Turnover Contagion: How Coworkers' Job Embeddedness and Job Search

Home , Job attitude, Job embeddedness

௠ Academy of Management Journal 2009, Vol. 52, No. 3, 545–561.

TURNOVER CONTAGION: HOW COWORKERS’ JOB EMBEDDEDNESS AND JOB SEARCH BEHAVIORS INFLUENCE QUITTING

WILL FELPS Erasmus University

TERENCE R. MITCHELL University of Washington, Seattle

DAVID R. HEKMAN University of Wisconsin–Milwaukee

THOMAS W. LEE University of Washington, Seattle

BROOKS C. HOLTOM Georgetown University

WENDY S. HARMAN Central Washington University

This research developed and tested a model of turnover contagion in which the job embeddedness and job search behaviors of coworkers influence employees’ decisions to quit. In a sample of 45 branches of a regional bank and 1,038 departments of a national hospitality firm, multilevel analysis revealed that coworkers’ job embeddedness and job search behaviors explain variance in individual “voluntary turnover” over and above that explained by other individual and group-level predictors. Broadly speaking, these results suggest that coworkers’ job embeddedness and job search behaviors play critical roles in explaining why people quit their jobs. Implications are discussed.

As the global economy becomes increasingly On the macro side, economic research often looks knowledge based, organizations that can success- at particular industries or localities to explain how fully retain their human resources have an advan- market forces such as unemployment rates or job tage over organizations that cannot. Indeed, a num- supply and demand affect the frequency with ber of studies have shown that turnover negatively which people leave their jobs (e.g., Banerjee & Gas- effects performance (e.g., Shaw, Gupta, & Delery, ton, 2004). Sociological research has also looked at 2005). Hatch and Dyer summarized such findings how turnover affects and is affected by institutional with the observation that “firms with high turnover changes within and across industries (e.g., Have- significantly under-perform their rivals” (2004: man, 1995), as well as organizational variables such 1155). As such, organizational leaders are inter- as size (Price, 1977). ested in understanding why people choose to leave The unique contribution of management scholar- their jobs and insights that might help with em- ship is not only to investigate the individual or ployee retention (Ulrich & Smallwood, 2006). Ac- institutional level, but also what emanates from the cordingly, researchers have spent considerable ef- careful exploration of “the space in between” fort developing and testing models to explain why (Bradbury & Lichtenstein, 2000). For this reason, people quit. organizational researchers are often encouraged to To explain the phenomenon of employee turn- do “meso-level” research, in which individuals are over, the social sciences have offered both psycho- studied in their social contexts (e.g., House, Rous- logical (i.e., micro) and organizational and eco- seau, & Thomas-Hunt, 1995; Johns, 2006). How- nomic (i.e., macro) explanations. On the micro ever, there is surprisingly little work on how social side, job satisfaction and organizational commit- relationships affect turnover. To quote Pfeffer, ment have captured most of the research interest. “Turnover has most often been examined as the

545 Copyright of the Academy of Management, all rights reserved. Contents may not be copied, emailed, posted to a listserv, or otherwise transmitted without the copyright holder’s express written permission. Users may print, download or email articles for individual use only. 546 Academy of Management Journal June consequence of an individual decision process, port, coworker support, and distributive justice with the individual acting in isolation.... Virtu- (Price, 1977; Price & Mueller, 1986). It is important ally all of the dominant models of turnover concep- to note that in Price and Mueller’s model, as in tualize it as an individual decision, without con- virtually all other traditional models, various fac- sidering the effect of social structure” (1991: 795). tors influence turnover through their impact on Although Pfeffer’s comment overlooks the work of organizational commitment and job satisfaction, economists and sociologists, he is broadly correct which in turn influence intent to leave, which then in stating that the bulk of management research on leads to voluntary turnover. turnover focuses on individual attitudes as the sole The result of subsequent scholarship based on precursor to leaving. The influence of one’s imme- these ideas is both impressive and troublesome. It diate coworkers on turnover decisions (what Pfeffer is impressive in that turnover theory and research describes as social structure) has been largely have proceeded programmatically in such a way ignored. that researchers can be confident about a pair of This article investigates the social dimensions of assertions. First, less satisfied and less committed quitting and offers a model of “turnover contagion” employees think about leaving, look for alternative in which the decision to stay at or leave a job is jobs, are more likely to quit, and do each of these to influenced by coworkers. We provide evidence that a greater degree when they believe that desirable turnover decisions are a domain in which cowork- job alternatives exist. Second, many individual- ers can influence an actor’s thoughts, judgments, and macro-level variables are related to turnover feelings, and behaviors (Salancik & Pfeffer, 1978). through satisfaction and commitment. However, Two field studies support the predictive validity of the turnover literature is also troublesome in that our model, offering new insights into the interper- even the most inclusive models leave the vast ma- sonal precursors of “voluntary turnover” (job leav- jority of variance unexplained (e.g., Griffeth, Hom, ing). We argue that this type of meso-level research & Gaertner, 2000; Maertz & Campion, 1998; Price & can widen researchers’ conceptual lenses, increase Mueller, 1986). A number of authors have therefore our ability to predict turnover, and enhance the suggested that scholars need to expand their con- utility of turnover research for practitioners. ceptual lenses to better understand employee turnover (e.g., Kammeyer-Mueller, Wanberg, Glomb, & Ahlburg, 2005; Maertz & Campion, 1998; Mitchell TOWARD A THEORY OF & Lee, 2001; Mossholder, Settoon, & Henagan, TURNOVER CONTAGION 2005). The framework we describe below, in which we outline the turnover contagion process, is such Turnover Research Heritage an expansionary attempt. March and Simon’s (1958) seminal book, Orga- nizations, marks the real beginning of the attempt The Turnover Contagion Process to develop an overall theory explaining why people leave their jobs. According to March and Simon’s The central theoretical claim made here is that theory, the two factors that determine whether an when an employee’s coworker engages in behaviors employee will leave his or her job are the perceived antecedent to leaving a job, these activities some- desirability of leaving the employing organization times spill over onto others in such a way that the (conceptualized as job satisfaction and organiza- affected others are more likely to leave. Put more tional commitment) and the perceived ease of leav- precisely, a coworker’s search for job alternatives or ing the organization (conceptualized as the quality actual quitting can spread, through a process of of job alternatives). The research focusing on job social contagion, to affect another employee’s quit- satisfaction and organizational commitment, in ting behavior. Like the contagion of illness, the particular, has been extensive. Mobley (1977) iden- process involves the transmission of something tified the sequential and intermediary variables from one individual to another. For us, the “some- leading from job dissatisfaction to eventual quit- thing” is the tendency to leave a job. Others have ting. In an exemplar of programmatic turnover re- used the contagion metaphor to understand the search, Price and Mueller (1986) added to this spread of burnout (Bakker & Schaufeli, 2000), emo- model by cataloging the antecedents of organizations (Barsade, 2002), and long work hours (Brett & tional commitment and job satisfaction, including Stroh, 2003). pay, social integration, instrumental communica- We believe that the primary mechanism in turn- tion, formal communication, centralization, routi- over contagion is people’s pervasive tendency to nization, role overload, promotional opportunity, compare themselves to others. Research on social professionalism, general training, supervisor sup- comparison has documented that this tendency is 2009 Felps, Mitchell, Hekman, Lee, Holtom, and Harman 547 among the most robust and ubiquitous of psycho- was related to individual employee absenteeism. logical phenomena (Kruglanski & Mayseless, 1990). And Eder and Eisenberger (2008) demonstrated “The notion that people rely on others to help that the average tardiness of work group members define reality in ambiguous circumstances has long is related to individual tardiness. They also been a core tenet in social psychology” (Degoey, showed, in a second study, that withdrawal behav- 2000: 58). Salancik and Pfeffer (1978) extended iors carried out at the group level, such as taking Festinger’s (1954) original work on social compar- undeserved work breaks or engaging in idle conver- ison to organizational behavior and job attitudes, sation, influence the probability that individuals and Bandura (1977) applied these insights to learn- do the same. Thus, the idea that withdrawal behav- ing theory. Social comparisons are especially likely iors of group members can influence an individu- to be made in novel, risky, or ambiguous situations al’s likelihood of engaging in those behaviors (Festinger, 1954; Tesser, Campbell, & Mickler, clearly has some precedent. Importantly, we do not 1983; Wooten & Reed, 1998). When comparisons presume here that either job satisfaction or organi- reveal differences with a relevant other’s thoughts, zational commitment plays a key role in the pro- feelings, or behaviors, an individual’s propensity to cess. The turnover contagion model highlights the change his or her understanding of a situation so role that simply observing others plays and sug- that thoughts, feelings, and behaviors become con- gests that a key determinant of whether quitting is sistent with those of the relevant other increases a viable option at any given point in time is (Festinger, 1954). Chartrand and Bargh stated this: whether coworkers are leaving. “Throughout the history of psychology, many have argued that the act of perceiving another person’s From Theory to Hypotheses behavior creates a tendency to behave similarly oneself” (1999: 813). Above we have presented a theory of turnover The application to turnover theory is straightfor- contagion whereby the tendency to quit spreads ward. Given that high levels of risk and uncertainty throughout a work group. We now offer two spe- often characterize job transition (Steel, 2002), we cific hypotheses about factors that are central to the expect employees to look to others in evaluating turnover contagion process. First, we hypothesize whether to seek alternative employment. When a that turnover contagion is most likely to occur number of coworkers are looking for other jobs, it when the coworkers around a focal employee are may increase the salience and perceived viability of not “embedded” in their jobs. We choose to focus leaving for a focal employee, especially since im- on job embeddedness, as opposed to job satisfac- mediate coworkers are likely targets for social com- tion or organizational commitment, because it is a parison (Kulik & Ambrose, 1992). Conversely, if broader construct that captures a greater range of few coworkers are looking for other jobs, it is likely factors that provoke leaving. In Mitchell, Holtom, that a focal employee will be less inclined to initi- Lee, Sablynski, and Erez’s (2001) original formula- ate the turnover process. In either case, social com- tion, the job embeddedness construct addressed parison helps to answer the question, “Should I how well people fit in their jobs (e.g., personal consider leaving?” We posit that the chance that skills are well suited to the work assigned) and the answer will be yes increases when many co- community (e.g., they like the amenities a commu- workers are looking for jobs. In this way, the trans- nity provides); the interpersonal links they have on mission of a tendency to leave occurs as employees and off the job (e.g., their number of ties to people watch and converse with their coworkers. The focal and groups); and what they would have to give up person may observe such job search behaviors in a or sacrifice in leaving their place of employment or dyadic interaction (e.g., “I am going on a job inter- community (e.g., what opportunities they would view this week”) or in a group interaction (e.g., forego). In sum, job embeddedness includes several “You all should probably know that I have a job individual-level factors that enmesh employees in interview this week”). Moreover, there are a variety their jobs, and numerous studies have shown it to of leaving behaviors that can be observed; the em- be a good predictor of an employee’s tendency to ployee might see a coworker update a re´sume´, quit (Allen, 2006; Crossley, Bennett, Jex, & Burn- search classified ads, or schedule interviews. In field, 2007; Holtom, Mitchell, & Lee, 2006; Holtom short, a range of behaviors may indicate that one or & O’Neill, 2004; Lee, Mitchell, Sablynski, Burton, & multiple coworkers are in the process of leaving. Holtom, 2004; Mitchell et al., 2001; Van Dijk & Some research has addressed the topic of with- Kirk-Brown, 2003; Zatzick & Iverson 2006). In drawal caused by group-level variables. Mathieu many of these studies, job embeddedness has gone and Kohler (1990), for example, found that the fre- beyond job satisfaction and organizational commit- quency of absenteeism among work group members ment in predicting variance in individual turnover. 548 Academy of Management Journal June

