Received: 13 February 2015 Revised: 30 January 2017 Accepted: 31 January 2017
Surveying the forest: A meta-analysis, moderator
investigation, and future-oriented discussion of
the antecedents of voluntary employee turnover
Alex L. Rubenstein1Marion B. Eberly2Thomas W. Lee2
1University of Memphis
2University of Washington
Alex L. Rubenstein, Fogelman College of Business
and Economics, University of Memphis, 323 FCB
Building, Memphis, TN 38152, USA.
Recent narrative reviews (e.g.,Hom, Mitchell, Lee, and Griffeth; Hom,
Lee, Shaw, and Hausknecht) advise that it is timely to assess the
progress made in research on voluntary employee turnover so as
to guide future work. To provide this assessment, we employed a
three-step approach. First, we conducted a comprehensive meta-
analysis of turnover predictors, updating existing effect sizes and
examining multiple new antecedents. Second, guided by theory, we
developed and tested a set of substantive moderators, considering
factors that might exacerbate or mitigate zero-order meta-analytic
effects. Third, we examined the holistic pattern of results in order
to highlight the most pressing needs for future turnover research.
The results of Step 1 reveal multiple newer predictors and updated
effect sizes of more traditional predictors, which have received sub-
stantially greater study. The results of Step 2 provide insight into the
context-dependent nature of many antecedent–turnover relation-
ships. In Step 3, our discussion takes a bird’s-eye view of the turnover
“forest” and considers the theoretical and practical implications of
the results. We offer several research recommendations that break
from the traditional turnover paradigm, as a means of guiding future
The question of why employees voluntarily leave their jobs has captivated researchers for 100 years (Hom, Lee, Shaw,
& Hausknecht, 2017). Given its impact on organizations’ functioning and survival, it is no surprise that research aimed
at understanding employee turnover remains an important topic for academics and practitioners (Holtom, Mitchell,
Lee, & Eberly, 2008; Hom, Mitchell, Lee, & Griffeth, 2012). Turnover is costly: Recent data show that organizations
incur costs often upwards of 200% of an employee’s annual pay to recruit, select, and train successors (Allen, Bryant,
& Vardaman, 2010). Less tangible costs are also noteworthy, including loss of tacit knowledge and social capital (Dess
& Shaw, 2001), reduced customer satisfaction (McElroy, Morrow, & Rude, 2001), or turnover contagion (Felps et al.,
Personnel Psychology.2017;1–43. wileyonlinelibrary.com/journal/peps c
2017 Wiley Periodicals, Inc. 1
2RUBENSTEIN ET AL.
Nearly 2 decades have passed since Griffeth, Hom, and Gaertner (2000) conducted the broadest meta-analysis of
the turnover literature. Since then, narrower meta-analyses have sharpened our empirical understanding of turnover
processes via more isolated snapshots of select precursors or outcomes (e.g., Berry, Lelchook, & Clark, 2012; Hancock,
Allen, Bosco, McDaniel, & Pierce, 2013; Jiang, Liu, McKay, Lee, & Mitchell, 2012), thereby offering some clarity in pre-
dicting what has proved an elusive behavior. Since the turn of the millennium, however, a sizable number of primary
studies have been published, also seeking to understand this phenomenon. Some effect sizes are out of date, whereas
other antecedent–turnover relationships have yet to be meta-analyzed. Moreover, given the large amount of hetero-
geneity existing around many effect size estimates, it is also worthwhile to scrutinize moderators that would account
for this variability.
The purpose of this paper is to present an updated and holistic picture of how oft-studied constructs oper-
ate within the turnover literature. Our focus is on individual voluntary turnover, deﬁned as “voluntary cessation
of membership in an organization, by an individual who receives monetary compensation for participation in that
organization” (Hom & Griffeth, 1995, p. 5).1With the abundance of theoretical and empirical work advanced in
recent years, it is prudent to assess the progress that has been made in describing, explaining, and predicting
In a recent narrative review of the literature, Holtom et al. (2008) tracked the nomological network development of
turnover research since March and Simon (1958) introduced the ﬁrst formal turnover theory, focusing mainly on the
desire to leave (i.e., job satisfaction) and the ease of leaving (i.e., job alternatives) as the primary reasons why people
quit. Almost 60 years have passed since this seminal theoretical work—and a century since the ﬁrst empirical (mostly
atheoretical) turnover articles surfaced (Hom et al., 2017)—such that the number of turnover predictors has dramati-
cally increased in this time. Based on an extensive review of the literature, Holtom et al. (2008) acknowledged a total
of 50 broad antecedents as having scientiﬁc value, of which some (like personality and role states) subsume even more
variables. This sheer number of antecedents highlights the myriad perspectives by which researchers have studied
turnover but also raises questions regarding which constructs have made the greatest impact. As such, it is timely, if
not necessary, to assess what progress has been made since the ﬁrst empirical tests of March and Simon emerged in
Our holistic effort begins with an updated meta-analytic empirical assessment of turnover research to assess main
effect relationships (Step 1). Since Griffeth et al. (2000), a bevy of new constructs have entered into the academic ver-
nacular, whereas other constructs have been studied in considerably more depth, perhaps warranting a revision of
earlier estimates. As a point of illustration, whereas the Griffeth et al. analysis examined 45 predictors and 843 effect
sizes, we include 57 predictors across 1,800 effect sizes (a 27% increase in constructs and a 114% increase in effects).
We provide insight into new and inﬂuential predictors such as engagement, justice, and job characteristics, as well as
examining potential changes in effect sizes compared to previous work.
Along with this initial meta-analysis, we developed and tested a set of substantive, a priori moderation hypotheses,
grounded in theory, to account for variability in predictor–turnover effects (Step 2). Due to the context-driven nature
of the turnover process, this section highlights those higher level factors (at the sample or economy level) that exacer-
bate or mitigate antecedent–turnover effects. Whereas some antecedents might not be strong predictors of turnover
in a zero-order sense, it is possible that they become more or less impactful after accounting for the context in which
they are embedded.
Third, because so many predictors have been tested in this literature, it is timely to highlight the major develop-
ments visible in our results and to draw conclusions for future work. We integrate the results of our ﬁrst two steps
to consider trends such as what variables appear to be predictive across a variety of contexts, what variables are
not predictive across contexts, what variables appear highly context driven, and what variables still require more
research. Humphrey (2011) asserted that every meta-analysis should answer the question, “Where are we going?” by
providing a substantive analysis of one’s results and what they mean. We concur, and conclude with a constructive
discussion with eyes looking toward the most promising constructs and methods, along with speciﬁc recommenda-
tions to guide future turnover study and practice. In short, we seek parsimony and improved focus for this very large
RUBENSTEIN ET AL.3
1STEP 1: INITIAL META-ANALYSIS
To begin our systematic exploration, we ﬁrst conducted a zero-order meta-analysis to determine correlational relation-
ships of multiple antecedents with turnover.For reasons of completeness, we coded all turnover antecedents examined
in previous studies.
1.1.1 Literature search and inclusion criteria
We ﬁrst identiﬁed published empirical articles that examined any turnover antecedents. The articles were identiﬁed
through an online search of the PsycINFO, EBSCO, JSTOR, and Google Scholar databases, as well as searching confer-
ence proceedings from the Academy of Management and Society for Industrial and Organizational Psychology annual
meetings. We restricted our search to the time frame of 1975 through July 2016, because it was in the mid-1970s
when the ﬁrst theory-based, empirical studies on turnover emerged (Hom et al., 2017). We reviewed the abstracts
from this initial search and eliminated studies of a theoretical nature (e.g., literature reviews), studies that only focused
on employee involuntary turnover, studies that treated employee voluntary turnover as an independent variable, pre-
vious meta-analyses, and studies that treated “turnover” as withdrawal cognitions or attitudes (i.e., turnoverintentions
as the criterion). To be included in the meta-analysis, the study had to report sufﬁcient data to calculate an effect size,
had to use a design such that it measured predictors before turnover took place, and had to measure individual turnover
behavior. In cases where there were not sufﬁcient results reported to calculate an effect size, we reached out to one of
the study authors to request information. These criteria led to a ﬁnal primary study population of 316 articles.
1.1.2 Coding procedures
In a ﬁrst step, the ﬁrst two authors extracted all information according to the coding scheme from the articles. Each
author coded approximately half of the total set of manuscripts. In a second step, we grouped the various independent
variables according to their conceptual overlap. In conducting this meta-analysis, we attempted to be as comprehen-
sive as possible in capturing all relevant and empirically investigated antecedents. At the same time, we recognized the
need to reduce the abundance of data to a manageable and interpretable level. For example,although there is undoubt-
edly much to be learned by examining the relationships among dimensions of commitment (i.e., affective, normative,
and continuance) or ﬁt (i.e., person-organization, person-job, person-vocation), we opted for parsimony over speci-
ﬁcity and combined dimensions to represent overarching constructs. For instance, leadership was aggregated using its
most common positive indicators: measures of leader–member exchange, consideration, and transformational leader
style. Further, we do not report effects for any variables for which there were k≤3 studies (this included 14 variables,
such as shocks, emotional labor strategies, and subjective norms). Throughout the coding process, we engaged in an
iterative process and continued discussion to identify and resolve other trade-offs between parsimony and speciﬁcity.
This resulted in a ﬁnal set of 57 predictors. In order to verify agreement in coding, the ﬁrst two authors independently
coded a random sample of 10% of the others’ articles. We found minimal differences, resolved by checking the original
manuscripts and then collaboratively discussing disagreements.
1.1.3 Meta-analytic procedures
Following Hunter and Schmidt (2004), we used psychometric procedures for a random effects meta-analytic model, in
which we corrected for attenuation due to unreliability as well as for sampling error by weighting each study’s effect
size by its sample size. Random effects models are preferred over ﬁxed effect models when the relationships tested
in primary studies are presumed to be heterogeneous across studies, which is typically the case in turnover studies
(Borenstein, Hedges, Higgins, & Rothstein, 2010). We computed composite correlations for those studies that included
multiple measures of the same construct. Studies that included multiple independent samples were coded separately.
We used Cronbach’s 𝛼to correct for measurement error in the predictors. These values were provided in the majority
4RUBENSTEIN ET AL.
of the studies. However, when studies did not report this reliability coefﬁcient, we used means across other studies.
Reliabilities for objective variables, such as for employee age, pay, and turnover, were set to 1.00. With additional cor-
rections for artifactual variance due to sampling error, we then performed random-effects meta-analyses based on the
average sample-size weighted correlations and corrected correlations. We also report Cochran’s Qstatistic and the
I2index. The former statistic is an absolute measure of the absence or presence of signiﬁcant heterogeneity, whereas
the latter statistic considers the percentage of variance attributable to heterogeneity in a predictor–turnover rela-
tionship, which, as a percentage, is advantageous for comparing across meta-analyses of different study sizes (Higgins,
Thompson, Deeks, & Altman, 2003). Values of I2between 0–40% represent minor heterogeneity, 40–70% medium
heterogeneity, and 70–100% substantial heterogeneity (Borenstein, Hedges, Higgins, & Rothstein, 2009).
Following previous research, relationships involving turnover behavior were corrected based on the nature of a
point-biserial correlation including a dichotomous criterion. Such correlations may be downwardly biased if they vio-
late assumptions of bivariate normality due to an uneven split of stayers and leavers (Kemery, Dunlap, & Griffeth,
1988). Hunter and Schmidt (2004) provide a correction formula, which requires coding for the mean turnover base
rate across samples in the meta-analysis. For this procedure, where turnover rate data were unavailable (∼4% of stud-
ies), we used the average rateacross studies in the most closely related industry (using standard industrial classiﬁcation
codes) from which the sample was obtained.
Across the primary studies coded, the mean sample size across studies was 1,053, mean base rate of turnover was
23.5% (SD =14.6%), mean time lag between predictor and turnover measurement was roughly 16 months (mode =
12 months), mean response rate was 67.9% (SD =23.6%), grand mean employee age was 35.3 (SD =6.7), grand mean
sex distribution was 53.3% male (SD =24.0%), and grand mean organizational tenure was 6.5 years employed (SD =
4.0). Among countries represented, a large majority (78.9%) came from the United States. Among occupations, 25.9%
were involved with some sort of hospital work, nursing, or healthcare services; 12.6% were involved with banking or
other ﬁnancial services; 12.3% performed what would be considered “blue collar” work (e.g., manufacturing, food pro-
cessing); 9.1% were military; the remaining studies represented smaller occupational percentages.
