

Delery, J. E. In press. Issues of fit in strategic humanresource management: Implications for research. Hu-man Resource Management Review.

Delery, J. E., & Doty, D. H. 1996. Modes of theorizing instrategic human resource management: Tests of uni-versalistic, contingency, and configurational per-spectives. Academy of Management Journal, 39:802-835.

Fishbein, M., & Ajzen, I. 1975. Relief, attitudes, inten-tion, and behavior. Reading, MA: Addison-Wesley.

Fisher, G. D., & Gitelson, R. 1983. A meta-analysis of thecorrelates of role conflict and ambiguity. Journal ofApplied Psychology, 68: 320-333.

Freeman, R. B., & Medoff, J. L. 1984. What do unions do?New York: Basic Books.

Goldstein, I. L. 1991. Training in work organizations. InM. D. Dunnette & L. M. Hough (Eds.), Handbook of


industrial and organizational psychology, vol. 2:507-619. Palo Alto: Consulting Psychologists Press.

Gordon, J. 1991. Training budgets; Recession takes a bite.Training, 28(10): 37-45.

Greenberg, }. 1990. Organizational justice: Yesterday, to-day, and tomorrow. Journal of Management, 16:399-432.

Gupta, N., & Jenkins, G. D., Jr. 1980. The structui-e ofwithdrawal: Relationships among estrangement,tardiness, absenteeism, and turnover. Springfield,VA: National Technical Information Service.

Gupta, N., & Jenkins, G. D., Jr. 1991. Rethinking dysfunc-tional employee behaviors. Human Resource Man-agement Review, 1: 39-59.

Hackman, J. R., & Lawler, E. E., III. 1971. Employeereactions to job characteristics. Journal of AppliedPsychology, 55: 259-286.

Hackman, ]. R., & Oldham, G. R. 1980. Work redesign.Reading, MA: Addison-Wesley.

Hulin, G. L. 1979. Integration of economics and atti-tude/behavior models to predict and explain turn-over. Paper presented at the annual meeting of theAcademy of Management, Atlanta.

Hulin, G. L., Roznowski, M., & Hachiya, D. 1985. Alter-native opportunities and withdrawal decisions: Em-pirical and theoretical discrepancies and an integra-tion. Psychological Bulletin, 97: 233-250.

Hunter, J. E., & Hunter, R. F. 1984. Validity and utility ofalternate predictors of job performance. Psychologi-cal Bulletin, 96: 72-98.

Hunter, J. E., & Schmidt, F. L. 1983. Quantifying theeffects of psychological interventions on employeejob performance and work-force productivity. Amer-ican Psychologist, 38: 473-478.

Huselid, M. A. 1995. The impact of human resourcemanagement practices on turnover, produ.ctivity,and corporate financial performance. Academy ofManagement Journal, 38: 635-672.

Jackson, S. E., & Schuler, R. 1985. A meta-analysis andconceptual critique of research on role ambiguityand role conflict in work settings. OrganizationalBehavior and Human Decision Processes, 36: 16-78.

Lemay, S. A., Taylor, G. S., & Turner, G. B. 1993. Driverturnover and management policy: A survey of truck-load irregular carriers. Transportation Journal,33(1): 15-21.

Leonard, J. S. 1987. Carrots and sticks: Pay, supervisionand turnover. Journal of Labor Economics, 5: s l36-sl52.

Locke, E. A. 1976. The nature and causes of job satisfac-tion. In M. D. Dunnette (Ed.), Handbook of indus-trial and organizational psychology: 1297-1349.Ghicago: Rand McNally.

MacDuffie, J. P. 1995. Human resource bundles and man-

ufacturing performance: Organizational logic andflexible production systems in the world auto indus-try. Industrial and Labor Relations Review, 48:197-221.

Meyer, A. D., Tsui, A. S., & Hinings, C. R. 1993. Guestco-editors' introduction: Gonfigurational approachesto organizational analysis. Academy of Manage-ment Journal, 36: 1175-1194.

Mobley, W. H. 1977. Intermediate linkages in the rela-tionships between job satisfaction and employeeturnover. Journal of Applied Psychology, 62: 237-240.

Mobley, W. H. 1982. Employee turnover: Causes, con-sequences, and control. Reading, MA: Addison-Wesley.

Mowday, R. T., Porter, L. W., & Steers, R. M. 1982. Em-ployee-organization linkages: The psychology ofcommitment, absenteeism, and turnover. NewYork: Academic Press.

Muchinsky, P. M., & Morrow, P. C. 1980. A multidimen-sional model of voluntary employee turnover. Jour-nal of Vocational Behavior, 17: 263-290.

Murphy, J. V. 1992. Driver friendly decisions. TrafficWorld, March 30: 4.

Murphy, K. R., & Cleveland, J. N. 1991. Performanceappraisal: An organizational perspective. Boston:Allyn & Bacon.

Naylor, J. C, Pritchard, R. D., & Ilgen, D. R. 1980. Atheory of behavior in organizations. New York: Ac-ademic Press.