When coworkers’ job embeddedness is low, we better understand the turnover process. A qualita- believe that the resultant social context will make tive analysis of the behaviors discussed by the fo- individuals more likely to entertain the possibility cus group members provides some information of changing jobs. When coworkers are not tethered about how employees may be influenced by their to—embedded in—a job, they are likely to be open coworkers’ comments about leaving. In Study 2, we to the possibility of leaving. It is this willingness to sought to specifically measure coworkers’ searches leave that transfers from low-embeddedness co- for alternative employment using the Job Search workers to a focal employee in their work unit. Behavior Index (Kopelman, Rovenpor, & Millsap, Thus, we expect the average job embeddedness of 1992). These authors reported that this measure coworkers to predict focal employee turnover. Fur- (aggregated to the group level) did an excellent job ther, since job embeddedness is a broad construct of predicting leaving and internal transfer and went that includes nonaffective elements such as the over and above eight affective, perceptional, attitu- number of links to important others and family ties, dinal, and intention measures (e.g., organizational we would expect this effect to be observed even commitment, intent to stay, and general job satis- when a focal employee is satisfied with the work faction) in such prediction. For our purposes, the itself or committed to his/her organization. argument is simple. When an employee sees and Let us briefly take some examples from the hears about coworkers looking for other jobs, leav- Mitchell et al. (2001) job embeddedness measure to ing becomes a more salient option for her/him, provide a more grounded understanding of how which leads to a greater propensity to quit. Figure 1 turnover contagion might operate. Imagine a work- presents a summary of these ideas. It should be place where most people strongly agree with the noted that although both of the studies described following statements: “I feel like I am a good match below measured coworkers’ job embeddedness, for my organization,” “I really love the place I live,” only the second study assessed coworker job search “I would sacrifice a lot if I left this job,” “My family behavior using the Kopelman et al. (1992) measure. roots are in this community,” and “I work closely Thus, Hypothesis 2 is only empirically tested in with my coworkers.” Interactions among employ- Study 2. ees who feel this way are likely to mutually rein- force each other’s perceptions that “I belong here, I Hypothesis 2. Coworkers’ job search behaviors should be here, and I must remain here.” In such a mediate the negative relationship between co- setting, people are unlikely to be looking at want workers’ job embeddedness and focal em- ads, talking about available jobs elsewhere, or say- ployee turnover. ing things that indicate they want to leave. Contrast this situation with a workplace populated by those From Hypotheses to Analytics who are less embedded in their jobs and communities (e.g., people who feel they don’t fit in their Although undisputedly important, pursuing work group or community, or people who have meso-level research can be challenging (House et little to sacrifice in renegotiating their relationships al., 1995). When one defines meso-level research as to their jobs). In this sort of environment, even if research that includes activities and processes that they like their jobs, employees have little to lose by take place between the micro and macro, this chal- voicing ideas about leaving or about alternative lenge comes into stark relief. “Micro” and “macro” avenues of employment (Bartunek, Huang, & are defined relative to each other, and there are a Walsh, 2008). Frequent discussions about leaving number of potentially relevant “levels” for both are likely to prime other employees, possibly even predictors and criteria, including individuals, dy- those who are fairly embedded, to consider quit- ads, small groups, organizations, industries, and ting. Thus, we hypothesize the following: societies. The number of possible combinations is extensive; comments by Klein and Kozlowski Hypothesis 1. Coworkers’ job embeddedness is (2000) were helpful for our definitional analysis. negatively related to voluntary turnover. We are particularly interested in how the behaviors The next question naturally follows: How does a that occur in dyads or existing groups (our inde- willingness to quit engendered by low job embed- pendent variable) influence individual members to dedness influence others? As noted previously, we quit (our dependent variable). Rousseau (1985) de- hypothesize that the transmission of this leaving scribed such an influence process as a cross-level tendency occurs as employees watch and converse phenomenon. In our case, the phenomenon of in- with coworkers searching for alternative employ- terest is turnover contagion. More specifically, each ment. In Study 1 (see below), we gathered data person’s job embeddedness reflects an overall through a series of focus groups designed to help us “stuckness” in the job (the inverse of which is a 2009 Felps, Mitchell, Hekman, Lee, Holtom, and Harman 549

FIGURE 1 The Turnover Contagion Model

Coworkers’ Job Embeddedness Coworkers’ Job Search (average job embeddedness Behavior (qualitative and score of employees in quantitative) department or branch)

Level 2 Variables

Level 1 Variables

Links–Organization

Links–Community

Individual Fit–Organization Individual Job Voluntary Embeddedness Fit–Community Turnover