1.2 Results and discussion
To facilitate interpretation of the results, we organized the 57 predictors by labels based on Holtom et al. (2008). Def-
initions and sample studies for each predictor are provided in Table 1, whereas Table 2 shows the results for each pre-
dictor. Although readers can examine the complete results in Table 2, there are two aspects of the results on which we
focus here: (a) what effects have signiﬁcantly changed compared to what Griffeth et al. (2000) reported and (b) what
predictors are new and noteworthy, given increased empirical study in recent years.
To compare our effect sizes with those reported by Griffeth et al. (2000) and determine which antecedent–turnover
relationships have signiﬁcantly changed, we could not perform independent samples comparisons because the corre-
lations we obtained were dependent with theirs (i.e., all antecedent–turnover samples we coded were partially redun-
dant with theirs). Their study also did not report conﬁdence intervals, so we were also unable to directly compare
nonoverlapping estimates. If samples are independent and conﬁdence intervals are available, these approaches would
be recommended to test for signiﬁcant effect differences (Steiger, 1980; Zou, 2007). To circumvent this issue, we used
three alternative approaches, which we list in order of greater to lesser objectivity. Ultimately, no approach is per-
fect, so we suggest a combination be used based on data availability rather than relying on any single method alone:
First, we reanalyzed all correlations in our data using only estimates from 2000 to 2016 (i.e., after the Griffeth et al.
article) and examined whether those values—which are independent—were signiﬁcantly different. To compare esti-
mates, we followed the modiﬁed asymptotic method of correlational differences, outlined in Zou (2007, equation 15).
This approach invokes Fisher’s r-to-ztransformation to compute a conﬁdence interval for the difference in correla-
tions. To test for signiﬁcance, we used the mean study sample size, excluding outliers (N=1,053). Second, we compared
whether the Griffeth et al. estimates were different than those of this study by comparing 95% conﬁdence intervals of
their effects to our own. This is an approach widely used in meta-analyses to compare effect size differences between
or among groups/subgroups. However, since Griffeth et al. did not report standard errors, we input those obtained for
RUBENSTEIN ET AL.5
TAB LE 1 Deﬁnitions of turnover antecedents
Variable name Deﬁnition/Clariﬁcation Sample studies
(1) Individual attributes
Abilities and skills Proﬁciency in doing something, whether learned or innate. In this context, it refers to abilities or skills
that relevant to the job (e.g., cognitive ability, managerial skills).
Kraimer, Seibert, Wayne,
Liden, and Bravo (2011);
Age Chronological number of years an individual has lived. Stumpf and Dawley (1981)
Agreeableness Personality trait capturing the extent to which individuals are courteous, trusting, cooperative, and
empathic. Part of the ﬁve-factor model of personality (the Big Five, McCrae & Costa, 1987).
Barrick and Mount (1996);
Children Number of children an individual reports having, primarily,the number of children living at home
Blegen, Mueller, and Price
(1988); Lee and Maurer
Conscientiousness Personality trait capturing the extent to which individuals are dependable, organized, persistent, and
achievement-oriented. Part of the ﬁve-factor model of personality (the Big Five, McCrae & Costa,
Barrick and Zimmerman
(2009); Timmerman (2006)
Education An individual’s maximum level of education attained. Bretz, Boudreau, and Judge
Emotional stability Personality trait capturing the extent to which individuals are calm, secure, and not moody; part of the
ﬁve-factor model of personality (the Big Five, McCrae & Costa, 1987).
Barrick and Zimmerman
Boswell, Roehling, and
Ethnicity/race An individual’s identiﬁcation as being 0 =White or 1 =non-White. Often, studies do not disclose how
this information is assessed. It is possible our dataset includes studies that inquired about biological
distinctions (race), whereas others inquired about cultural distinctions (ethnicity).
Elvira and Cohen (2001);
Zatzick, Elvira, and Cohen
Extraversion Personality trait capturing the extent to which individuals are sociable, talkative, active, and ambitious
(Barrick & Mount, 1996); part of the ﬁve-factor model of personality (the Big Five,McCrae&Costa,
Roehling, and Boudreau
(2000); Liao and Joshi
Internal motivation Motivation based on internal factors such as enjoyment of the task, a sense of conﬁdence, or a
normative belief in the importance of work in general.
Iverson (1999); Mirvis and
Locus of control (higher =external) Personality trait capturing the extent to which an individual attributes the cause of personal
events—whether successes or failures—either to the self (internal locus of control) or to the
environment (external locus of control).
Allen, Weeks, and Mofﬁtt
2005; Blau (1987)
Marital status Whether an individual is currently married (0 =nonmarried, 1 =married). Hom and Hulin (1981)
6RUBENSTEIN ET AL.
TAB LE 1 (Continued)
Variable name Deﬁnition/Clariﬁcation Sample studies
Openness to experience Personality trait capturing the extent to which individuals are imaginative, cultured, broad-minded, and
ﬂexible (Barrick & Mount, 1996); part of the ﬁve-factor model of personality (the Big Five,McCrae&
Barrick and Mount (1996);
Sex Biologically based categories of 0 =female and 1 =male. Lyness and Judiesch (2001)
Tenu re Time employed with one’s current organization (measured in years). Benson, Finegold, and
(2) Aspects of the job
Instrumental communication Degree to which organizations formally transmit information to employees about their job. Blegen, Mueller, and Price
(1988); Price and Mueller
Job characteristics (VISAF) Composite including ﬁve core features of Hackman and Oldham’s (1976) job characteristics model:
variety (degree to which the job offers different proﬁciencies), task identity (degree to which the job
provides tasks with visible outcomes), task signiﬁcance (degree to which the job is important for other
people), autonomy (degree to which the job provides discretion in how to do the work), and feedback
(degree to which the job provides speciﬁc and actionable information about one’s
Colarelli, Dean, and Konstans
(1987); Schmidt and
Job security Degree to which one is conﬁdent about having stable present or future employment in their current job. Arnold and Feldman (1982);
Iverson and Pullman
Participation Degree to which an individual can exercise power, has opportunity for input, and can make decisions on
the job (Price & Mueller, 1981).
Allen, Shore, and Griffeth
(2003); Price and Mueller
Pay Amount of money an individual receives for the job. We include all basic monetary compensation such
as base salary and commission.
Motowidlo (1983); Tekleab,
Bartol, and Lui (2005)
Role ambiguity Degree to which role expectations are unclear. Spector (1991)
Role conﬂict Degree to which employee role expectations are conﬂicting or incompatible. Iverson (1999); Spector
Routinization Degree to which the job is repetitive and is subject to rules and regulations (Price & Mueller, 1981). Martin (1980); Parasuraman
and Alutto (1984)
Task complexity Degree to which the job is intricate and has many cognitive requirements. Parasuraman and Alutto
RUBENSTEIN ET AL.7
TAB LE 1 (Continued)
Variable name Deﬁnition/Clariﬁcation Sample studies
Workload Amount of work required by the individual; includes mostly measures of hours workedor how hard and
fast an individual works.
Huffman, Adler, Dolan, and
Castro (2005); Spector
(3) Traditionaljob attitudes
Job involvement Degree to which an individual identiﬁes with his or her job. Hollenbeck and Williams
Job satisfaction Degree to which an individual likes his or her job, measured globally or via the job’s different facets (e.g.,
pay, the work itself, peers, etc.).
Boswell, Boudreau, and Tichy
(2005); Judge (1993)
Organizational commitment Degree to which an individual experiences loyalty to the organization and desires to stay. Includes all
three facets of organizational commitment: affective (a desire to stay), normative (an obligation to
stay), and continuance (a need to stay) commitment (Meyer & Allen, 1991)
Lee, Ashford, Walsh, and
Other commitment Degree to which an individual experiences loyalty to targets other than the organization, such as his or
her occupation or career.
Blau (1989); Cohen (2000)
Other satisfaction Degree to which an individual likes other aspects relevant to employment, such as his or her career or
life in general.
Boswell, Boudreau, and
(4) Newer personal conditions
Coping An individual’s abilities to manage internal and external demands that are perceived as exceeding
Riolli and Savicki (2006);
Wright and Bonett (1993)
Engagement Degree to which an individual invests their entire self into their work role; their dedication, vigor,and
devotion toward work.
Owens, Johnson, and
Mitchell (2010); Spell, Eby,
and Vandenberg (2014)
Stress/exhaustion Various well-being-related variables such as strain (an aversive psychological reaction to stressful
work), burnout (a state where emotional resources are depleted and can no longer be recovered),
tension (being bothered by stressful work incidents), and emotional exhaustion (the experience of
being emotionally depleted).
de Croon, Sluiter, Blonk,
Riolli and Savicki (2006);
Sheridan and Vredenburgh
(5) Organizational context
Centralization Degree to which power is concentrated and held by few individuals in an organization at higher
Bluedorn (1982); Martin
8RUBENSTEIN ET AL.
TAB LE 1 (Continued)
Variable name Deﬁnition/Clariﬁcation Sample studies
The shared experiences, perceptions, and behavioral tendencies among a group of employees. A higher
score denotes a more positive climate.
Spreitzer and Mishra (2002);
Suszko and Breaugh
Organization prestige Degree to which an organization is well regarded or is perceived as being high-status (Mael & Ashforth,
Joseph, Ang, and Slaughter
(2015); Ramesh and
Organization size Number of employees of a focal organization. Elvira and Cohen (2001)
Organization support Degree to which an individual believes that the organization values their contributions and cares about
their well-being (Eisenberger, Huntington, Hutchison, & Sowa, 1986).
Allen, Shore, and Griffeth
Eisenberger, and Armeli
Reward contingency Degree to which compensation is based on an individual’s performance. Allen and Griffeth (2001);
Rewards offered Rewards provided to employees beyond pay. Includes beneﬁts, career/growth opportunities, and
Kraimer, Seibert, Wayne,
Liden, and Bravo (2011)
(6) Person–context interface
Fit “Compatibility between an individual and a work environment that occurs when their characteristics
are well matched” (Kristof-Brown, Zimmerman, & Johnson, 2005, p. 281). Includes various facets, but
most prominently coded as person-organization ﬁt and person-job ﬁt.
Cable and DeRue (2002);
Ramesh and Gelfand
Inﬂuence The effect an individual can have on another person, for example, by persuasion or informal leadership. Buchko (1992); Clausen and
Job embeddedness A broad constellation of inﬂuences on why an individual stays within their job. An individual’s
“stuckness” within a larger social system as a function of external forces within the organization
(on-the-job) and the community (off-the-job). Encompasses links (connectionsto other people and
activities), ﬁt (environmental compatible with the individual’s values and needs), and sacriﬁce (what
an individual would give up by quitting).
Harris, Wheeler, and Kacmar
(2011); Mitchell, Holtom,
Lee, Sablynski, and Erez
Justice Experience of fairness within one’s work. Includes all facets of justice: distributive (fairness of
outcomes), procedural (fairness of processes used to arrive at outcomes), informational (fairness of
information provided), and interpersonal (fairness of interpersonal treatment received) justice
(Colquitt, Conlon, Wesson, Porter, & Ng, 2001).
Erdogan and Bauer (2010);
Spreitzer and Mishra
RUBENSTEIN ET AL.9
TAB LE 1 (Continued)
Variable name Deﬁnition/Clariﬁcation Sample studies
Leadership “A process whereby intentional inﬂuence is exerted by one person over other people to guide, structure,
facilitate activities and relationships in a group or organization” (Yukl, 2002). Primarily measured with
leadership styles (e.g., transformational leadership) and quality of leader–follower relationship (e.g.,
leader-member exchange). A higher score denotes more positive leadership.
Graen, Liden, and Hoel
(1982); Tse, Huang, and
Met expectations Degree to which what an individual encounters at work is in line with the expectations they had for such
experiences (Porter & Steers, 1973).
Hom, Griffeth, Palich, and
Peer/group relations Degree to which an individual reports positive interpersonal experiences with coworkers or group
members. Most often measured by coworker support but also with feelings of cohesion or social
Price and Mueller 1981;
Psychological contract breach An individual’s cognition that the employer has failed to meet one or more obligations of the
employment relationship (Morrison & Robinson, 1997).
and Aselage (2016)
Work–life conﬂict Degree to which one’s work role interferes with nonwork roles, and vice versa. Hom and Kinicki (2001);
Huffman, Casper, and
(7) External job market
Alternatives Availability of employment opportunities. Includes subjective (e.g.,perceived alternatives) and objective
(e.g., unemployment rate) measures.