Naylor, J. C, & Shine, L. C. 1965. A table for determiningthe increase in mean criterion score obtained byusing a selection device. Journal of Industrial Psy-chology, 3: 33-42.

Osterman, P. 1987. Turnover, employment security, andthe performance of the firm. In M. M. Kleiner, R. N.Block, M. Roomkin, & S. W. Salsburg (Eds.), Humanresources and the performance of the firm: 275-317. Madison, WI: Industrial Relations Research As-sociation.

Osterman, P. 1994. How common is workplace transfor-mation and who adopts it? Industrial and LaborRelations Review, 47: 173-188.

Park, H. Y., Ofori-Dankwa, J., & Bishop, D. R. 1994. Or-ganizational and environmental determinants offunctional and dysfunctional turnover: Practical andresearch implications. Human Relations, 47: 353-366.

Parsons, D. O. 1977. Models of labor market turnover: Atheoretical, and empirical summary. In R. G. Elren-bery (Ed.), Research in labor economics: 185-223.Greenwich, GT: JAI Press.

Perser, J. H. 1994. Lumping, long hours, absence fromhome continue to fuel truckload driver shortage.Traffic World, February 4: 31-33.


Powell, I., Montgomery, M., & Cosgrove, J. 1994. Com-pensation structure and establishment quit and firerates. Industrial Relations, 33: 229-248.

Price, J. L. 1977. The study of turnover. Ames: Iowa StateUniversity Press.

Quinn, R. P., & Mangione, T. W. 1973. The 1969-1970survey of working conditions: Chronicles of an un-finished enterprise. Ann Arbor, MI: Survey Re-search Center.

Roberts, K. H., Hulin, C. L., & Rousseau, D. M. 1978.Developing an interdisciplinary science of organi-zations. San Francisco: Jossey-Bass.

Schmidt, F. L., & Hunter, }. E. 1983. Individual differ-ences in productivity: An empirical test of estimatesderived from studies of selection procedure utility.Journal of Applied Psychology, 68: 407-414.

Schmidt, F. L., Hunter, J. E., Outerbridge, A. N., & Tratt-mer, M. H. 1986. The economic impact of job selec-tion methods on size, productivity, and payroll costsof the federal workforce: An empirically-based ex-amination. Personnel Psychology, 39: 1-30.

Smith, M. J., Cohen, B. G., & Stammerjohn, L. W., Jr.1981. An investigation of health complaints and jobstress in video display operations. Human Factors,23: 387-400.

Taylor, H. C, & Russell, J. T. 1939. The relationship ofvalidity coefficients to the practical effectiveness oftests in selection. Journal of Applied Psychology,23: 565-578.

Thibaut, J., & Walker, L. 1975. Procedural justice: Apsychological analysis. Hillsdale, NJ: Erlbaum.

Tsai, S. P., Bernacki, E. J., & Lucas, L. J. 1989. A longitu-dinal method of evaluating employee turnover. Jour-nal of Business and Psychology, 3: 465-473.

Tsui, A. S., Pearce, J. L., Porter, L. W., & Tripoli, A. M.1997. Alternative approaches to the employee-orga-nization relationship: Does investment in employeespay off? Academy of Management Journal, 40:1089-1121.

Wachter, M. L., & Kim, C. 1979. Labor supply: Finalreport for ONR. Philadelphia: Wharton Economet-ric Forecasting Associates.

Wexley, K. N., & Latham, C. P. 1991. Developing andtraining human resources in organizations. NewYork: HarperCollins.

Williams, M. L., & Dreher, G. F. 1989. Compensationsystems attributes and applicant pool characteris-tics. Academy of Management Journal, 35: 571-595.

Williamson, O. 1981. The modern corporation: Origins,evolution, attributes. Journal of Economic Litera-ture, 19: 1537-1568.

Wilson, N., & Peel, J. M. 1991. The impact of absenteeismand quits on profit-sharing and other forms of em-ployee participation. Industrial and Labor Rela-tions Review, 44: 454-468.

Jason D. Shaw is an assistant professor of management atDrexel University. He received his Ph.D. in managementfrom the University of Arkansas. His current researchinterests include the consequences of financial incen-tives, the link between compensation systems and orga-nizational effectiveness, and person-environment con-gruence issues.

John E. Delery is an associate professor of management atthe University of Arkansas. He holds a Ph.D. in manage-ment from Texas A&M University. His current researchinterests include the strategic management of human re-sources, the structure of human resource managementsystems, and selection interviews.

G. Douglas Jenkins, Jr., was a professor of management atthe University of Arkansas. He received his Ph.D. inorganizational psychology from the University of Michi-gan. His research interests included compensation sys-tems, job design, and research methods. He is now de-ceased.

Nina Gupta is a professor of management at the Univer-sity of Arkansas. She received her Ph.D. in organizationalpsychology from the University of Michigan. Her currentresearch interests include the behavioral and organiza-tional consequences of reward systems, political aspectsof human resource management, and the dynamics ofabsenteeism and turnover.

an organization-level analysis of voluntary and … · 2006-02-01 · g. douglas jenkins, jr., is...

Documents