Sacrifice–Organization

Sacrifice–Community

willingness to leave), which can be contagiously embeddedness or job search behavior in any mul- transferred through modeling or direct interaction tilevel statistical test (Klein & Kozlowski, 2000). All with coworkers. In actual work situations, individ- subsequent analyses do this. Second, this measure- uals may work at various times with just one other ment process does not depend on employees com- individual (i.e., dyadically) or, as is increasingly ing to some sort of socially agreed upon consensus common, in project teams, departments, and inde- about job embeddedness or about job search behav- pendent branches (i.e., in a small group). In a small iors. For example, one popular type of meso-level group, any given individual is likely to send turn- research links group-level consensus about some- over contagion stimuli to a number of others as well thing (e.g., norms, mood, etc.) to individual behav- as receive this sort of leaving stimuli from a number ior. Chan described these as “direct consensus of others. models” (1998). Our theoretical model is not one of If a researcher wants to capture the cumulative direct consensus. As such, the methodological influence of coworkers on a focal actor’s turnover standards used to verify direct consensus effects decision, it is necessary to somehow combine the (i.e., high degrees of agreement as assessed by in- contagious effects of multiple group members traclass correlations, or Rwg statistics) would be (Klein & Kozlowski, 2000). The simplest way to do meaningless for our analysis. Instead, coworkers’ this is to aggregate the scores of coworkers who are job embeddedness represents what Chan (1998) known to work closely with an individual on those called an “additive index model” that does not variables that have been theorized to be associated hinge upon agreement, but is instead about with the turnover contagion process—in short, job whether relevant social comparisons prompt look- embeddedness and job search behaviors. Thus, we ing for a different job. propose to test whether the average of coworkers’ job embeddedness scores for a natural group influences individual employee turnover and, if so, STUDY 1 whether the coworkers’ average level of job search Methods behaviors mediates this relationship. It is important to point out two statistical issues Sample. Our first research site was a large recre- related to this conceptualization. First, given that a ation and hospitality organization, hereafter re- focal individual is nested in a group, it is important ferred to as Funcorp. This organization operates to control for a focal individual’s own level of job roughly 200 golf courses, country clubs, private 550 Academy of Management Journal June business and sports clubs, and resorts. Funcorp used to measure job embeddedness. This measure provides services to about 200,000 member fami- was developed using data collected from 769 cor- lies throughout the United States. Our initial sam- rections officers. Given the fact that the short-form ple consisted of 14,981 Funcorp employees who items are also represented in the long form, we serve its members. Nine thousand seventy-nine em- would expect this correlation to be very high. More ployees completed our survey, for a response rate importantly, after job satisfaction was controlled of 60.6 percent. Missing values reduced the number for, the long-form measure of individual job embed- of usable observations to 8,663, or 57.8 percent of dedness significantly predicted voluntary turnover the initial sample. Within our usable sample were (p Ͻ .001), as did the short-form measure (p Ͻ 1,037 club departments. Overall, 39.3 percent of .001), which provides evidence of predictive valid- the respondents were women; the average age was ity for this shorter measure. Further, there was no 39.0 years; the average tenure with the organization difference in the amount of variance in turnover was 6.2 years; and 32.6 percent were nonwhites. explained by the two forms of the instrument. The average department size was 14.4, and the In both samples, the respondents indicated on a average number of survey respondents per depart- five-point scale the extent to which they agreed ment was 8.35. The firm provided demographic with 18 of the 21 items. The other 3 items involved data for all employees, allowing us to statistically yes or no answers. We standardized and averaged compare respondents and nonrespondents. These each individual’s scores for each item to create an comparisons yielded no significant differences in individual-level job embeddedness score. These in- gender, age, tenure, race, or turnover rate, provid- dividual job embeddedness scores were then aver- ing some confidence that nonresponse bias was not aged across employees in each department to create a concern. an aggregate of departmental job embeddedness Measures. Voluntary turnover was measured in (i.e., departmental coworkers’ job embeddedness). Study 1 as whether an employee voluntarily left the The Appendix reports the survey’s items. Because organization in the 18 months immediately follow- individual job embeddedness is a formative (or in- ing the survey. An 18-month period was reasonable dicator) construct, high internal consistency (e.g., because it allowed enough time for the indepen- as measured by coefficient alpha) and unidimen- dent variables to influence employees’ turnover de- sionality (e.g., as shown by one-factor-model supe- cisions and provided us with a large enough sam- riority) are not the standards by which construct ple to reliably run statistical tests. Specifically, validity should be judged (Diamantopoulos & Win- 2,001 of the employees surveyed choose to leave klhofer, 2001). However, for descriptive purposes, Funcorp. This number corresponds to an 18-month we note that coefficient alpha was high (␣ ϭ .88). voluntary turnover rate of 23.1 percent (or 15.4 Control variables. Given that we wanted to test percent annually). coworkers’ job embeddedness as a predictor of fo- In prior studies of job embeddedness, researchers cal employee turnover, we sought to control for have relied on a 40-item measure to capture the six other variables that might provide alternative ex- subdimensions that were then aggregated to create planations. These control variables included both composite measures (e.g., Mitchell et al., 2001). In the individual (level 1) factors of job embedded- defining the construct, Mitchell et al. (2001) char- ness, job satisfaction, organizational commitment, acterized job embeddedness as a formative indica- part-time versus full-time status, age, gender, race, tor construct, in that multiple variables are asso- and tenure, as well as the group (level 2) factors of ciated with the embeddedness construct and coworkers’jobsatisfaction,coworkers’organization- predictive validity represents the major mechanism al commitment, department size, and local unem- for validation of its conceptual meaning (Edwards, ployment rate. Job satisfaction assessed the degree 2001). In other words, job embeddedness captures a to which employees expressed satisfaction with ten large set of things that enmesh people in their jobs dimensions of their jobs (e.g., pay, coworkers, pro- and that predict voluntary turnover. motion, etc.) using a shortened version of Spector’s In the present study, we assessed the degree to (1985) job satisfaction measure. Spector’s original which an employee’s coworkers were enmeshed in scale includes 36 items, but because of survey the organization and community (coworkers’ job length constraints our shortened measure included embeddedness) using a 21-item measure of job em- only the 2 best-loading items for each subscale (as beddedness developed and validated by Holtom, based on Spector [1985]). Thus, the respondents Mitchell, Lee, and Tidd (2006). In their measure indicated on a five-point scale the extent to which development study, the product-moment correla- they agreed with 20 items assessing satisfaction tion showed a strong relationship between the orig- with various aspects of their jobs. Coefficient alpha inal long form and the revised short form (r ϭ .92) for job satisfaction was .93. We measured organi- 2009 Felps, Mitchell, Hekman, Lee, Holtom, and Harman 551 zational commitment using four items from Meyer, TABLE 1 Allen, and Smith’s measure of affective organiza- Study 1 Means, Standard Deviations, and a tional commitment. Respondents indicated on a Correlations of Level 2 Variables five-point scale the extent to which they agreed Variables Mean s.d. 1 2 3 4 with the items. Coefficient alpha for this measure was .85. The employees’ full or part-time work sta- 1. Group size 8.35 1.50 tus was determined from organizational records at 2. Local unemployment rate 4.55 1.01 Ϫ.08 the time the employee completed the survey (0 ϭ 3. Coworkers’ organizational 4.04 0.48 Ϫ.13 .01 “full time,” 1 ϭ “part time”). Part-time employees commitment 4. Coworkers’ job satisfaction 3.90 0.42 Ϫ.02 .04 .63 worked a maximum of 32 hours per week and did 5. Coworkers’ job 3.79 0.31 Ϫ.09 .00 .64 .59 not receive benefits, whereas full-time employees embeddedness were expected to work at least 40 hours per week a and received benefits. We obtained the demo- k ϭ 1,037 departments; all correlations greater than .03 are Ͻ graphic variables age, gender, race, and tenure from significant at p .01. the organizations’ records and entered them as controls. We included these employee demographic tic regression software called hierarchical general- variables in the model because we wanted to have ized linear modeling (HGLM; Guo & Zhao, 2000). confidence that effects were not based on employ- The main difference between hierarchical linear ee’s life experiences, social categories, or career modeling (HLM) and HGLM is that the latter allows position. for binary outcome variables (e.g., stay/quit). In addition, the analysis contained several group HGLM was ideal for our tests because it is designed (level 2) controls because they could also constitute to account for nonindependence between group- potential alternative explanations. These include level predictor variables. Given that HGLM is sim- coworkers’ job satisfaction and coworkers’ organi- ply a type of multilevel logistic regression analysis, zational commitment, which are the individual- normally distributed outcome variables and error level variables of job satisfaction and organization- terms are not necessary. It has been used to study al commitment averaged over department. We multilevel predictors of a wide range of binary out- should reiterate that we were adding group-level comes, including whether a person drops out of job satisfaction and group-level organizational high school, completes college, marries, divorces, commitment simply as conservative controls. Since or goes bankrupt (see Guo and Zhao, 2000). HGLM they are major predictors of turnover at the indi- helps one to disentangle individual-level effects vidual level, they may also control variance in turn- from social effects by statistically disaggregating over when assessed at the group level. However, individual (level 1) and group (level 2) effects. In we are not postulating that they necessarily op- sum, we used this form of analysis because it pro- erate through a contagion process similar to co- vided the least biased and most informative workers’ job embeddedness (although they method of hypothesis testing in this context. could). Department size was assessed as the number of employees in each branch or department. Results Local unemployment rate was obtained from the Bureau of Labor Statistics for each zip code Tables 1 and 2 report the means, standard devi- where a club was located. ations, and correlation coefficients between the dependent, independent, and control variables for the level 1 and level 2 variables. Table 3 presents the Analysis results of the HGLM analysis. Employees who share a department have the Hypothesis 1 posited a negative relationship be- same coworkers’ satisfaction, coworkers’ commit- tween coworkers’ job embeddedness and voluntary ment, and coworkers’ embeddedness scores. To ig- turnover. Table 3 (model 1) shows a negative and nore this dependence by using normal logistic re- significant relationship between coworkers’ job gression would violate a core assumption of embeddedness and individual voluntary turnover regression analysis. Even excluding the focal actor (␤ ϭϪ.19, p Ͻ .001). We further suggested that from each aggregated score would leave highly in- coworkers’ job embeddedness would predict turn- terdependent aggregated scores. In fact, aggregated over even when coworkers’ job satisfaction and scores with the focal actor excluded are almost organizational commitment are controlled. As identical to aggregated scores with the focal actor shown in Table 3 (model 2), coworkers’ job embed- included (i.e., the average correlation is .95). There- dedness remains significantly predictive of turn- fore, the data were analyzed with multilevel logis- over (␤ ϭϪ.16, p Ͻ .001), and neither coworkers’ 552 Academy of Management Journal June