Boswell, Boudreau, and
Dunford 2004; Trevor
10 RUBENSTEIN ET AL.
TAB LE 1 (Continued)
Variable name Deﬁnition/Clariﬁcation Sample studies
(8) Attitudinal withdrawal
Withdrawal cognitions Thoughts about leaving or related withdrawal attitudes. Encompasses turnover intentions, thoughts of
quitting, search intentions, and expected utility of withdrawal (Hom, Caranikas-Walker, Prussia, &
Allen, Weeks, and Mofﬁtt
2005; Griffeth, Steel, Allen,
and Bryan (2005)
(9) Employee behaviors
Selection process performance How an individual has performed on a given task during the selection process (i.e., on prehire
performance tasks or an interview).
Hausknecht, Trevor, and Farr
(2002); Kinicki, Lockwood,
Hom, and Griffeth (1990)
Job search Behavioralmethods to general alternative employment opportunities. Bretz, Boudreau, and Judge
(1994); Linnehan and Blau
Absenteeism Missing from work entirely when one is expected to be in attendance. McElroy, Morrow, and
Fenton (1995); Rosse 1988
Lateness Arriving to work at a time beyond expected by the employer. Blau (1994); Clegg (1983)
Employee performance Activitiesthat contribute to the technical core of the organization, often prescribed by an employee’s
Organizational citizenship behaviors
Individual discretionary actions that are not explicitly recognized by an organization’s reward system,
but that promote organizational effectiveness in the broader psychological and social environment.
Chen, Hui, and Sego (1998);
Lee, Mitchell, Sablynski,
Burton, and Holtom (2004)
RUBENSTEIN ET AL.11
TAB LE 2 Meta-analytic predictors of voluntary turnover
Variable name kN
interval (CV) [LB, UB] SE ̂
interval (CI) [LB, UB] Q-Statistic I2Index (%)
(1) Individual attributes
Abilities and skills 15 17,651 −.06 −.06 .11 [−.20, .08] .03 [−.12, .00] 217.38** 93.56
Age 121 209,588 −.21 −.21 .13 [−.39, −.04] .01 [−.24, −.19] 4,247.00** 97.17
Agreeableness 62,449 −.07 −.08 .08 [−.18, .02] .04 [−.15, .00] 14.78 66.17
Children 25 40,201 −.19 −.20 .11 [−.34, −.05] .02 [−.24, −.15] 458.33** 94.76
Conscientiousness 83,409 −.15 −.16 .14 [−.34, .03] .05 [−.26, −.05] 78.83** 91.12
Education 51 59,574 .04 .04 .17 [−.18, .26] .02 [−.01, .08] 1,105.78** 95.66
Emotional stability 16 7,593 −.17 −.19 .21 [−.45, .08] .05 [−.30, −.08] 490.23** 96.94
Ethnicity/race (0 =White,
29 457,562 .02 .02 .03 [−.01, .06] .01 [.01, .03] 494.06** 94.33
including Hom et al.
28 53,510 .02 .02 .09 [−.10, .14] .02 [−.02, .05] 492.57** 94.52
Extraversion 13 6,795 .02 .02 .07 [−.07, .11] .02 [−.03, .06] 38.01** 68.43
Internal motivation 17 5,960 −.13 −.16 .16 [−.36, .05] .04 [−.24, −.07] 150.28** 89.35
Locus of control
13 3,187 .08 .10 .13 [−.07, .27] .04 [.02, .18] 56.57** 78.79
Marital status (0 =
nonmarried, 1 =
27 134,505 −.10 −.10 .09 [−.21, .02] .02 [−.13, −.06] 1,099.32** 97.63
Openness to experience 5 1,009 .13 .14 .12 [−.01, .29] .06 [.02, .26] 16.08** 75.12
Sex (0 =female, 1 =male) 89 602,869 .00 .00 .12 [−.15, .16] .01 [−.02, .03] 2,078.58** 95.77
Sex not including Hom
et al. 2008 outlier
88 198,817 −.01 −.01 .07 [−.09, .08] .01 [−.02, .01] 1,066.54** 91.84
Tenu re 118 669,753 −.20 −.20 .10 [−.33, −.07] .01 [−.21, −.18] 7,652.76** 98.45
Tenu re not including
Hom et al. 2008 outlier
117 194,295 −.27 −.27 .17 [−.48, −.05] .02 [−.30, −.23] 6,659.12** 98.23
12 RUBENSTEIN ET AL.
TAB LE 2 (Continued)
Variable name kN
𝝆90% CV SE ̂
𝝆95% CI Q-Statistic I2Index (%)
(2) Aspects of the job
85,185 −.14 −.14 .03 [−.25, −.04] .02 [−.17, −.10] 33.46** 36.78
Job characteristics (VISAF) 16 12,869 −.16 −.18 .12 [−.33, −.02] .03 [−.23, −.11] 74.94** 79.98
Job security 52,155 −.21 −.23 .06 [−.32, −.14] .04 [−.30, −.16] 12.80** 68.75
Participation 5 1,895 −.11 −.13 .01 [−.14, −.11] .03 [−.17, −.08] 5.27* 24.10
Pay 55 177,634 −.17 −.17 .07 [−.26, −.08] .01 [−.19, −.15] 1,043.22** 94.82
Role ambiguity 8 5,765 .14 .15 .11 [.02, .29] .04 [.07, .23] 59.38** 88.21
Role conﬂict 10 10,903 .13 .15 .09 [.04, .27] .03 [.10, .21] 77.19** 88.34
Routinization 6 4,106 −.10 −.12 .09 [−.23, .00] .04 [−.20, −.04] 30.25** 83.47
Task complexity 10 3,117 −.01 −.01 .10 [−.14, .12] .04 [−.08, .06] 35.53** 74.67
Workload 21 82,204 −.10 −.10 .07 [−.18, −.01] .02 [−.13, −.07] 330.19** 93.94
(3) Traditionaljob attitudes
Job involvement 19 5,158 −.17 −.19 .15 [−.38, −.01] .04 [−.26, −.12] 84.62** 78.73
Job satisfaction 174 107,625 −.25 −.28 .18 [−.52, −.05] .01 [−.31, −.26] 2,046.12** 91.54
129 71,862 −.26 −.29 .14 [−.47, −.11] .01 [−.31, −.26] 1,365.33** 90.62
Other commitment 12 3,601 −.30 −.34 .17 [−.55, −.12] .05 [−.44, −.24] 95.38** 88.47
Other satisfaction 16 14,811 −.38 −.43 .22 [−.71, −.15] .05 [−.54, −.32] 777.93*** 98.07
(4) Newer personal conditions
Coping 7 880 −.32 −.39 .24 [−.69, −.08] .09 [−.57, −.20] 48.82** 87.71
Engagement 41,408 −.19 −.20 .04 [−.24, −.15] .03 [−.26, −.14] 5.60* 46.43
Stress/exhaustion 32 18,740 .20 .21 .15 [.03, .40] .03 [.17, .26] 384.85** 91.94
RUBENSTEIN ET AL.13
TAB LE 2 (Continued)
Variable name kN
𝝆90% CV SE ̂
𝝆95% CI Q-Statistic I2Index (%)
(5) Organizational context
Centralization 64,128 −.06 −.06 .15 [−.26, .14] .07 [−.19, .07] 94.41** 94.70
Climate 8 2,711 −.21 −.24 .18 [−.47, −.01] .06 [−.33, −.09] 78.73** 91.11
Organization prestige 52,433 −.05 −.06 .13 [−.23, .11] .07 [−.18, .06] 41.13** 90.27
Organization size 15 30,422 .03 .03 .13 [−.14, .20] .03 [−.04, .10] 531.27** 97.36
Organization support 16 8,256 −.19 −.19 .18 [−.43, .04] .04 [−.28, −.10] 291.81** 94.86
Reward contingency 4 678 −.17 −.20 .21 [−.40, .06] .11 [−.42, .03] 26.93** 88.86
Rewards offered 25 30,743 −.28 −.28 .15 [−.48, −.09] .03 [−.35, −.22] 754.02** 96.89
(6) Person–context interface
Fit 17 4,146 −.25 −.29 .25 [−.60, .03] .06 [−.41, −.17] 229.91** 93.04
Inﬂuence 7 24,331 −.08 −.09 .05 [−.15, −.02] .02 [−.13, −.04] 71.88** 91.65
Job embeddedness 29 31,158 −.25 −.26 .09 [−.39, −.14] .02 [−.30, −.22] 309.89** 90.96
Justice 30 17,556 −.15 −.17 .10 [−.30, −.05] .02 [−.21, −.13] 151.30** 80.83
Leadership 42 28,637 −.23 −.24 .16 [−.45, −.03] .03 [−.29, −.19] 748.09** 94.52
Met expectations 11 3,236 −.10 −.12 .23 [−.42, .17] .07 [−.26, .02] 118.59** 91.57
Peer/group relations 24 11,104 −.13 −.14 .13 [−.30, .03] .03 [−.19, −.08] 126.36** 81.80
7 8,083 .17 .18 .06 [.11, .25] .02 [.13, .22] 29.30** 79.52
14 RUBENSTEIN ET AL.
TAB LE 2 (Continued)
Variable name kN
𝝆90% CV SE ̂
𝝆95% CI Q-Statistic I2Index (%)
Work–life conﬂict 7 12,107 .16 .19 .06 [.11, .27] .03 [.14, .24] 45.73** 86.88
(7) External job market
Alternatives 79 58,512 .20 .23 .19 [−.01, .48] .02 [.19, .27] 1210.91** 93.56
(8) Attitudinal withdrawala
Withdrawal cognitions 211 73,405 .50 .56 .27 [.20, .90] .02 [.52, .59] 8,004.92** 97.38
(9) Employee behaviors
Absenteeism 36 44,405 .23 .23 .19 [−.02, .48] .03 [.16, .29] 766.22** 95.45
Employee performance 86 473,624 −.07 −.08 .11 [−.23, .06] .01 [−.11, −.06] 4,746.24** 98.21
not including Hom
et al. 2008 outlier
85 111,562 −.17 −.21 .19 [−.45, .03] .02 [−.25, −.17] 3,200.19** 97.38
Job search 27 18,685 .38 .40 .14 [.22, .58] .03 [.35, .46] 410.92** 93.67
Lateness 5 1,431 .14 .14 .06 [.06, .22] .04 [.07, .22] 10.87** 63.20
OCBs 96,047 −.09 −.10 .05 [−.16, −.02] .02 [−.12, −.07] 19.79** 72.91
4 3,016 −.10 −.11 .03 [−.14, −.06] .02 [−.16, −.06] 6.20 51.61
Note. *p<.05, **p<.01.
aWithdrawal cognitions are a weighted aggregate of search intentions, expected utility of withdrawal, withdrawal cognitions, and withdrawal intentions.
RUBENSTEIN ET AL.15
each construct in our study to create a 95% conﬁdence interval of their effects to thereby allow for comparison. Third,
we statistically compared the effect size differences between Griffeth et al. and our own as if they were independent,
again using Zou’s (2007) formula. This third approach effectively allows one to compare updated correlations to older
estimates as a means of assessing effect size stability based on increased study, including all available data. Invoking
all comparisons together, we deemed that if at least two of three approaches yielded signiﬁcant differences, then the
meta-analytic correlations were considered reliably different.
1.2.1 Individual attribute predictors
Among individual attributes, tenure (𝜌=–.27, outlier excluded) age (𝜌=–.21), children (𝜌=–.20), emotional stability
(𝜌=–.19), and conscientiousness (𝜌=–.16) demonstrate the strongest effects. Perhaps more important; however, age
validities signiﬁcantly differed compared to the Griffeth et al. (2000) analysis (hereafter, GHG: –.11, here: –.21), as did
the effect size for abilities/skills (GHG: .02, here: –.06). Implications of this larger age effect in particular (i.e., also more
negative for post-2000 compared to pre-2000 studies), merit comment. If older workers are less likely to quit, younger
workers are, equally, more likely to quit. Some scholars (e.g., Bal & Jansen, 2016) ﬁnd support for the idea that younger
workers hold higher—perhaps even unrealistic— expectations than do older workers regarding what they want from
their employers. Looking forward, researchers might monitor this trend, and if/how the broader deﬁnitions of careers
and work relationships change, and what that means for theory and practice.
Within this category there are also newly meta-analyzed constructs. Speciﬁcally, we ﬁnd that those individuals with
a more internal locus of control and those more internally motivated, are less likely to quit (𝜌=–.10 and 𝜌=–.16, respec-
tively). In face of stressful job demands or performance setbacks that inevitably occur during work life, such individuals
appear more likely to persevere and try to overcome such obstacles rather than quit. These variables are some of the
stronger individual difference predictors, and could be promising selection tools.
1.2.2 Predictors reﬂecting aspects of the job
Within this category are many newly estimated relationships, such as job characteristics (𝜌=–.18), job security (𝜌=
–.23), task complexity (𝜌=–.01), and workload (𝜌=–.10). Many of these effects generate interesting thoughts for future
research. For instance, the moderate negative relationship for job characteristicsmay indicate that managers can make
active efforts to reduce an individual’s turnover likelihood rather than assuming such decisions are made purely on
the basis of general dissatisfaction or dispositional factors (Humphrey, Nahrgang, & Morgeson, 2007). The variability
around these relationships also points to future research opportunities. For instance, a primary study might look at
how workload interacts with family demands, role integration, or internal motivation to predict turnover. It may be
that a high workload is only problematic for those who must also devote signiﬁcant portions of their time to other roles
(Brief, van Sell, & Aldag, 1979; Ilies, Wilson, & Wagner, 2009). Equally, complex work might only be detrimental to the
degree that work is overly burdensome and stressful.
Our results show that pay (GHG: 𝜌=–.11, here: 𝜌=–.17), role ambiguity (GHG : 𝜌=.24,here: 𝜌=.15), and ro le conﬂict
(GHG: 𝜌=.22, here: 𝜌=.15) all have signiﬁcantly different effects than found by Griffeth et al. (2000). Although the
face validity of the consistent negative effect for pay seems intuitive, challenges remain to explain whythis effect is not
stronger (compared to other predictors) or what it is about higher pay that reduces quitting. One explanation might be
found in job embeddedness theory (Mitchell et al., 2001): This would suggest that pay affects off-the-job embedded-
ness as it provides support for one’s lifestyle in a particular neighborhood or social caste; or, alternatively, pay might
more strongly increase on-the-job embeddedness by providing an objective signal of the “worth” of one’s employment,
as an aspect of their job that he or she would be reluctant to sacriﬁce by departing.
1.2.3 Traditional job attitudes predictors
For this category, we ﬁnd some of the stronger attitudinal predictors. Broadly speaking, these variables constitute
what March and Simon 1958 construed as “desirability of leaving” antecedents. Whereas the relationship for orga-
nizational commitment remains stable, we see stronger effects for job involvement (GHG: 𝜌=–.12, here: 𝜌=–.19) and job
16 RUBENSTEIN ET AL.
satisfaction (GHG: 𝜌=–.22, here: 𝜌=–.28). This latter effect is interesting, because with the addition of over four times
as many employees sampled, we now see nearly identical relationships between satisfaction and commitment, and
turnover (𝜌=–.28 and –.29, respectively). Such convergent validity is promising theoretically, as it offers greater justi-
ﬁcation to treat these variables as a single latent job attitude (Chang, Rosen, & Levy, 2009; Harrison, Newman, & Roth,
Equally noteworthy are the sizable effects for the newly meta-analyzed antecedents other commitment (𝜌=–.34)
and other satisfaction (𝜌=–.43). Per our deﬁnitions (see Table 1), these consist of factors like career commitment and
life satisfaction and, though based on fewer studies, demonstrate some of the strongest negative relationships with
turnover. These ﬁndings may open new avenues for research contributions. For example, whereas theorizing often
construes turnover as a response to proximal job conditions, the reported correlations suggest that quitting can also
be a function of a more distal and general life dissatisfaction. Given that people’s identities are often reﬂected by their
choice of occupation, life dissatisfaction could signal individuals seeking a “fresh start” and a need for control over one’s
life via a career change.
1.2.4 Newer personal conditions predictors
We ﬁnd the relationship for stress is signiﬁcantly larger than in earlier years (GHG: 𝜌=.16, here: 𝜌=.21, post-2000:
𝜌=.23). Indeed, the correlation between study year and stress effect sizes was also positive and signiﬁcant (r=.16,
p<.05). This ﬁnding perhaps alludes to generational effects, such as with regard to stress reactivity or work–life bal-
ance concerns. Perhaps individuals who have recently entered the workforce are less prepared to manage stress? Or,
do such employees perceive stress differently than earlier generations? With increasingly blurred lines between work
and nonwork roles, and demands to be “on 24/7,” it is possible that what were previously considered positive challenge
stressors (e.g., workload, time pressures) are, for some, slowly evolving into negative hindrance stressors (Podsakoff,
LePine, & LePine, 2007). Supporting this logic, surveys ﬁnd that younger generations report the most stress—
particularly caused by work—and the least relief (American Psychological Association, 2012).
We also provide new meta-analyses for employee coping (𝜌=–.39) and engagement (𝜌=–.20) and turnover. These
effect sizes are encouraging—particularly for engagement—which has recently gained increased study and practitioner
interest (Bakker & Leiter, 2010; Wefald & Downey, 2009, see also our Section 5.3). These articles discuss how there
is limited research on engagement consequences, so our results provide some preliminary testament to its predic-
tive validity. Continued study of engagement will be quite useful, particularly as technological advances modify how
physical work is performed and how the meaning of “engagement” changes (e.g., automation, virtual work, e-business,
increased connectivity). It may be the case that engagement explains unique turnover variance beyond traditional
1.2.5 Organizational context predictors
The organizational context has generally been ignored in turnover research until recently (Hom et al., 2017). As such,
most meta-analyses within this category are new. We see notable effects for climate perceptions (𝜌=–.24), organi-
zational support (𝜌=–.19), and rewards offered (𝜌=–.28). Such results provide compelling evidence that the broader
context does matter for turnover (Johns, 2006). We expand on this idea later, with moderators examining how context
affects antecedent–turnover relationships. Further, because contextual factors like those above tend to be less affec-
tive in nature, in a multivariate sense, we might expect them to account for unique turnovervariance beyond traditional
attitudes (Carr, Schmidt, Ford, & DeShon, 2003).
Comparatively weakereffects were found for variables like organization size (𝜌=.03) and prestige (𝜌=–.06). Although
we only compiled ﬁve studies for prestige, the results suggest that individual turnover decisions are less a function of
what the organization is like in an absolute sense but rather are more strongly determined by how a given employee
is immersed within that organization. However, especially for prestige, there was considerable variability around this
effect (SE =.07), suggesting possible moderators. For instance, an organization’s prestige might only affect turnover
for those who desire it (i.e., as it indicates high status).
RUBENSTEIN ET AL.17
1.2.6 Person–context interface predictors
Across all categories, we ﬁnd some of the most promising results for person-context interface predictors. Many effects
have changed, or are now conceptually different from previous work. For instance, with three times as many employ-
ees surveyed, we ﬁnd signiﬁcantly stronger ﬁt effects compared to the Kristof-Brown, Zimmerman, and Johnson (2005)
meta-analysis (theirs: 𝜌=–.08, here: 𝜌=–.29), signiﬁcantly weaker effects for met expectations (GHG: 𝜌=–.18, here: 𝜌=
–.12), as well as slightly stronger (although not signiﬁcantly different) effects for overall job embeddedness compared to
Jiang et al. (2012) (theirs [average on- and off-the-job embeddedness]: 𝜌=–.16, here: –.26). These sizable effects offer
some indirect support for person–environment ﬁt and attraction–selection–attrition theories, both of which empha-
size how employees seek work environments that align with their demography, personality, and values. We elaborate
more on this personal ﬁt idea in our Step 2 moderator analysis.
This is also the ﬁrst large-scale meta-analysis linking all aspects of justice to turnover behavior (𝜌=–.17), as previous
studies only examined its relationship to intentions or only looked at distributive justice. Our updated meta-analysis
with nearly three times as many studies as Griffeth et al. (2000) reveals a signiﬁcantly stronger effect (GHG: 𝜌=–.11).
This effect is also relatively strong among the many attitudinal predictors examined, emphasizing how much individuals
value fair and equitable treatment/outcomes from their employers. Indeed, many organizations have gone so far as to
ban discussing pay differences or other equity issues in the workplace, which makes sense given the consequences
shown here, should injustice occur.
1.2.7 External job market predictors
For alternatives, our results with almost four times as many samples show a signiﬁcantly increased effect size (GHG: 𝜌=
.15, here: 𝜌=.23). Given that many turnover process models (e.g.,Hom, Griffeth, & Sellaro, 1984; March & Simon, 1958;
Mobley,1977; Price & Mueller, 1981) include a stage in which employees compare alternatives to one’s present job, the
full explanatory power of such models depends on signiﬁcant effects at each mediated link. Although a validity of .15 is
modest, our updated estimate of .23 lends support to job alternatives as one of the more important antecedents of quit
decisions. However, other theoretical approaches, such as the unfolding model (Lee & Mitchell, 1994), propose that
some employees quit without an alternative job in hand. Although some studies position alternatives as a boundary
condition to relationships such as job search–turnover (Swider, Boswell, & Zimmerman, 2011), there still remain many
unanswered questions in this literature. In Step 2, we seek to add clarity to this literature by examining the moderating
role of job market alternatives.
1.2.8 Attitudinal withdrawal predictors
Not surprisingly, withdrawal cognitions such as intent to leave have the strongest correlation with turnover, and we ﬁnd
this effect to also be stronger compared to Griffeth et al. (2000), now having compiled three times as many studies
(GHG: 𝜌=.45, here: 𝜌=.56). One noticeable implication of this result is that proximal withdrawal attitudes and other
on-the-job judgments tend to better predict turnover than do more distal person or organizational characteristics. A
second implication is that, although the turnover cognitions-behavior effect is quite strong, it is still not an identity. We
would therefore caution researchers against treating turnover cognitions or intentions as a simple proxy for behavior,
because there is slippage between intent and action. We elaborate on this issue in our Step 3 recommendations.
1.2.9 Employee behaviors predictors
Lastly, regarding behaviors, we generally ﬁnd similar results to other meta-analyses for predictors such as performance
(𝜌=–.21; Hom, Roberson, & Ellis, 2008 outlier excluded), citizenship behavior (OCBs, 𝜌=–.10), lateness (𝜌=.14), and
absenteeism (𝜌=.23) (Berry et al., 2012; Griffeth et al., 2000; Podsakoff, Whiting, Podsakoff, & Blume, 2009). Still, there
remain few studies on lateness and turnover,and theory development is still needed, particularly with respect to isolat-
ing reasons for lateness itself: Is it an overt and conscious act of insubordination, does it often coincide with job search,
is it largely dispositional (i.e., low conscientiousness), or is it a random behavior (i.e., difﬁcult-to-time commutes)? Pro-
gression theories of withdrawal support lateness as a catalyst to absenteeism and subsequent quitting (Rosse, 1988),
18 RUBENSTEIN ET AL.
yet work remains as to understanding what drives lateness in the ﬁrst place or how employees perceive lateness to
reinforce possible job dissatisfaction (Koslowsky, Sagie, Krausz, & Singer, 1997).
For job search, however, we do ﬁnd a signiﬁcantly stronger effect size (GHG:𝜌=.31, here: 𝜌=.40), and even stronger
among post-2000 studies (k=16; 𝜌=.43). Although employees may have various reasons for searching (Boswell,
Boudreau, & Dunford, 2004), the overall trend we observe is that those who search for alternatives are more likely
to quit. Yet, this effect also has a relatively wide conﬁdence interval, hinting at possible moderators (e.g., Swider et al.,
1.3 Progress assessment
In surveying the considerable growth of research that has accumulated in the 21st century, we have obtained a much
clearer picture of the turnover forest for many predictors and their relationships with turnover behavior. Many effects
increased, others decreased, whereas some have remained stable. Interestingly, many new predictors emerged that
show promise. Although it is possible some effect size increases could be due to studies in contexts yielding larger
effects (e.g., from occupations with high turnover), the increase in studies augurs well for the trustworthiness of the
results. We know that traditional attitudes such as commitment and satisfaction are strong predictors of turnover,
and withdrawal behaviors such as absenteeism and job search also signal impending exit. However, rather than hewing
to only these predictors, we advise researchers and practitioners to also consider other factors that we ﬁnd to also
inﬂuence exit decisions, for there are many future opportunities for contribution and understanding. The increased
relevance of personality, engagement, job embeddedness, and the organizational context suggests at the very least
continued attention to these constructs will be worthwhile. For instance, core self-evaluations (CSEs) are a recently
introduced meta-construct comprises emotional stability, locus of control, generalized self-efﬁcacy (i.e., an internal
motivation construct), and self-esteem (Judge & Bono, 2001). Although we found no primary studies examining if, how,
or why CSEs as a whole predict turnover, speciﬁc CSE dimensions do independently contribute to exit decisions (see
also Zimmerman, Swider,Woo, & Allen, 2016 for an integrative review regarding how distal individual differences man-
ifest into work withdrawal). Theory and testing about combinations of these attributes may be quite fruitful. Further,
there are opportunities to build off of our content-related results (i.e., what predicts exit), with primary studies model-
ing processes and competitive theory testing (i.e., how and why antecedents predict exit).