TABLE 2 Study 1 Means, Standard Deviations, and Correlations of Level 1 Variablesa

Variables Mean s.d. 12345678

1. Voluntary turnover 0.23 0.42 2. Age 39.98 14.45 Ϫ.21 3. Tenure 7.20 6.79 Ϫ.17 .41 4. Gender 0.39 0.49 .01 Ϫ.07 Ϫ.03 5. Race 0.33 0.47 Ϫ.08 Ϫ.04 .05 Ϫ.12 6. Work status 0.26 0.44 .07 Ϫ.05 Ϫ.24 .05 Ϫ.17 7. Organizational commitment 3.97 0.87 Ϫ.19 .21 .14 Ϫ.06 .12 Ϫ.13 8. Job satisfaction 3.89 0.72 Ϫ.08 .00 Ϫ.02 Ϫ.06 .12 .04 .62 9. Job embeddedness 3.76 0.55 Ϫ.17 .20 .15 Ϫ.01 .06 Ϫ.01 .63 .57

a n ϭ 8,663 individuals; all correlations greater than .02 are significant at p Ͻ .01. job satisfaction nor coworkers’ organizational com- cant, negative predictors of voluntary turnover mitment remains as a significant predictor of turn- (model 2: satisfaction, ␤ ϭϪ.07, p Ͻ .05; commit- over in this “competitive test” model. Thus, Hy- ment, ␤ ϭϪ.16, p Ͻ .01; embeddedness, ␤ ϭϪ.09, pothesis 1 is supported. Although replication was p Ͻ .01). not the focus of this research, Table 3 does show results that replicate prior research. Specifically, Supplementary Qualitative Analysis individual-level job satisfaction, organizational commitment, and job embeddedness are signifi- Our model suggests that aggregate job embeddedness influences individual turnover through conta- TABLE 3 gion of job search behaviors. In Study 1, we did not Study 1 HGLM Logistic Regression Results Predicting quantitatively measure job search behaviors. In- Individual Voluntary Turnovera stead, we conducted content analyses of a deduc- tive nature based on 11 focus groups at both of our Individual Turnover research sites. To get a wide variety of responses, we selected Funcorp and Cashcorp1 sites where Variables Model 1 Model 2 employee turnover from prior years was high (six focus groups) and other sites where turnover was Level 2 Group size Ϫ.01 Ϫ.01 low (five focus groups). Eisenhardt and Graebner Local unemployment rate Ϫ.03 Ϫ.03 (2007) called this method “contrasting polar Coworkers’ job embeddedness Ϫ.19*** Ϫ.16*** types.” Focus groups were conducted before the Coworkers’ organizational Ϫ.06 survey was sent out in both samples. We asked commitment focus group participants to tell us about their jobs Coworkers’ job satisfaction .01 and why people stayed or left. The focus groups Level 1 lasted 90–120 minutes each and were attended by Age Ϫ.47*** Ϫ.48*** an average of eight employees each. All interviews Ϫ Ϫ Tenure .62*** .61*** were audiotaped and later transcribed verbatim. Gender Ϫ.05* Ϫ.05* Race Ϫ.14*** Ϫ.15*** One of the leading measures of job search behav- Work status .09** .09** ior (Kopelman et al., 1992) asks respondents to note Job embeddedness Ϫ.10** Ϫ.09** which, if any, of ten different search behaviors they Organizational commitment Ϫ.17*** Ϫ.16*** have engaged in during the prior year. The behav- Ϫ Ϫ Job satisfaction .05* .07* iors include revising one’s re´sume´, going on a job interview, and talking with coworkers about getting Log-likelihood Ϫ794.45 Ϫ802.81 a new job. In the focus groups, we were careful not Ϫ ␤ Ϫ ␤ b 2[L( reduced) L( full)] 16.72*** to put any of the participants under pressure by a To enhance ease of interpretation, we report standardized asking questions about revising re´sume´s or going coefficients. n ϭ 8,663 individuals; k ϭ 1,037 departments. on job interviews. However, when we asked about b The significant change in log-likelihood indicates that the reasons why people stay, we noted that many model 2 is significantly worse at predicting turnover than model 1. The loss of two degrees of freedom may be the cause. * p Ͻ .05 1 ** p Ͻ .01 This is our pseudonym for the bank providing Study *** p Ͻ .001 2 data. 2009 Felps, Mitchell, Hekman, Lee, Holtom, and Harman 553 spontaneous comments about leaving emerged. ses about what causes people to search and leave. Moreover, it seemed to be much more acceptable to The group’s average levels of satisfaction (r ϭϪ.10, discuss leaving in the high-turnover locations. n.s.) and commitment (r ϭϪ.27, n.s.) were not Consequently, we asked two of the authors who significantly correlated with the number of com- were not involved with conducting the focus ments about leaving. However, and consistently groups to use Atlas.ti qualitative software (a quali- with the turnover contagion model, the group’s tative analysis tool that helps users organize, lo- level of job embeddedness was significantly, nega- cate, code, and annotate findings from large vol- tively correlated with the number of comments umes of qualitative documents) to independently about leaving (r ϭϪ.64, p Ͻ .05). This qualitative count the comments about leaving (e.g., reasons for finding regarding coworkers’ job embeddedness is leaving, alternative job options, people who had considerably more speculative than our subsequent left or were considering leaving). To ensure the quantitative findings reported for Study 2. As men- coding process was blind, all focus-group-identify- tioned by Lee (1999), qualitative research is not ing information was removed from the transcripts. suited to discussions of prevalence, generalizabil- Before coding, the two judges discussed how they ity, or calibration. Qualitative research, however, is would count leaving reasons. For example, they well suited to discussions of description, interpre- agreed that when a single focus group participant tation, and explanation. Thus, these findings in- listed several leaving reasons, they would count creased our confidence in our conceptual under- each reason as unique. One coder counted 158 leav- standing and encouraged us to further test whether ing reasons in the 11 focus groups, and the other coworker job search mediates the relationship be- coder counted 163 leaving reasons. Together, the tween coworkers’ job embeddedness and focal em- two judges identified 168 leaving reasons, of which ployee turnover. 156 were the same, for 93 percent agreement. All disagreements were resolved in a discussion between the two judges, and the judges ultimately STUDY 2 agreed on a final count of 158 leaving reasons. Hypotheses To assess spontaneous discussions about leaving, the coders counted the number of reasons for leav- In Study 2, we sought not only to replicate the ing that employees publicly stated in each group. results of Study 1 but also to gain greater under- Employees mentioned that their coworkers left to standing of the coworker behaviors that explain the obtain more pay, better opportunities or benefits, effect of coworkers’ job embeddedness on individ- and less physically demanding jobs, or to go back to ual employee turnover. Recall that Hypothesis 2 school. For example, one Cashcorp employee at a holds that coworkers’ job search behavior mediates branch where the coworkers’ job embeddedness the effect of coworkers’ job embeddedness on indi- score was low made the comment, “Did you know vidual turnover. As described in the theory devel- that at [alternative company], the pay starts at $9 or opment section, contagion is a process by which $10 and they reimburse 100% of tuition? If I saw turnover propensity spreads from coworkers to a that they were hiring, I could see myself leaving.” focal actor. This process is hypothesized to occur After conducting the focus groups, we surveyed the when employees model each other’s leaving-re- employees as part of the broader quantitative por- lated behaviors (i.e. resume revision, reading the tion of our study, as we describe in the Methods classifieds, going on job interviews, etc.). However, sections pertaining to Studies 1 and 2. From each in Study 1 we did not directly measure job search focus group participant’s individual job embedded- behavior. In Study 2 we attempted to assess di- ness, commitment, and satisfaction scores, we cal- rectly if job search behavior was more common culated each focus group’s average level of job em- where employees were not embedded and if co- beddedness, commitment, and satisfaction. We workers’ search behavior mediated the relationship imputed the organizational average score to each of between coworkers’ job embeddedness and focal the 5 focus group participants (out of 88) who did employee turnover. not fill out a survey. The survey data gathered from focus group participants were also included in the Methods broader HGLM analysis. The number of coded comments about leaving (a proxy for job search behav- Sample. Our second site was a retail bank in the ior) was then correlated with the group’s average U.S. Midwest. Cashcorp owns and operates 45 level of job embeddedness, commitment, and branch offices in two states and has roughly two satisfaction. billion dollars in assets. We sent a survey to all 486 Our findings were consistent with our hypothe- employees. Three-hundred and twenty employees 554 Academy of Management Journal June