Beyond these zero-order results, though, many constructs had a signiﬁcant amount of heterogeneity around effect
sizes. To better account for this, in Step 2, we developed theoretical arguments for multiple substantive moderators of
turnover relationships. Due to the context-bound nature of turnover, such an analysis might yield greater understand-
ing and more precise predictions as to who quits, when they quit, and why (Holtom et al., 2008).
2STEP 2: META-ANALYTIC MODERATOR ANALYSES
Given the variability observed around our estimates, we considered potential moderators of antecedent–turnover
effects. Although meta-analytic methods generally preclude the testing of individual-level moderators (e.g., examining
if the effect of job satisfaction on turnover depends on employees’ personality traits), they do provide opportunities to
test contextual or sample-level moderators, which are typically more challenging to capture in primary studies (Park&
Shaw, 2013). Speciﬁcally, our moderation analyses broadly reﬂect how the organizational and economic context can
inﬂuence the magnitude of the relationships between individual antecedents and turnover behavior. Johns (2006)
noted that turnover studies rarely attend to context, despite contextual inﬂuences likely factoring into employee’s
turnover decisions and accounting for variance beyond individual antecedents alone (see also Cappelli & Sherer, 1991;
Holtom et al., 2008; Steel, 2002). Following Johns (2006), in Step 2, we tested a set of contextual moderators in terms
of cross-level effects, where the mean levels of a variable in the sample or organization (e.g., mean age across employ-
ees, mean base rates of turnover in an organization) or in the economy (e.g., unemployment rates when data were col-
lected) are expected to render relationships between individual-level antecedents and turnover stronger or weaker.
RUBENSTEIN ET AL.19
Our meta-analytic dataset provides us information to examine how organizational and economic factors (Level-2)inﬂu-
ence individual-level antecedent–turnover effect sizes (Level-1; see also Park & Shaw, 2013 for a similar examination
of sample-level moderation regressions).
Context can be conceptualized and operationalized in different ways. In this paper, we view context from four differ-
ent perspectives: personal ﬁt, attitudinal climate, the job market, and turnovercontagion. Context is important because
it provides unique meaning as to how similar or different one is relative to others on a given variable (personal ﬁt per-
spective): For instance, an employee who is unable to sell much of a product to customers may be likely to quit (out of
frustration at their ability), but this relationship might be weakened if the employee were to ﬁnd out that the average
sales volume across all employees in the organization was not much different than their own. Context also matters
because it provides a salient and social benchmark against which to gauge one’s own attitudes and behaviors. In other
words, the context or organizational climate helps an employee make sense of their own attitudes and behaviors, as
it allows for self-comparison to others in the organization. In turn, this social comparison might buffer or amplify the
likelihood of turnover. For example, an employee with high job satisfaction may be even more likely to stay if he or
she is surrounded by many others who share a high level of satisfaction (climate perspective). Finally, context matters
because it can elicit situational opportunities or constraints on behavior: The efﬁcacy of an employee’s search for a new
job will likely be tempered by overall job availability in the external market (job market perspective), just as one’s job
dissatisfaction may create more motivation to leave if turnover rates in the organization are high (turnover contagion
perspective; Johns, 2006).
Following these four perspectives (i.e., ﬁt, climate, job market and turnover contagion) in the next sections, we
developed a set of moderation hypotheses. Guided by theory, stipulating a sufﬁcient number of effect sizes to test
for moderation (≥15), examining moderators only where the zero-order effects results indicated signiﬁcant hetero-
geneity, and excluding those studies where contextual information was not available,2we arrived at 10 moderators
to test the personal ﬁt perspective (sample mean levels of an antecedent moderating its respective individual-level
antecedent–turnover relationship), two moderators to test the climate perspective (mean levels of job satisfaction
and organizational commitment), two moderators to test the external job market perspective (mean levels of per-
ceived alternatives and annual unemployment rate), and three moderators to test the turnover contagion perspec-
tive (turnover base rate, mean levels of withdrawal cognitions, and job search). We test these moderators for the
following antecedent–turnover relationships: absenteeism, age, alternatives, education, employee performance, job
embeddedness, job satisfaction, job search, organizational justice, organizational commitment, organizational tenure,
pay, sex, stress/exhaustion, and withdrawal cognitions. However, not every moderator is tested for every antecedent–
turnover relationship, partly due to data unavailability issues and partly due to lacking theoretical rationale for a
2.1 Personal ﬁt
Past theory and research suggests that people are more likely to seek out other individuals and remain in environ-
ments that are similar to themselves, in terms of biographical and personality factors, attitudes/beliefs/values, behav-
ioral habits, and other sociodemographics. As a result, social networks tend to be relatively homogenous on these
organizing characteristics, termed the homophily principle (McPherson, Smith-Lovin, & Cook, 2001). On the other hand,
should individuals ﬁnd themselves to be dissimilar to others on these factors, they will often remove themselves from
such environments. Person–environment ﬁt theory (Kristof, 1996) and, particularly germane to turnover, attraction–
selection–attrition theory (Schneider, 1987), highlight this idea, in which homophily is reinforced as employees are
attracted to organizations similar to themselves and are more likely to quit organizations—or more speciﬁcally, the
people in those organizations—if they deem themselves dissimilar,or a misﬁt. In this way, Schneider theorized turnover
as a means of correcting an error in homophily judgment, if the reality of the organization is or becomes divergent on
one or more of these characteristics.
Although in our initial meta-analysis we did examine the “ﬁt” construct, we expand the notion of ﬁt here to repre-
sent the extent to which an employee is unique on demographic, attitudinal, or behavioral factors compared to other
20 RUBENSTEIN ET AL.
employees. For demographics, for example, we expect to ﬁnd that the relationship between sex (coded 0 =female,
1=male) and turnover will be more negative when the sex makeup of an organization is predominately male (i.e., men
will be less likely to quit if most other employees are also male). In terms of attitudes, we expect the ﬁt effect to operate
such that turnover effect sizes will be more positive when employees aversely trend away from the mean-level atti-
tudes of employees in a sample. For example,the negative individual-level job satisfaction-turnover effect may become
more negative (further away from zero) when sample mean-level satisfaction is higher (i.e., highly satisﬁed employ-
ees may be even less likely to quit if everyone else is also satisﬁed). Or, the positive relationship between individual
stress and turnover is expected to be more positive when average stress levels across employees are lower (i.e., highly
stressed employees may be evenmore likely to quit if others are not very stressed). More formally, we hypothesize that
the relationship between a given antecedent and turnover at the individual level will be moderated by the respective
antecedent’s sample mean level. Based on the inclusion criteria noted above, we examined the following moderators:
age, education, employee performance, job embeddedness, job satisfaction, justice, organizational commitment, orga-
nizational tenure, sex, and stress.
Hypothesis 1: The individual-level relationships between employee age, education, performance, job embeddedness,
job satisfaction, justice, organizational commitment, organizational tenure, sex and stress/exhaustion
and turnover behavior will be moderated by the antecedent’s respective sample mean-level, such that
relationships will become more positive (or less negative) when employees are more dissimilar (i.e., a
misﬁt) to others on that antecedent.
2.2 Attitudinal climate
Turnover relationships may also vary as a function of the organizational climate, deﬁned here as the shared expe-
riences, perceptions, and behavioral tendencies among a group of employees (Schneider, Ehrhart, & Macey, 2013).
Ostroff (1993) offered a taxonomy of climate perceptions, representing three broad facets: affective, cognitive,
and instrumental (see also Carr et al., 2003). Drawing from this theoretical template, we hypothesized that some
antecedent–turnover relationships might vary as a function of whether attitudinal climates are more or less favor-
able in valence. From our initial meta-analysis, we ﬁnd employees are less likely to quit when working in more posi-
tive climates. We extend this notion to moderators, suggesting that a positive attitudinal climate might buffer certain
antecedents from translating into turnover. From a social interactionist perspective, research has shown that climate
can serve such a moderating role, because individuals are attuned to their environments, seek to cohere with them,
and use them to derive important signals about their attachment to the organization (Eisenbeiss, van Knippenberg, &
Boerner, 2008; Li, Liang, & Crant, 2010; Smith-Crowe, Burke, & Landis, 2003).
Speciﬁcally, we posit that turnover will be less likely (i.e., effects more negative) when sample mean job satisfac-
tion and organizational commitment are more favorable. For example, as we ﬁnd from Step 1 that withdrawal cog-
nitions positively predicts turnover, we expect such a relationship to become less positive (i.e., employees who think
about quitting will be not as likely to quit) when mean job satisfaction and organizational commitment are higher (e.g.,
Liu, Mitchell, Lee, Holtom, & Hinkin, 2012). A generally more satisﬁed/committed workforce might deﬂect an individ-
ual’s thoughts of leaving in an effort to maintain the positive environment or because a positive climate might become
embedding insofar as it would be a sacriﬁce to give up should one leave—and job embeddedness has been shown to
have a buffering effect on turnover (Allen, Peltokorpi, & Rubenstein, 2016; Holtom & Inderrieden, 2006; Swider et al.,
With this hypothesis, we must note two caveats: First, like our ﬁt hypothesis, we measured “climate” in our initial
meta-analysis. However, the key difference in our treatment of climate as a moderator is by using sample mean per-
ceptions of satisfaction and commitment. Most primary studies of climate tend to instead use a referent shift approach
(Chan, 1998), where employees ratethe attributes of their unit or organization rather than gauging their own attitudes.
However, we would argue that mean ratings across employees are perhaps a better indicator of climate than is a ref-
erent shift, for mean levels assess how employees themselves, on average, actually feel about the favorability of the
RUBENSTEIN ET AL.21
organization rather how they think others feel. Second, although one could argue that mean stress, performance, with-
drawal cognitions, or even demographics (e.g., an older workforce) can be construed as other “climates,” we did not
examine them in the climate category, in part because some moderators were already subsumed in other categories
(i.e., under the personal ﬁt perspective), or due to insufﬁcient data. Based on data availability, we tested the climate per-
spective for the following antecedents: age, alternatives, commitment, tenure, pay, sex, and withdrawal cognitions.
Hypothesis 2: The individual-level relationships between employee age, alternatives, commitment, tenure, pay, sex
and withdrawal cognitions, and turnover behavior will be moderated by sample mean-level job satis-
faction and organizational commitment, such that the relationships will become less positive (or more
negative) when the attitudinal climate is more favorable (i.e., when mean levels are higher).
2.3 Job market
Third, we considered job market factors that might constrain turnover. Such moderators align with March and Simon’s
(1958) notion of “ease of leaving,” suggesting that turnover will be less likely if there is general scarcity of alternative
employment available. Although alternatives positivelypredict turnover, we also expect alternatives to moderateother
antecedent–turnover relationships, in that more/less available jobs (actual or perceived) might expand/limitthe extent
to which antecedents are acted upon. For example, research has found that job search (Swider et al., 2011) and job
satisfaction (Trevor, 2001) are less likely to result in quitting if one perceives few alternatives and that the majority of
quitters do leave with replacement jobs in hand (Lee & Mitchell, 1994; Lee, Mitchell, Holtom, McDaniel, & Hill, 1999).
We tested two job market moderators: Unemployment rates at the time data were collected (for U.S.-based sam-
ples only) and sample mean job alternatives. When unemployment rates are higher, and average job alternatives per-
ceptions are lower, we take this to mean that there are fewer jobs available in the external market and as such, positive
(negative) turnover relationships should weaken (strengthen). The moderating effect of mean alternatives could only
be tested for the following antecedents: alternatives, job satisfaction, and withdrawal cognitions. Yet, for unemploy-
ment rates as a moderator, all antecedents were examined (a total of 15).
Hypothesis 3: The individual-level relationships between employee absenteeism, age, alternatives, educa-
tion, performance, embeddedness, satisfaction, search, justice, commitment, tenure, pay, sex,
stress/exhaustion and withdrawal cognitions, and turnover behavior will be moderated by sample
mean-level job alternatives and U.S. yearly unemployment rates, such that relationships will become
less positive (or more negative) when the unemployment rates were higher when data were collected
and when sample mean-level job alternatives are lower.
2.4 Turnover contagion
Turnover contagion theory (Felps et al., 2009) suggests that an employee’s propensity to leavea job can partly be inﬂu-
enced by whether other employees have also left their jobs or are intending to do so. This line of theorizing describes
a process whereby employees compare themselves and their attitudes to others in order to determine if they should
quit. That is, one’s own turnover propensity can be affected by the salience of others’ attitudes about quitting and oth-
ers’ actual quitting behavior. This moderator was considered in Griffeth et al. (2000); however, they only examined one
relationship, pay-turnover, and dichotomized turnover base rates as more or less than 15%. In our analyses, we exam-
ined base rates as a continuous moderator with many more samples (thereby improving the statistical power of such
tests), along with considering other potential moderated antecedent–turnover relationships.
To test the original turnover contagion hypotheses, Felps et al. (2009) operationalized one’s coworkers’ (low) job
embeddedness and job search behaviors as representing a high contagion environment. In the present study, we
offer an even stronger test of turnover contagion theory by considering sample-level contagion indicators, thereby
allowing comparisons across organizations. A clear assertion of turnover contagion would be reﬂected by actual
turnover rates in a sample, mean withdrawal cognitions across employees or mean job search rates (we considered job
22 RUBENSTEIN ET AL.
embeddedness, but it was ultimately omitted due to insufﬁcient primary study data). Such moderation would involve
testing whether, for example, employees are more likely, on average, to translate higher alternatives, job search, or
withdrawal cognitions into quitting when a sizable proportion of employees have already quit, are thinking about quit-
ting, or are searching for new work. Such questions cannot be answered as easily through primary studies but are read-
ily accessible as cross-level meta-analytic moderators.
Although both job search ×mean-level job search and withdrawal cognitions ×mean-level withdrawal cognitions
could be interpreted under the personal ﬁt perspective, we treat them here as turnover contagion, insofar as we expect
to see positive relationships. That is, if an employee is thinking about quitting or is searching for alternative employ-
ment, if everyone else is also thinking about quitting or is searching for new work, we expect the relationships between
withdrawal cognitions and job search to turnover to be even stronger. Speciﬁcally, we tested the moderating role of
turnover base rates for all 15 antecedents, sample mean-level withdrawal cognitions for 13 antecedents (excluding
education and stress), and sample mean-level job search only for job search (technically, this also provides a test of the
Hypothesis 4: The individual-level relationships between employee absenteeism, age, alternatives, educa-
tion, performance, embeddedness, satisfaction, search, justice, commitment, tenure, pay, sex,
stress/exhaustion and withdrawal cognitions, and turnover behavior will be moderated by sample
mean-level turnover base rates, withdrawal cognitions, and job search, such that relationships
with turnover will become more positive (or less negative) when turnover base rates, withdrawal
cognitions, and job search are higher.
2.5.1 Coding and procedure
For each primary study, we coded for 14 sample-level moderators. They were sample means for age, tenure, sex makeup
(percent male), education, job satisfaction, organizational commitment, stress/exhaustion, job embeddedness, alterna-
tives, withdrawal cognitions, job search, employee performance, and the mean sample turnover base rate. Using only
studies from the United States, unemployment rate was gathered from the year when the data were collected; if not
noted, we used the year prior to the study being published (to allow for publishing time lag). We obtained yearly unem-
ployment data from the Bureau of Labor Statistics website.
To obtain moderator sample means, we reviewed the “Methods” sections and correlation matrices of each primary
study (i.e., mean and SD reports). As noted earlier, we excluded studies collapsing multiple organizations into a single
sample to isolate moderation effects to only the sample for which mean values were applicable. For mean age, tenure,
sex, turnover base rates, and unemployment rates, we used reported values (or website data, for unemployment),
although for tenure we transformed all codings into tenure in years (some studies reported in months or weeks). Forthe
other moderators, rescaling was necessary so as to interpret means equally across studies using different scale ranges.
To do this, we divided reported values by the maximum of the given scale used in the study (Aguinis, Gottfredson,
& Culpepper, 2013). For instance, if job satisfaction were rated on a seven-point Likert scale, a study reporting mean
satisfaction levels of 3.50 would yield a standardized value of 0.50. We were conservative in this regard, in that we only
coded what was reported in study text. Although a mean and SD might indirectly imply scale maximums (e.g., a mean of
3.50 with a SD of 0.77 might suggest a ﬁve-point scale), we did not extrapolate beyond what was reported.
We tested hypotheses using SPSS version 23, using a weighted least squares (WLS) regression approach. WLS is
advantageous compared to testing for signiﬁcant bivariate correlations between effect sizes and moderator levels,
comparing hierarchical subgroups (two methods used in Griffeth et al., 2000), or ordinary least squares regression,
because WLS accounts for correlations among moderators while proportionally weighing primary studies based on
the inverse of the sampling error variance. As such, WLS regression is altogether a more statistically powerful method
to test for moderation (Steel & Kammeyer-Mueller, 2002).
RUBENSTEIN ET AL.23
2.6 Results and discussion
Descriptive statistics for standardized moderators (i.e., a proportion out of 1.00) were as follows: For education (many
studies used scales rather than absolute years educated), the grand mean was 0.53, which on a ﬁve-pointscale indicates
roughly between some college and a bachelor’s degree. The grand mean for job satisfaction was 0.70 (SD =0.11); for
organizational commitment, 0.67 (SD =0.08); for stress/exhaustion,0.53 (SD =0.09); for job embeddedness, 0.54 (SD =
0.15); for alternatives, 0.68 (SD =0.16); for withdrawal cognitions, 0.49 (SD =0.15); for job search, 0.42 (SD =0.13); and
for employee performance, 0.66 (SD =0.18).
Moderation results are shown in Table 3. The results for the personal ﬁt, attitudinal climate, job market and turnover
contagion perspectives are shown in the second through ﬁfth columns, respectively. Each row lists antecedents for
each hypothesis test. Cells ﬁlled in by a “–” had insufﬁcient data or theoretical rationale for inclusion, and cells with
a “#” denote an effect examined in a different moderator category (the result is shown in a different column). WLS
regressions were simultaneously performed for all moderators across each antecedent row.
To aid readers in interpreting the results, we note that a negative moderator effect would make an antecedent–
turnover relationship more negative (or less positive), whereas a positive moderator effect would make a relationship
more positive (or less negative). This is because we interpret a zero-order relationship in terms of how higher levels of
an antecedent translates to higher (positive effect) or lower (negative effect) turnover likelihood. For example, we ﬁnd
that the negative age–turnover relationship (𝜌=–.21, older employees are less likely to quit, on average) is negatively
moderated by respective mean sample age (i.e., personal ﬁt perspective), suggesting that in organizations where the
mean age is higher, individual-level age–turnover correlations are even more negative. Mathematically, Table 3 shows
that for each year increase in the mean age of across employees in a sample, the correlation between individual age
and turnover becomes negative by a further .02, or for every SD increase in mean employee age (roughly 6.7 years),
the age–turnover relationship becomes further negative by .81 SDs (about .14 correlation points). Conversely, the
alternatives-turnover relationship (𝜌=.23) is positively moderatedby mean withdrawal cognitions across employees in
asample(𝛽=.33, p<.01), meaning that in organizations where employees as a whole are thinking more about quitting,
the alternatives–turnover relationship becomes further positive (i.e., turnover contagion perspective).
Hypothesis 1 proposed a set of personal ﬁt moderators, suggesting that positive (negative) individual-level
antecedents–turnover relationships would become stronger (weaker) when employees are more dissimilar from sam-
ple mean-level of that respective antecedent. Column 2 in Table 3 summarizes these results. We found numerous sig-
niﬁcant moderated effects. Speciﬁcally,signiﬁcant effects were seen for employee age (𝛽=–.81, p<.01), education (𝛽=
–.38, p<.05), sex (𝛽=–1.13, p<.05), job satisfaction (𝛽=–.39, p<.05), organizational commitment (𝛽=–.21, p<.01),
tenure (𝛽=–.61, p<.01), and stress (𝛽=–.57, p<.01).
As noted above, we ﬁnd that the negative relationship between individual age and turnover becomes increasingly
negative in samples with higher mean-level age, whereas the nonsigniﬁcant relationship between education and
turnover becomes signiﬁcantly more negative in samples where the mean education level is higher. Interestingly, we
also ﬁnd that the null zero-order relationship between sex and turnover becomes signiﬁcantly more negative when the
sex makeup of the sample is increasingly male. A similar personal ﬁt result is also seen for tenure, where the negative
relationship between tenure and turnover becomes more negative in organizations mostly made up of senior-level
employees. Pfeffer (1983) documented such an organizational demography phenomenon for age and tenure in the
United States railroad industry, where younger, newcomer employees were deterred from remaining in the industry
because most workers were older and of senior level. Subsequently, newcomers perceived greater misﬁt to others
in the trade, along with limited opportunities for promotion/advancement. Here, we generalize this ﬁt effect to many
more organizations. One other ﬁnding is also noteworthy: Satisfaction and commitment as ﬁt moderators are both
signiﬁcant and negative, suggesting that, in organizations where employees are more satisﬁed and committed, on
average individual satisfaction/commitment–turnover relationships are even stronger (i.e., unhappy employees are
even more likely to leave when surrounded by mostly happy peers). That is, not only does dissimilarity to others on
surface-level factors affect turnover, but also employees are cognizant of attitudinal dissimilarity between themselves
and others and that this dissimilarity can amplify negative perceptions. In a related vein, the moderation effect for
24 RUBENSTEIN ET AL.
TAB LE 3 Weighted least squares regression analysis of individual-level antecedent–turnover correlations as a function of contextual sample-level moderators
Moderator group Perspective 1: Personal ﬁt Perspective2: Attitudinal climate Perspective 3: Job market
moderator Mean respective antecedent Mean job satisfaction
commitment Mean alternatives
Absenteeism – – – – – – – – – – – – .06 .04 .47
Age −.02 .01 −.81** −.18 .11 −.29 −.24 .22 −.30 – – – .02 .01 .26
Alternatives ## #−.03 .46 02 −.50 .73 −.23 −.05 .25 −.06 .01 .02 .18
Education −.18 .10 −.38*– – – – – – – – – .05 .01 .70*
Employee performance .38 .85 .20 – – – – – – – – – .03 .04 .20
Job embeddedness .28 .68 .14 – – – – – – – – – −.04 .06 −.33
Job satisfaction −.58 .30 −.39*## #–– –−.97 .41 −.70*.04 .03 .34
Job search # # # – – – – – – – – – .00 .03 .01
Justice −.53 .43 −.43 – – – – – – – – – .02 .02 .23
−.35 .09 −.21** −.14 .23 −.12 # # # – – – −.01 .02 −.10
Organizational tenure −.002 .001 −.61** −.46 .74 −.12 – – – – – – .01 .03 .04
Pay – – – .64 .50 .39 – – – – – – −.01 .02 −.12
Sex (0 =female, 1 =
−.61 .18 −1.13** −.20 .20 −.30 – – – – – – .08 .03 .76*
Stress/exhaustion –.73 .18 −.57** – – – – – – – – – −.02 .01 −.29*
Withdrawal cognitions # # # 1.03 .47 .54*.88 .46 .36 .97 .42 .45*−.06 .04 −.32
RUBENSTEIN ET AL.25
TAB LE 3 (Continued)
Moderator category Perspective 4: Turnover contagion
Speciﬁc moderator Mean turnover base rate Mean withdrawal cognitions Mean job search behaviors
Independent variable BSE𝜷BSE𝜷BSE𝜷
Absenteeism .64 .38 .50 −.20 .53 −.11 – – –
Age −.70 .21 −.46** −.30 .18 −.27 – – –
Alternatives −.20 .59 −.12 .43 .11 .33** –––
Education .08 .16 .09 – – – – – –
Employee performance −.45 .53 −.29 1.54 .76 .78*–––
Job embeddedness 1.29 1.92 .30 –1.54 1.27 −.45 – – –
Job satisfaction −.85 .38 −.48*−.17 .25 −.14 – – –
Job search −.48 .50 −.40 −.64 .96 −.22 .41 .20 .67*
Justice .40 .32 .35 −.53 .28 −.54*–––
Organizational commitment −.04 .16 −.04 −.47 .19 −.51** – – –
Organizational tenure −.32 .23 −.26 1.11 .51 .46*–––
Pay −.06 .47 −.05 .78 .34 .71*– – –
Sex (0 =female, 1 =male) .50 .34 .38 −.42 .40 −.32 – – –
Stress/exhaustion −.41 .27 −.20 – – – – – –
Withdrawal cognitions −.57 .57 .26 .82 .36 .55*–––
Note. Perspective 1 “mean respective antecedent” represents each row’s antecedent moderated by its own sample mean-level(e.g., individual age–turnover effect moderated by sample mean-
level employee age; stress-turnover moderated by sample mean-level stress). Sex moderator coded as percent male. Each study was weighted by the inverse of the sampling error variance.