TABLE 4 Study 2 Means, Standard Deviations, and Correlations of Level 2 Variablesa, b

Variables Mean s.d. 123456

1. Group size 5.20 6.07 2. Local unemployment rate 4.08 2.46 Ϫ.08 3. Coworkers’ job alternatives 3.34 0.41 .08 Ϫ.14 4. Coworkers’ organizational commitment 2.97 0.35 .04 .03 Ϫ.53 5. Coworkers’ job satisfaction 3.27 0.31 .01 Ϫ.09 Ϫ.52 .66 6. Coworkers’ job embeddedness 2.36 0.27 .19 Ϫ.02 Ϫ.19 .52 .46 7. Coworkers’ search behavior 0.45 0.19 Ϫ.12 .13 .38 Ϫ.60 Ϫ.55 Ϫ.60

a The correlation between coworkers’ job embeddedness and comments about leaving is Ϫ.64, p Ͻ .05. The relationships between coworkers’ job satisfaction and coworkers’ organizational commitment and comments about leaving are not significant. b k ϭ 45 departments; all correlations greater than .19 are significant at p Ͻ .05. completed the survey, for a response rate of 66 Study 1 only assessed affective commitment, and percent. Missing values reduced the number of us- Study 2 used a more comprehensive 18-item mea- able observations to 234 and the final response rate sure of organizational commitment that includes to 48 percent. In our usable sample, respondents affective, normative, and continuance commitment were from 45 branches; 77.1 percent were women; (Meyer, Allen, & Smith, 1993). Both studies used the average age was 37.8 years; the average tenure the same short version of Holtom and colleagues’ with the organization was 6.1 years; and 8.2 per- (2006) job embeddedness scale. Once again, the cent were nonwhites. The average branch size was reliability scores for these scales were high (␣ ϭ 10.8 (5.2 survey respondents per branch). In the .93, job satisfaction; ␣ ϭ .89, organizational com- two years following the survey, 60 employees who mitment; and ␣ ϭ .82, job embeddedness). Finally, completed our survey voluntarily left Cashcorp, we used Griffeth and Hom’s (1988) 5-item Index of which equates to a two-year voluntary turnover rate Perceived Job Alternatives at the individual level to of 25.7 percent (or 12.9 percent per year). Finally, control for the effect of employee perceptions on nonresponse bias was unlikely because employees the job alternative–turnover relationship. Finally, who completed our survey and those who did not as in Study 1, we aggregated five variables (i.e., job were not significantly different in terms of gender, search, job embeddedness, job satisfaction, organi- age, tenure, race, or turnover rate. zational commitment, and job alternatives) to the Measures. The measures and methods were only unit level. Our rationale for aggregating job satis- slightly different from those for Study 1. The most faction, organizational commitment, and job alter- important addition was that Study 2 included the natives was the same as for Study 1. They were ten-item Job Search Behavior Index (Kopelman et seen as conservative controls. al., 1992; ␣ ϭ .83). We aggregated this measure Analysis. The analytic technique also remained to the unit level in order to assess the amount of the same (i.e., HGLM), with the exception that we job search activity occurring in a particular bank tested mediation using both the traditional Baron branch. This index seeks to tap the actual behaviors and Kenny (1986) standard and the Sobel test, involved in looking for a new job and includes which directly assesses the statistical significance items such as, “[During the past year, have you] of the change in regression coefficients when the revised your re´sume´?” “. . . read the classified/ mediator is added to the equation. As Baron and help wanted advertisements in the newspaper?” Kenny (1986) suggested, the Sobel test offers a con- “. . . sent copies of your re´sume´ to a prospective firmatory and rigorous test of mediation. employer?” “. . . talked to coworkers about getting a new job?” “. . . gone on a job interview?” The re- Results sults of Kopelman and colleagues (1992) suggest that this index may be a better and more behavior- Tables 4 and 5 present the level 1 and level 2 ally grounded predictor of employee turnover than descriptive statistics. Table 6 presents results of are intention to leave and attitudinal variables. the regression of the job attitude variables on the Study 2 also employs slightly better measures of mediator (coworkers job search behavior) as well job satisfaction and organizational commitment. as results of the HGLM analysis for actual quit- Whereas the first study used a shortened 20-item ting. The main purpose of this study, though, was version of Spector’s Job Satisfaction Index (1985), to test whether coworkers’ job search behavior Study 2 used the full 36-item scale. In addition, would mediate the relationship between cowork- 2009 Felps, Mitchell, Hekman, Lee, Holtom, and Harman 555

TABLE 5 Study 2 Means, Standard Deviations, and Correlations of Level 1 Variablesa

Variables Mean s.d. 12345678910

1. Voluntary turnover 0.26 0.44 2. Age 37.51 12.15 Ϫ.37 3. Tenure 5.89 7.05 Ϫ.26 .47 4. Gender 0.73 0.45 .10 Ϫ.15 Ϫ.03 5. Race 0.09 0.28 .20 Ϫ.09 Ϫ.04 .06 6. Work status 0.13 0.33 .16 Ϫ.22 Ϫ.13 .13 .09 7. Job alternatives 3.36 0.95 .07 .02 .03 .07 Ϫ.13 Ϫ.06 8. Organizational commitment 2.97 0.62 Ϫ.26 .06 .13 .05 .04 .01 Ϫ.33 9. Job satisfaction 3.28 0.53 Ϫ.24 .04 Ϫ.04 Ϫ.07 Ϫ.02 Ϫ.02 Ϫ.34 .55 10. Job embeddedness 2.41 0.47 Ϫ.28 .30 .27 Ϫ.04 Ϫ.14 Ϫ.17 Ϫ.23 .56 .47 11. Job search behavior index 0.43 0.31 .45 Ϫ.37 Ϫ.33 .03 .10 .06 .19 Ϫ.44 Ϫ.37 Ϫ.49

a n ϭ 234 for all variables; all correlations greater than .13 are significant at p Ͻ .05. ers’ job embeddedness and focal employee vol- workers’ organizational commitment, coworkers’ untary turnover. Results indicate that coworkers’ job satisfaction, and coworkers’ perceptions of job embeddedness was significantly and nega- job alternatives. Next, we regressed voluntary tively related to coworkers’ job search behavior turnover on coworkers’ job embeddedness (mod- (␤ ϭϪ.41, p Ͻ .001) even after we controlled for els 1 and 2 in Table 6). Replicating our first bank branch size, local unemployment rate, co- study’s findings, coworkers’ job embeddedness

TABLE 6 Study 2 OLS and HGLM Regression Results Examining the Influence of Coworkers’ Job Embeddednessa

Individual Turnover

Variables Coworker Job Search Model 1 Model 2 Model 3

Level 2 Group size Ϫ.09 Ϫ.04 Ϫ.10 Ϫ.21* Ϫ.19* Local unemployment rate .20 .20 .16 .06 Ϫ.04 Coworkers’ job alternatives .13 .18 Ϫ.17 Ϫ.26 Coworkers’ organizational commitment Ϫ.32* Ϫ.24 Ϫ.11 Ϫ.04 Coworkers’ job satisfaction Ϫ.18 Ϫ.08 Ϫ.27 Ϫ.29 Coworkers’ job embeddedness Ϫ.41*** Ϫ.92*** Ϫ.68** Ϫ.46 Coworkers’ job search .59**

Level 1 Age Ϫ.58** Ϫ.52* Ϫ.54* Tenure Ϫ.26 Ϫ.18 Ϫ.18 Gender .08 .16 .15 Race .29 .30 .33 Work status .29 .27 .28 Job embeddedness .12 .10 .12 Organizational commitment Ϫ.15 Ϫ.15 Ϫ.15 Job satisfaction Ϫ.10 Ϫ.07 Ϫ.10 Job alternatives Ϫ.10 Ϫ.11 Ϫ.10 Job search behavior index .87*** .90*** .87***