Number of studies and blank cells vary due to speciﬁed hypotheses and/or insufﬁcient data. Each row is a single WLS regression of all moderators regressed on a given antecedent–turnover
effect size. “–” denotes insufﬁcient available data to test moderator. “#” denotes a moderator examined in a different theoretical perspective. A negative moderation effect means an antecedent–
turnover relationship becomes more negative, or less positive, as the moderator value increases in absolute magnitude. A positive moderation effect means that an antecedent–turnover
relationship becomes more positive, or less negative, as the moderator value increases in magnitude.
26 RUBENSTEIN ET AL.
stress-turnover ×sample mean-level stress was also signiﬁcant, such that the positive zero stress-turnover effect
size became more negative (i.e., less positive) in organizations where employees as a whole reported higher stress
levels. Although nonsigniﬁcant moderation effects were found for performance, job embeddedness, and justice, most
ﬁndings are signiﬁcant, so we conclude that Hypothesis 1 is generally supported.
Hypothesis 2 considered climate moderators of mean-level job satisfaction and organizational commitment. These
results are presented in Table 3, column 3. Unfortunately, few studies reported on these variables’ sample mean values
along with antecedents, so we were limited in the scope of our moderator tests. The only signiﬁcant effect found was for
satisfaction moderating withdrawal cognitions-turnover (𝛽=.54, p<.05), but surprisingly, this effect is in the opposite
direction as hypothesized. Thus, we generally ﬁnd no support for Hypothesis 2.
Hypothesis 3 posited that job market conditions would moderate antecedent–turnoverrelationships such that rela-
tionships would be more positive (less negative) when more jobs are available. Again unfortunately, few studies mea-
sured both alternatives and other antecedents to fully examine this moderator, but we were able to test all moderating
effects for unemployment rates (for U.S. samples only). As shown in Table 3, column 4, some moderation effects for
alternatives or unemployment rate are not signiﬁcant, but the following are: withdrawal cognitions ×mean alterna-
tives (𝛽=.45, p<.05), stress ×unemployment (𝛽=–.29, p<.05), job satisfaction ×mean alternatives (𝛽=–.70, p<
.05), education ×unemployment (𝛽=.70, p<.05), and percent male ×unemployment (𝛽=–.76, p<.05). These results
suggest, for instance, that the positive relationship between thinking about quitting and actually quitting is stronger
when others perceive there to be many available alternatives in the job market and that men are more likely to quit (or,
women are less likely to quit) when the job market is tighter and fewer jobs are available. Taken together, we conclude
that Hypothesis 3 is partially supported.
Hypothesis 4 concerned the turnover contagion perspective, shown in Table 3, column 5. We hypothesized that
when sample mean-level turnover base rates, withdrawalcognitions, and job search behaviors are higher, antecedent–
turnover relationships would become more positive or less negative. As shown in the table, numerous moderatorswere
signiﬁcant. Speciﬁcally, when an organization’s turnover base rate is higher, age (𝛽=–.46, p<.01) and job satisfaction
(𝛽=–.48, p<.05) are even more strongly negatively related to turnover. When mean withdrawal cognitions are higher,
the positive relationships to turnover of alternatives (𝛽=.33, p<.01) and withdrawal cognitions (𝛽=.55, p<.05)
become even more positive. Similarly, the negative relationships of justice (𝛽=–.54, p<.05) and organizational com-
mitment (𝛽=–.51, p<.01) to turnover become more negative. The negative relationships of performance (𝛽=.78,
p<.05), tenure (𝛽=.46, p<.05), and pay (𝛽=.71, p<.05) become less negative as others think more about leav-
ing. Said differently, although higher performance, longer tenure and higher pay typically keep employees in their jobs,
higher performers, longer tenured employees, and higher paid employees may actually be more likely to quit when oth-
ers around them think about leaving. The moderation effect for job search–turnover ×sample mean-level job search
was also signiﬁcant, being more positive in samples with higher average search behavior (𝛽=.67, p<.05). When many
people are searching for new employment, we ﬁnd that employees are more likely to translate their own job search
behavior into quitting.
Surprisingly,turnover base rates did not moderate relationships for predictors such as commitment, embeddedness,
or withdrawal cognitions. This may partly be a function of limited relative studies available,such as embeddedness (i.e.,
a possible TypeII error), but it also points to an interesting consideration: Possibly, who leaves (i.e., a “bad apple” versus
a supportive colleague) is more important than the raw metric of how many employees leave. In job embeddedness the-
ory, considering the quality of links—beyond mere quantity—could be insightful in this regard (Lee, Burch, & Mitchell,
2014). In general, though the results show that when the work environment is one in which withdrawal, thoughts of
withdrawal, and actual quitting are higher, antecedents such as alternatives, search, and withdrawal cognitions have
even more positive effects, whereas factors that typically prevent turnover (e.g., high performance, longer tenure and
higher pay) have weaker (i.e., closer to zero) effects. We conclude that Hypothesis 4 is generally supported.
Overall, we conclude that the personal ﬁt, job market, and turnover contagion perspectives received the greatest
support. When examining how employees arrive at a possible turnover decision, it seems prudent then to take into
consideration their (dis)similarity to others, perceived or actual job alternative availability, and the extent to which
others in the organization have quit or are signaling their intent to do so. The moderation results present a relatively
RUBENSTEIN ET AL.27
conservative effect in that we did not tease apart contextual nuances. It is possible that the context plays a greater role
than what is shown here, due to the imprecise nature in which we were able to operationalize it. Contextual inﬂuences
should be greater, the more narrowly deﬁned the context becomes. For example, an employee might consider looking
for a new job when immediate coworkers in the same department do the same but would not exhibit such tendencies
when coworkers in a different department engage in search. Our analyses operationalized context at the sample level,
and although we excluded samples of employees from multiple organizations, heterogeneity in each sample’s composi-
tion could be quite high because some spanned multiple departments, business units, teams, or even geographical loca-
tions (though still from the same organization). Meanwhile, other contexts were more narrowly deﬁned. We suspect
that viewing the context in narrower ways would yield more signiﬁcant moderation effects and possibly also provide
more support for the attitudinal climate perspective.
3STEP 3: WHERE ARE WE GOING? AN INTEGRATIVE,
The overarching goal of this study was to assess the current state of the turnover literature, considering the progress
made resulting from the substantial growth of empirical work in the 21st century and what that progress means going
forward. To these ends, we conducted the most comprehensive turnover meta-analysis to date and identiﬁed those
predictors that most meaningfully contribute to individual quitting decisions. Second, guided by theory,we conducted a
series of moderator analyses to gain more insight into the contextual nature of turnover. In this discussion, we consider
the theoretical and practical implications of the results and integrate our ﬁndings by developing a research “road map,”
so to speak, for the future of turnover study. Speciﬁcally, after reviewing hundreds of articles, we feel it is important to
discuss what we believe to be the most necessary conceptual and methodological challenges and opportunities going
forward, so as to improve theory testing and prediction.
3.1 Theoretical implications
A primary goal of this study was to aid in interpreting the vast turnover landscape with an eye toward future research.
Part of this task was accomplished with a thorough an initial meta-analytic review (Step 1). Still, we believe there is
beneﬁt in offering a general picture of these results. In order to summarize all tested relationships, we offer a scatter
plot, shown in Figure 1, which organizes each antecedent according to its absolute corrected meta-analytic correlation,
on the Y-axis, and its corresponding standard error, on the X-axis. The former metric describes effect size magnitude,
whereas the latter describes effect size variability (computed as SD/√N, see Table 2). The number of studies, k,for
each antecedent is also included in Figure 1, distinguished by point marker quartiles. Using this plot as a visual aid, we
seek to identify potentially stronger and weaker areas of inquiry and areas that require further study. Although these
results cannot speak to how any given effect might operate in a multivariate model, when a turnover researcher or
human resources analytics team is considering what variables to include in their study, and survey length constraints
are a factor, Figure 1 is a useful starting point.
A few variables stand out as they are the most predictive and tend to exhibit relatively low variability across con-
texts. These variables are shown in the upper-left area of the ﬁgure and include proximal work perceptions and behav-
iors such as withdrawal cognitions, job search, organizational commitment, job satisfaction, rewards offered beyond
pay, justice, embeddedness, and performance. It also includes distal factors such as age, tenure, and children. Interest-
ingly, a greater number of antecedents in this area of the plot are more proximal rather than distal.
Most of these antecedents also have been researched quite extensively (i.e., 3rd or 4th quartile of study). We
are therefore quite conﬁdent about the robustness of these ﬁndings and would anticipate such variables will be
consistently predictive of quitting. If a holistic account of why people quit were a researcher or practitioner’s goal,
we would recommend that inquiry start with these predictors. Equally, if a researcher seeks to predict turnover with a
28 RUBENSTEIN ET AL.
FIGURE 1 Summary of meta-analytic turnover antecedent estimates (as effect sizes-by-standard errors)
Note. Correlation signs indicated in parentheses. OCB =organizational citizenship behavior.PC breach =psychological
contract breach. Due to visual overlap, we note extraversion, OCBs and organizational support are in the 2nd quartile
of studies (k) accumulated; ethnicity, job involvement, marital status and workload are in the 3rd quartile.
new construct, it would be helpful to ﬁrst incorporate these antecedents as controls or covariates. For example, when
the job embeddedness construct was ﬁrst introduced, Mitchell et al. (2001) demonstrated evidence of its incremental
predictive validity beyond that of similar constructs like job satisfaction and organizational commitment.
Viewing standard errors as a signal of the context-dependent nature of turnover reveals antecedents showing
moderate-to-strong predictive magnitude but also relatively high variability. These antecedents are located toward
the top-right of Figure 1. They include constructs such as ﬁt, climate, coping, other satisfaction, and other commitment.
Thus, although job satisfaction and organizational commitment tend to be predictive across occupations and samples,
more narrow predictors like career satisfaction and occupational commitment might be more context speciﬁc. Also
possible, such variables may emerge as more consistent predictors when matched to their respective target (e.g., occu-
pational commitment predicting occupational turnover; Blau, 2007).
Another inference from the data markers is that many effect sizes must be interpreted with caution. For instance, it
is not especially surprising that coping and reward contingency are outliers, given their limited study.At the same time,
one cannot interpret this ﬁnding as a reason to dismiss their effects entirely, or conversely, to automatically include
them as study controls. Clearly, more empirical work is needed on these variables in order to draw ﬁrmer conclusions.
Noting both ﬁt and climate having both sizable effect sizes and high standard errors speaks to the general conclusion
of our research that the organizational context matters and should be explored in more depth (Johns, 2006). Our mod-
eration tests reported here were aimed to more systematically guide and inspire such efforts. Speciﬁcally, we found
numerous ﬁt-related boundary conditions, revealing that when an employee makes a claim that he or she “ﬁts” with
the company culture, this perception can be made in terms of demographics or attitudes/cognitions, as shown here, as
well as in terms of job demands relative to personal abilities or personal needs relative to job supplies (Kristof-Brown
et al., 2005). Researchers might capitalize on methodological advancements in the form of response surface analysis
and self–other congruence indices to further explore these promising ideas (Edwards, 2007).
RUBENSTEIN ET AL.29
We also note antecedents with higher standard errors, coupled with smaller relative effects. These antecedents are
plotted toward the bottom right of Figure 1. Met expectations, centralization, and openness to experience stand out as
examples. Although these constructs’ effect magnitudes are comparatively lower, their higher variability suggests that
they might be especially impactful in certain contexts yet matter little in others. From both theoretical and practical
perspectives, there is value in identifying what such boundary conditions might be, and why they emerge for a given
antecedent. Unfortunately, we were unable to perform moderator tests for these variables due to insufﬁcient primary
studies, which also makes it challenging to draw ﬁrm conclusions as to population parameter values. For example, late-
ness has a modest magnitude (𝜌=.14) but a relatively larger standard error (SE =.04). Lateness has a longstanding
history in turnover theory but has received surprisingly limited empirical attention. Given the criticality of lateness
behavior to “progression of withdrawal” models (Berry et al., 2012; Harrison et al., 2006), where lateness is viewed as a
minor form of withdrawal, progressing in increasing severity to absenteeism and eventuallyquitting, greater scrutiny of
this mediated process seems warranted. Such models argue that absenteeism fully mediates lateness–turnover rela-
tionships, such that a direct lateness–turnover effect is nonsigniﬁcant. However, such models essentially ignore the
heterogeneity of lateness–turnover direct effects, such that its relationship might indeed matter under particular cir-
cumstances, varying as a function of dispositional factors, the organizational climate, company policies, job market con-
ditions, or contagion pressures.