R2 .44 .55 ⌬R2 from previous model .11*** Log-likelihood Ϫ330.25 Ϫ330.55 Ϫ326.63 Ϫ ␤ Ϫ ␤ 2[L( reduced) L( full)] 7.84**

a To enhance ease of interpretation, we report standardized coefficients. n ϭ 234 individuals; k ϭ 45 branches. * p Ͻ .05 ** p Ͻ .01 *** p Ͻ .001 556 Academy of Management Journal June significantly predicted turnover (␤ ϭϪ.92, p Ͻ rion, (3) explaining additional variance over and .001). Then we regressed voluntary turnover si- above that explained by competitive constructs, multaneously on coworkers’ job search behavior and (4) showing the process by which something and coworkers’ job embeddedness (model 3 in has an effect (i.e., establishing mediation). It Table 6). Coworkers’ job embeddedness becomes should be noted that all four of these criteria have nonsignificant when the mediator is added. been demonstrated for coworkers’ job embedded- Moreover, the Sobel test shows that the change in ness. In analyses (1) controlled for demographic the regression coefficient for coworkers’ job em- characteristics (e.g., tenure, age, work status, and beddedness is itself significant (Sobel t ϭϪ8.6, gender), perceived and objective measures of job p Ͻ .001). Thus, Hypothesis 2 is supported. Cowork- alternatives, and department size, coworkers’ job ers’ job search behavior appears to mediate the rela- embeddedness (2) predicts turnover in two dis- tionship between coworkers’ job embeddedness and tinct samples, (3) exceeds the prediction of sim- turnover. As in our first study, we found that when ilarly aggregated variables such as coworkers’ job we included coworkers’ satisfaction and commitment satisfaction and coworkers’ organizational com- variables in the model, only coworkers’ job embed- mitment, (4) and can plausibly be seen to operate dedness was still significant. through the observation of coworkers’ job search The implications of our results are apparent in behaviors. Thus, we can have some confidence the effect sizes and statistical ramifications that coworkers’ job embeddedness is one impor- within our sample. On average, using a simple tant driver of turnover contagion. log-odds transformation, we found that a one But going beyond job embeddedness, we also standard deviation increase in coworkers’ job em- provide evidence through both qualitative and beddedness decreased the probability of an indi- quantitative analyses that coworkers’ job search vidual voluntarily leaving from 15.4 percent per may act as a critical mechanism in the turnover year to 8.5 percent per year at Funcorp, and from contagion process. These findings are conservative 12.9 percent per year to 4.2 percent per year at in that job search behaviors have more recently Cashcorp. This equates to a decrease in voluntary evolved to include Internet job search, job clearing- turnover of 45 percent at Funcorp and 67 percent house websites, and e-mail correspondence about at Cashcorp, given controls for other variables in positions. The Job Search Behavior Index (Kopel- the model. We speculate that the Cashcorp re- man et al., 1992) does not account for these new sults are stronger because the units are smaller ways of searching for a job. and more exclusive. In the confined space of a A particular strength of the study is the replica- bank branch, people saw their unit members tion of findings for two large samples in two very more frequently and were exposed to only their different settings. Cashcorp employees worked in fellow branch members’ leaving behaviors. self-contained branches with relatively few people. Across the two samples, a one standard deviation The employees at Funcorp, in contrast, were decrease in coworkers’ job search behavior de- grouped according to department within a larger creased the probability of an individual “turning organizational unit (a club). These department over” by 35 percent. Moreover, in comparing the members were not isolated from other employees effect sizes of the individual and coworker vari- in different departments. We suspect that this fac- ables, we found them to be roughly equal predictor dilutes the influence of coworkers’ job embed- tors of focal actor quitting. Thus, the job embeddedness. As such, finding that coworkers’ job em- dedness and job search behavior of coworkers beddedness influences voluntary turnover at had a sizable influence on focal actor turnover Funcorp represents a more rigorous test of our hy- decisions. potheses. Taking the results of these samples together enhances our confidence in the robustness of our inferences. DISCUSSION Extending social comparison theory to the do- Limitations and Future Research main of turnover, we investigated the role of coworkers’ attitudes and behaviors on individual We have argued that taking an average of cowork- employee turnover propensity. In two separate ers’ job embeddedness and coworkers’ job search samples, we found that aggregated coworkers’ job behaviors makes sense as a way to capture the embeddedness was a valid predictor of individ- turnover contagion stimuli to which a focal indi- ual voluntary turnover. There are a variety of vidual is exposed. However, we note that the as- ways to demonstrate validity: (1) controlling for sessment of turnover contagion was indirect. Nei- alternative explanations, (2) predicting a crite- ther the more speculative qualitative data gathered 2009 Felps, Mitchell, Hekman, Lee, Holtom, and Harman 557 from the focus groups nor the more rigorous job turnover on the six job embeddedness items to get search behaviors gathered in Study 2 measured the weights of each of the dimensions. We then what a focal person actually heard or saw cowork- multiplied each individual’s score on each dimen- ers do. Such behavioral data are difficult to gather sion by the weight for each dimension and added but would seem to be a necessary component of the six resulting products together. This created the future research on this topic. Moreover, there are weighted job embeddedness score. We then aggre- other variables that might signal how likely it is for gated these weighted individual job embeddedness contagion to occur—for instance, how close desks scores among the members of each department to are situated to each other, how often or effectively create our measure of coworkers’ job embedded- coworkers communicate with each other, friend- ness. The results of this analysis are virtually iden- ship level, and status similarity. Given that such tical to those obtained from the straight aggregation data were not available in our samples, the current approach but with slightly improved predictive va- research employed a simpler (and more conserva- lidity, which we would expect (Edwards, 2001; tive) measurement of turnover contagion. Future Howell, Breivik, & Wilcox, 2007; Law et al., 1998). research could productively build on these find- We chose not to report or base our conclusions on ings to identify alternative operationalizations of the results using these weights for three reasons. the turnover contagion process, as well as modera- First, such weights capitalize on sample-specific tors of these effects. variance and error. Second, findings reported in Further, our research has not included all vari- other studies suggest that the contributions to turn- ables known to be related to turnover. In partic- over of the dimensions vary by samples (Allen, ular, a valuable contribution to future research 2006; Lee et al., 2004, Zatzick & Iverson, 2006). would be to include more macro variables, such Thus, using weights means that the construct is as organizational support, leadership quality, essentially different with every sample, which and compensation policies, that might be fruit- makes it difficult to meaningfully compare results fully integrated as antecedent, moderator, or al- across studies (Howell et al., 2007) and thus re- ternative mechanisms for the turnover contagion duces generalizability and complicates theory model developed here. In particular, it is possible building. Third, it was more conservative (e.g., less that norms about the legitimacy of leaving might likely to capitalize on chance) and in line with develop and could affect turnover (Abelson, previous research to use the aggregate job embed- 1993). Qualitative work by Bartunek et al. (2008) dedness score. However, we should add that this and Rumery (2003) has suggested that such col- variation in weights points to the need for future lective norms can develop and that they may research into the potential moderators of the rela- affect turnover attitudes and behaviors. Unfortu- tionship between the subdimensions of job embed- nately, our data cannot speak to this issue. More- dedness and turnover or performance. Better infor- over, it should be pointed out that if such norms mation is still needed about how and under what were to exist, they would be predicated on exten- conditions job embeddedness subdimensions in- sive social comparison (Bartunek et al, 2008) and fluence turnover. Finally, though we did not use thus would act as a complementary rather than the weights in the research reported here, we rec- substitute mechanism for turnover contagion. ognize that such sample-specific information may Thus, although a normative factor could add to be what is most valuable in making prescriptions the prediction of individual turnover, we do not for any given organization. believe that effect will replace or be as strong as the contagion effect captured here. Said differ- Managerial Implications ently, we have attempted to control for the variables most likely to provide alternative explana- Organizations can use the results of this study to tions for our findings, yet we have not controlled design specific interventions aimed at reducing for all of them, nor have we included all the voluntary turnover. A primary implication is that, variables that may be involved in the process. at the group level, job embeddedness is an impor- The inclusion of these additional variables may tant antecedent to eventual turnover. Beyond just both clarify and extend the current research. affecting individual decision making, it also influ- Finally, another potential limitation concerns the ences whether the social environment incites leav- issue of weights for the subdimensions of job em- ing. Of particular interest in the context of this beddedness. As a robustness check, we ran all the research is a study by Allen (2006). He found that analyses using the weighted approach suggested by collective socialization tactics—where newcomers Law, Wong, and Mobley (1998). Specifically, we experience common learning experiences with a ran a logistic regression in which we regressed group or cohort—increase embeddedness in an or- 558 Academy of Management Journal June ganization. Such socialization tactics provide a Job search behavior may also have managerial common message about the organization, roles, and implications, and these are potentially more con- appropriate responses. This common message may troversial. Specifically, managers could prohibit shape how groups of people interpret organization- gossiping about people who are looking for other al events such as the loss of a respected coworker or jobs, especially “on company time.” Such a prohi- a large number of coworkers simultaneously. In bition might inhibit the spread of contagious infor- short, organizations could actively manage the con- mation, but bald attempts at concertive control may tent of collective socialization experiences as well provoke reactance, ill-will, and even sabotage. Per- as attend to influential individuals in social net- haps a more realistic alternative is for managers to works (Bartunek et al., 2008; Mossholder et al., track job embeddedness and turnover at the team 2005). level. Where embeddedness is low and/or turnover Second, individual-level factors that increase in- is high, they might actively try to raise embedded- dividual job embeddedness should also be consid- ness scores or reconstitute a group with some peo- ered (Mitchell et al., 2001). Prior research has iden- ple who have high embeddedness. Such changes tified a number of antecedents to on-the-job might reduce the job search behaviors demon- embeddedness. For example, personality variables strated by group members. such as conscientiousness, extraversion, and agree- ableness have demonstrated a strong, positive relationship with on-the-job embeddedness (Giosan, Concluding Thoughts Holtom, & Watson, 2005). Thus, reducing voluntary Tackling turnover theory at the meso level is not turnover through selection is one clearly actionable new; it has been advocated at the organization cul- approach (Barrick & Zimmerman, 2005). Further, ture level (Abelson, 1993) and even empirically both perceived supervisor support and perceived examined on occasion (cf. Feeley & Burnett, 1997. organizational support have been demonstrated to However, it is our belief that these approaches have positively predict levels of on-the-job embedded- not focused enough on social factors—specifically, ness (Giosan et al., 2005) and reduced voluntary the attitudes, characteristics, and behaviors of focal turnover (Maertz, Griffeth, Campbell, & Allen, employees’ coworkers. Although researchers per- 2007). Other suggestions include developing haps do not naturally think of quitting as a social schedules that fit employee needs (Holtom, Lee, & phenomenon, our research suggests that it is and Tidd, 2002), providing creative benefit alternatives that additional research regarding the social predic- or cafeteria plans, tailoring benefits to meet indi- tors of turnover is warranted. vidual needs and enhance work-life balance, allowing employees input into designing work environ- ments, and providing incentives or perks based on REFERENCES tenure (Giosan et al., 2005). Off-the-job embeddedness can also be increased Abelson, M. A. 1993. Turnover cultures. In G. Ferris & G. in a number of ways. For example, one firm was Rowland (Eds.), Research in personnel and human resource management, vol. 11: 339–376. Green- able to increase community embeddedness and wich, CT: JAI Press. subsequent retention by recruiting and hiring from communities close to their facilities and avoiding Allen, D. G. 2006. Do organizational socialization tactics relocating employees whenever possible (Holtom influence newcomer embeddedness and turnover? Journal of Management, 32: 237–256. et al., 2006). Similarly, another firm increased links in the community by supporting community ser- Bakker, A. B., & Schaufeli, W. B. 2000. Burnout contagion vice by employees (e.g., giving two days off per year processes among teachers. Journal of Applied So- for community service), allowing them to volunteer cial Psychology, 30: 2289–2308. in local student programs as mentors, and encour- Bandura, A. 1977. Social learning theory. New York: aging professional involvement in community- General Learning Press. based professional organizations (Holtom et al., Banerjee, D., & Gaston, N. 2004. Labour market signaling 2006). Finally, one organization augmented em- and job turnover revisited. Labour Economics, 11: ployee-perceived community-related sacrifice and 599–622. subsequent retention by providing home-buying as- Baron, R. M., & Kenny, D. A. 1986. The moderator-medi- sistance (Holtom et al., 2006). In sum, there are ator distinction in social psychological research: many ways that organizations can systematically Conceptual, strategic, and statistical considerations. seek to reduce the rate of voluntary, avoidable turn- Journal of Personality and Social Psychology, 51: over by enacting programs designed to increase job 1173–1182. embeddedness at the meso and micro levels. Barrick, M. R., & Zimmerman, R. D. 2005. Reducing vol- 2009 Felps, Mitchell, Hekman, Lee, Holtom, and Harman 559

untary turnover, avoidable turnover through selec- izational and market factors, Journal of Organiza- tion. Journal of Applied Psychology, 90: 159–166. tional Psychology, 5: 31–44. Barsade, S. G. 2002. The ripple effect: Emotional conta- Griffeth, R. W., & Hom, P. W. 1988. A comparison of gion and its influence on group behavior. Adminis- different conceptualizations of perceived alterna- trative Science Quarterly, 47: 644–675. tives in turnover research. Journal of Organiza- tional Behavior, 9: 103–111. Bartunek, J. M., Huang, Z., & Walsh, I. J. 2008. The development of a process model of collective turn- Griffeth, R. W., Hom, P. W., & Gaertner, S. 2000. A meta- over. Human Relations, 61: 5–38. analysis of antecedents and correlates of employee turnover: Update, moderator tests, and research im- Bradbury, H., & Lichtenstein, B. M. B. 2000. Relationality plications for the millennium. Journal of Manage- in organizational research: Exploring the space be- ment, 26: 463–488. tween. Organization Science, 11: 551–564. Guo, G., & Zhao, H. 2000. Multilevel modeling for binary Brett, J. M., & Stroh, L. K. 2003. Working 61 plus hours a data. In K. Cook & J. Hagan (Eds.), Annual review of week: Why do managers do it? Journal of Applied sociology, vol. 26: 441–462. Stanford, CA: Annual Psychology, 88: 67–78. Reviews. Chan, D. 1998. Functional relations among constructs in Hatch, N. W., & Dyer, J. H. 2004. Human capital and the same content domain at different levels of anal- learning as a source of sustainable competitive ad- ysis: A typology of composition models. Journal of vantage. Strategic Management Journal, 25: 1155– Applied Psychology, 83: 234–246. 1178. Chartrand, T. L., & Bargh, J. A. J. 1999. The chameleon Haveman, H. A. 1995. The demographic metabolism of effect: The perception-behavior link and social inter- organizations: Industry dynamics, turnover, and ten- action. Journal of Personality and Social Psychol- ure distributions. Administrative Science Quar- ogy, 76: 893–910. terly, 40: 586–618. Crossley, C. D., Bennett, R. J., Jex, S. M., & Burnfield, J. L. Holtom, B. C., Lee, T. W., & Tidd, S. 2002. The relation- 2007. Development of a global measure of job em- ship between work status congruence and work-re- beddedness and integration into a traditional model lated attitudes and behaviors. Journal of Applied of voluntary turnover. Journal of Applied Psychol- Psychology, 87: 903–915. ogy, 92: 1031–1042. Holtom, B. C., Mitchell, T. R., & Lee, T. W. 2006. Increas- Degoey, P. 2000. Contagious justice: Exploring the social ing human and social capital by applying job embed- construction of justice in organizations. In B. M. dedness theory. Organizational Dynamics, 35(4): Staw & R. I. Sutton (Eds.), Research in organization- 316–331. al behavior, vol. 22: 51–102. Greenwich, CT: JAI Holtom, B. C., Mitchell, T. R., Lee, T., & Tidd, S. 2006. Press. Less is more: Validation of a short form of the job Diamantopoulos, A., & Winklhofer, H. M. 2001. Index embeddedness measure and theoretical exten- construction with formative indicators: An alterna- sions. Paper presented at the annual meeting of the tive to scale development. Journal of Marketing Academy of Management, Atlanta. Research, 38: 269–277. Holtom, B. C., & O’Neill, B. S. 2004. Job embeddedness: A Eder, P., & Eisenberger, R. 2008. Perceived organizational theoretical foundation for developing a comprehen- support: Reducing the negative influence of co- sive nurse retention plan. Journal of Nursing Ad- worker withdrawal behavior. Journal of Manage- ministration, 34: 216–227. ment, 34(1): 55–68. House, R. J., Rousseau, D. M., & Thomas-Hunt, M. 1995. Edwards, J. R. 2001. Multidimensional constructs in The meso-paradigm: A framework for the integration organizational behavior research: An integrative of micro and macro organizational behavior. In B. M. analytical framework. Organizational Research Staw & L. L. Cummings (Eds.), Research in organi- Methods, 4: 144–192. zational behavior, vol. 17: 71–114. Greenwich, CT: JAI Press. Eisenhardt, K. M., & Graebner, M. E. 2007. Theory build- Howell, R. D., Breivik, E., & Wilcox, J. B. 2007. Recon- ing from cases: Opportunities and challenges. Acad- sidering formative measurement. Psychological emy of Management Journal, 50: 25–32. Methods, 12: 205–218. Feeley, T. J., & Barnett, G. A. 1997. Predicting employee Johns, G. 2006. The essential impact of context on organ- turnover from communication networks. Human izational behavior. Academy of Management Re- Communication Research, 23: 370–387. view, 31: 386–408. Festinger, L. 1954. A theory of social comparison pro- Kammeyer-Mueller, J. D., Wanberg, C. R., Glomb, T. M., & cesses. Human Relations, 7: 117–140. Ahlburg, D. 2005. The role of temporal shifts in Giosan, C., Holtom, B., & Watson, M. 2005. Antecedents turnover processes: It’s about time. Journal of Ap- to job embeddedness: The role of individual, organ- plied Psychology, 90: 644–658. 560 Academy of Management Journal June

Klein, K. J., & Kozlowski, S. W. J. 2000. From micro to of voluntary turnover and embeddedness: Founda- meso: Critical steps in conceptualizing and conduct- tions for a comprehensive theory of attachment. In ing multilevel research. Organizational Research B. M. Staw & R. I. Sutton (Eds.), Research in organ- Methods, 3: 211–236. izational behavior, vol. 23: 189–246. Greenwich, Kopelman, R. E., Rovenpor, J. L., & Millsap, R. E. 1992. CT: JAI Press. Rationale and construct validity evidence for the Job Mobley, W. H. 1977. Intermediate linkages in the rela- Search Behavior Index: Because intentions and New tionship between job satisfaction and employee Year’s resolutions often come to naught. Journal of turnover. Journal of Applied Psychology, 62: 237– Vocational Behavior, 40: 269–287. 240. Krackhardt, D., & Porter, L. W. 1985. When friends leave: Mossholder, K. W., Settoon, R. P., & Henagan, S. C. 2005. A structural analysis of the relationship between A relational perspective on turnover: Examining turnover and stayers’ attitudes. Administrative Sci- structural, attitudinal, and behavioral predictors. ence Quarterly, 30: 242–261. Academy of Management Journal, 48: 607–618. Krackhardt, D., & Porter, L. W. 1986. The snowball effect: Pfeffer, J. 1991. Organization theory and structural per- Turnover embedded in communication networks. spectives on management. Journal of Management, Journal of Applied Psychology, 71: 50–55. 17: 789–803. Kruglanski, A. W., & Mayseless, O. 1990. Classic and Price, J. L. 1977. The study of turnover. Ames: Iowa State current social comparison research: Expanding the University Press. perspective. Psychological Bulletin, 108: 195–208. Price, J. L., & Mueller, C. W. 1986. Absenteeism and Kulik, C. T., & Ambrose, M. L. 1992. Personal and situa- turnover of hospital employees. Greenwich, CT: JAI tional determinants of referent choice. Academy of Press. Management Review, 41: 55–67. Rousseau, D. M. 1985. Issues of level in organizational Law, K. S., Wong, C., & Mobley, W. H. 1998. Toward a research. In L. L. Cummings & B. M. Staw (Eds.), taxonomy of multidimensional constructs. Acad- Research in organizational behavior, vol. 7: 1–37, emy of Management Review, 23: 741–755. Greenwich, CT: JAI Press. Lee, T. W. 1999. Using qualitative methods in organi- Rumery, S. M. 2003. The influence of work group cul- zational research. Thousand Oaks, CA: Sage. ture on employee turnover. Unpublished doctoral Lee, T. W., Mitchell, T. R., Sablynski, C. J., Burton, J. P., dissertation, University of Connecticut, Storrs. & Holtom, B. C. 2004. The effects of job embedded- Salancik, G. R., & Pfeffer, J. 1978. A social information ness on organizational citizenship, job performance, processing approach to job attitudes and task de- volitional absences, and voluntary turnover. Acad- sign. Administrative Science Quarterly, 23: 224– emy of Management Journal, 47: 711–722. 252. Maertz, C. P., & Campion, M. A. 1998. 25 years of volun- Shaw, J. D., Gupta, N., & Delery, J. E. 2005. Alternative tary turnover research: A review and critique. Inter- conceptualizations of the relationship between national Review of Industrial and Organizational voluntary turnover and organizational perfor- Psychology, 13: 49–81. mance. Academy of Management Journal, 48: Maertz, C. P., Griffeth, R. W., Campbell, N., & Allen, D. G. 50–68. 2007. The effects of perceived organizational sup- Spector, P. E. 1985. Measurement of human service job port and perceived supervisor support on employee satisfaction: Development of the Job Satisfaction turnover. Journal of Organizational Behavior, 28: Survey. American Journal of Community Psychol- 1059–1075. ogy, 13: 693–712. March, J. G., & Simon, H. A. 1958. Organizations. New Steel, R. P. 2002. Turnover theory at the empirical inter- York: Wiley. face: Problems of fit and function. Academy of Man- Mathieu, J., & Kohler, S. 1990. A cross-level examination agement Review, 27: 346–360. of group absence influences on individual absence. Tesser, A., Campbell, J., & Mickler, S. 1983. The role of Journal of Applied Psychology, 75: 217–220. social pressure, attention to the stimulus, and self- Meyer, J. P., Allen, N. J., & Smith, C. A. 1993. Commit- doubt in conformity. European Journal of Social ment to organizations and occupations: Extension Psychology, 13: 217–223. and test of a three-component conceptualization. Ulrich, D., & Smallwood, N. 2006. How leaders build Journal of Applied Psychology, 78: 538–551. value: Using people, organization, and other in- Mitchell, T. R., Holtom, B. C., Lee, T. W., Sablynski, C. J., tangibles to get bottom-line results. Englewood & Erez, M. 2001. Why people stay: Using job embed- Cliffs, NJ: Wiley. dedness to predict voluntary turnover. Academy of Van Dijk, P., & Kirk-Brown, A. 2003. The relationship Management Journal, 44: 1102–1121. between job embeddedness and the intention to Mitchell, T. R., & Lee, T. W. 2001. The unfolding model leave an organization. Paper presented at the APS 2009 Felps, Mitchell, Hekman, Lee, Holtom, and Harman 561

Psychology of Relationships Interest Group, 3rd An- nual Conference, Deakin University, Melbourne. Wooten, D. B., & Reed, A. 1998. Informational influence Will Felps ([email protected]) is an assistant professor at the and the ambiguity of product experience: Order ef- Rotterdam School of Management, Erasmus University. fects on the weighting of evidence. Journal of Con- He received his Ph.D. in management from the Univer- sumer Research, 7(1): 79–99. sity of Washington, Seattle. His diverse topics of interests Zatzick, C. D., & Iverson, R. D. 2006. Job embeddedness, are linked together by a desire to create more effective gender, and voluntary turnover: Further exten- and humane organizations through more valid and co- sions of a new construct. Paper presented at the herent theories. annual meeting of the Academy of Management, At- Terence Mitchell ([email protected]) is a professor lanta. of management and organization and of psychology at the Foster School of Business at the University of Wash- ington, Seattle. He received his Ph.D. from the University APPENDIX of Illinois. His current research interests are organization- Job Embeddedness Scale, Short Forma al attachment, motivation, and leadership.

1. My job utilizes my skills and talents well. David R. Hekman ([email protected]) is an as- 2. I feel like I am a good match for my organization. sistant professor of health care management at the Lubar 3. If I stay with my organization, I will be able to achieve School of Business, University of Wisconsin–Milwaukee. most of my goals. He earned his Ph.D. in management from the University 4. I really love the place where I live. of Washington, Seattle. His research interests include 5. The place where I live is a good match for me. employee attachment to the organization, courageous 6. The area where I live offers the leisure activities that I like employee actions, subtle discrimination, and under- (sports, outdoor activities, cultural events & arts). standing why organizations fail. 7. I have a lot of freedom on this job to pursue my goals. 8. I would sacrifice a lot if I left this job. Thomas W. Lee ([email protected]) is the Hughes 9. I believe the prospects for continuing employment with M. Blake Professor of Management and associate dean for my organization are excellent. academic and faculty affairs at the Foster School of Busi- 10. Leaving the community where I live would be very hard. ness, University of Washington, Seattle. He earned his 11. If I were to leave the community, I would miss my non- Ph.D. in management from the University of Oregon. His work friends. primary research interests include employee loyalty, re- 12. If I were to leave the area where I live, I would miss my neighborhood. tention and turnover, and work motivation. 13. I am a member of an effective work group. Brooks C. Holtom ([email protected]) is an assistant profes- 14. I work closely with my coworkers. sor of management in the McDonough School of Business 15. On the job, I interact frequently with my work group at Georgetown University. He received his Ph.D. in man- members. 16. My family roots are in this community. agement from the University of Washington, Seattle. His 17. I am active in one or more community organizations (e.g., research interests include the attraction, development, churches, sports teams, schools, etc.). and retention of human and social capital. 18. I participate in cultural and recreational activities in my Wendy S. Harman ([email protected]) is vis- local area. 19. Are you currently married? iting assistant professor of business administration at Cen- 20. If you are currently married, does your spouse work tral Washington University. She received her Ph.D. in man- outside the home? agement from the University of Washington Business 21. Do you own a home (with or without a mortgage)? School. Her primary research interest focuses on using various methods of inquiry to positively impact work life. a Source: Holtom et al., 2006.