It is also worthwhile to address antecedents with relatively lower effect magnitudes and lower standard errors. One
might be quick to dismiss these variables; however,we would caution against such hasty value judgments, instead alert-
ing researchers to reﬂect on their necessity as controls and to make theory-driven decisions regarding their use as pre-
dictors. Employee sex, education, and ethnicity all havesmall effect sizes, along with lower standard errors. We noticed
that these demographics are often included in turnover studies, although most are not discussed in detail within the
context of prediction but rather are included as controls. The compiled evidence here suggests that such automatic
practice might not be compulsory. Becker (2005) recommends eliminating controls that have little or no relationship
with outcomes, offering |r| <.10 as a potential cut-off for inclusion. Ultimately, considering one’s research question
and the study context, these variables can sometimes prove useful, and careful thought should go into each variable
modeled—for as Kurt Lewin professed, “nothing is so practical as a good theory” (1945, p. 129). For instance, we see
that sex does actually matter for turnover when considering ﬁt (i.e., the sex makeup of the organization). Yet, it might
be less impactful in other situations. On the other hand, age, tenure, and having children do seem to matter across a
wider breadth of contexts. Controlling for these latter demographics might therefore be more routinely appropriate
(Becker et al., 2016; Bernerth & Aguinis, 2016).
3.2 A road map for future turnover research
So far, this paper has focused on the past and present. We solidiﬁed past ﬁndings, learned about new promising predic-
tors, and highlighted the contextual nature of turnover by ﬁnding support for our ﬁt, job market, and contagion argu-
ments. The ﬁnal goal of this paper was to use the results to identify the most pressing needs for future research. We
present the following 10 recommendations targeted toward study design and broader methodological improvement.
3.2.1 Recommendation #1: Measure voluntary turnover behavior
During our primary article search, it was surprising to ﬁnd a large portion of studies that treated cognitions or inten-
tions to quit as proxies for turnover behavior or studies that failed to distinguish voluntary from involuntary exit. First,
although intentions to engage in a behavior are the best predictor of engaging in that behavior (Ajzen & Fishbein, 1980),
our estimated effect size for withdrawal cognitions and turnover (𝜌=.56) demonstrates that the two are not identical
and should not be treated as such. We would argue that doing so might mislead researchers as to what antecedents
are most predictive (i.e., not all things that predict cognitions similarly predict behavior) and could yield improper con-
clusions if mediation tests model intentions as the endogenous outcome. In a predictive sense, if one were to exam-
ine the relationship between job attitudes and turnover intentions, this effect size would likely decrease as a function
of increased time between measurement, based on common method variance issues (Podsakoff, Mackenzie, Lee, &
30 RUBENSTEIN ET AL.
Podsakoff, 2003). However, if behavior were the outcome, this relationship would likely increase with greater measure-
ment separation (Holtom, Tidd, Mitchell & Lee, 2013), as more employees would have had more time to make a quit
decision. As such, although intentions can indeed be a legitimate research interest—to identify those who might be on
the verge of leaving, in order to intervene—it is fallacious to assume that the presence of one (i.e., thoughts) assures the
other (i.e., behavior). Instead, for the literature to progress, researchers must measure turnover behavior qua turnover
behavior and must test their research questions appropriately (e.g., with hazard modeling, logistic regression, or net-
work analysis, as opposed to ordinary least squares methods; Hom et al., 2017). As we mentioned earlier, there is also
room for more research on why intentions do not lead to action (e.g., Allen, Weeks, & Mofﬁtt, 2005; Vardaman, Taylor,
Allen, Gondo, & Amis, 2015).
3.2.2 Recommendation #2: Pursue combinational approaches
Merely studying a motley collection of predictors does not capture the likely complexity among relationships. As
such, we recommend that researchers emphasize predictor combinations in more thoughtful, theory-driven ways. For
instance, Hom et al. (2012) proposed a combinational approach by introducing the idea of “proximal withdrawalstates.”
They proposed four proximal withdrawalstates derived from two dimensions: the employee’s desire for leaving or stay-
ing and the employee’s perceived control over this decision (whether they can freely act on their desire or are bound
by external factors). Crossing these two dimensions leads to four groups of employees: those who want to leave and
can (enthusiastic leavers), those who want to leave but think they cannot (reluctant stayers), those who want to stay
and can (enthusiastic stayers), and those who want to stay but think they cannot (reluctant leavers). Although empirical
work has begun to test this theory (Li, Lee, Mitchell, Hom, & Griffeth, 2016), such ideas are a compelling example of the
kinds of combinational approaches the literature needs to pursue in order to increase prediction precision. Alterna-
tively, one could consider creating employee proﬁles: For example, Stanley, Vandenberghe, Vandenberg, and Bentein
2013 developed proﬁles based on employee reports of affective, normative, and continuance commitment, found this
to improve turnover prediction (see also Maertz & Campion, 2004).
3.2.3 Recommendation #3: Augment the standard predictive design
In the 1970s, the turnover literature made a signiﬁcant leap by adopting what was later called the “standard research
design” (Hom et al., 2017; Steel, 2002). Since then, the typical turnover study measures predictors at Time 1 and
turnover at Time 2, typically with a 3- to 12-month lag (the mode study lag time we observed was 12 months). This
design has advantages in terms of temporal separation to assess causality but also has drawbacks: By not capturing
what happens between measurement points, data useful to prediction are foregone (e.g., attitudinal shifts, disruptive
events, unexpected job offers). Furthermore, by measuring most all predictors at Time 1, one cannot meaningfully
depict the process of how individual turnover unfolds—that is, the process of accumulating dissatisfaction, lessening
commitment and embeddedness, searching for alternatives, and other withdrawal feelings and behaviors. Although
our results help identify what predictors hold promise in a general sense, the standard predictive design constrains
many conclusions that can be gleaned from reality. With the next three recommendations, we therefore broadly pro-
pose some methodological shifts so that future meta-analytic work can move to more complex multilevel path models.
3.2.4 Recommendation #4: Capture turnover processes in real time to establish temporal order
Considering that most studies measure predictors at one time and turnover at another, the ﬁeld is not currently in a
strong position to draw deﬁnitive conclusions regarding the temporal nature of exit as it unfolds. We appear to be in
need of studies that measure multiple turnover-related predictors at multiple measurement points to establish the
proper temporal sequence of variables, to and more precisely ascertain the complete chain of employee withdrawal.
For example, a study that simultaneously measures the most predictive proximal constructs—cognitions, satisfaction,
commitment, embeddedness, search, lateness, absenteeism, and alternatives—at multiple times would be a great start-
ing point to determine which sequence ﬁts the data best (and also perhaps assess how and why this might vary between
individuals). Such a study may not seem very sexy because it would not introduce any new constructs, but a clearly
RUBENSTEIN ET AL.31
deﬁned temporal model that competitively tests theoretical sequences could offer signiﬁcant insight into how these
mediators interrelate and interact (for example, do employees engage in job search and lateness simultaneously, does
one precede the other, for whom does either apply, when does it apply, and why?). Such knowledge would also increase
our ability to develop more precise practical counsel regarding the avoidance of unwanted turnover in the workplace.
3.2.5 Recommendation #5: Measure events
The typical predictive design of assessing predictors at Time 1 and turnover at Time 2 does not allow one to capture
key eventsthat may occur between measurement periods. Increasing the number of measurement points, coupled with
qualitative or quantitative inquiry about events, can help alleviatethis concern (cf. Morgeson, Mitchell, & Liu, 2015). For
example, the unfolding model of turnover suggests that shocks signiﬁcantly alter how employees think about attach-
ment to an employer. Hale, Ployhart, and Shepherd (2015) recently showed how speciﬁc individual employee and/or
manager turnover events can negatively affect the collective performance of bank branches, as turnover disrupts a
unit’s core processes, and the branch must subsequently recover by rebuilding its lost social and human capital. Relat-
edly, Ballinger, Lehman, and Schoorman (2010) found that if a leader were to quit, subordinates who had high-quality
exchange relationships with that leader would be more likely to quit themselves after the succession event, although
they would be more likely to stay if no leader succession event had occurred (see also Ballinger & Schoorman, 2007;
Shapiro, Hom, Shen, & Agarwal, 2016). Given these ﬁndings, those who report high satisfaction at Time 1 may sud-
denly quit due to unforeseen events (e.g., unsolicited offers, health concerns, pregnancy, turnover of a close colleague,
manager turnover, etc.) that occur subsequent to that assessment. However, the majority of current work would not
capture such dynamics, thereby limiting explained turnover variance.
3.2.6 Recommendation #6: Examine predictor change over time
Whereas Recommendation #4 was a call to examine various turnover-related variables over time in order to improve
the accuracy of temporal sequences regarding how quitting unfolds, Recommendation #6 is an explicit call for
researchers to continue to examine how predictors change overtime (i.e., growth, decline, or stagnancy). Most research
providing the foundation for this meta-analysis has taken a static approach, where turnover at Time 2 is regressed on
an individual’s predictor level at Time 1 (Steel, 2002). Such an approach assumes that predictors remain largely sta-
ble over time. Yet, research has established that this assumption is ﬂawed, many variables do ﬂuctuate signiﬁcantly,
and such changes are systematically meaningful (Kammeyer-Mueller, Wanberg, Glomb, & Ahlburg, 2005). Chen, Ploy-
hart, Thomas, Anderson, and Bliese (2011) found that job satisfaction and turnover intentions both changed over time
for many participants. Similarly, Hausknecht, Sturman, and Roberson (2011) found employee justice perceptions var-
ied over the course of just one year. Even seemingly stable dispositions may undergo change (Judge, Simon, Hurst, &
Kelley, 2014; Wille, Hofmans, Feys, & De Fruyt, 2014). Thus, adopting a dynamic view and accounting for trajectories
(i.e., using latent growth modeling) can signiﬁcantly increase turnover variance explained. Notably, Liu et al. (2012)
calculated employee job satisfaction trajectories, incorporating reports at three time points. Such trajectories pre-
dicted turnover after controlling for average levels. Bentein, Vandenberg, Vandenberge, and Stinglhamber (2005) also
found that employees experiencing steeper declines over time in their organizational commitment were more likely
to quit. We encourage future researchers to build on such results. For instance, studies might examine whether and
how other predictors (e.g., absenteeism, job search, job embeddedness, work–life conﬂict) exhibit similar change over
time, and whether this change predicts quitting. Equally, the degree of change over time is also worth consideration:
More measurement occasions at shorter intervals could enhance precision as to just how much some predictors ﬂuc-
tuate compared to others. For example, whereas Hausknecht et al. (2011) looked at justice change on the order of one
year, Matta, Scott, Colquitt, Koopman, and Passantino (in press) found justice perceptions to vary within just a 3-week
period. Researchers might also consider the interaction of mean predictor levels and predictor change over time in
predicting exit. It is possible that before deciding to quit, an employee reference past changes over time in his or her
attitudes (i.e., as having improved, worsened, or stayed the same) as a means of putting present attitude levels into
context. Such an interaction might account for unique variance beyond mean levels or trajectories alone.
32 RUBENSTEIN ET AL.
3.2.7 Recommendation #7: Increase attention to additional antecedent–turnover moderators
We strongly encourage researchers to systematically continue to examine predictors with high effect variability (i.e.,
SE ≥.03), such as ﬁt, climate, and job security. For example, for ﬁt, Vidyarthi, Erdogan, Anand, Liden, and Chaudhry
(2014) found that for employees with two leaders, satisfaction and turnover were inﬂuenced by the ﬁt (mis)alignment
in relationships that employees have with their two leaders. For many of these predictors, simply an increase in
the quantity of studies can be illuminating insofar as additional data bolsters conﬁdence about the stability (or
fragility) of effect sizes across contexts. It would also be valuable to explore the interactive effects of these predic-
tors with other turnover-relevant concepts. For example, there is evidence that locus of control moderates the with-
drawal cognitions-turnover link such that those with an external locus are more likely to turn thoughts into behavior
(Allen et al., 2005).
3.2.8 Recommendation #8: Scrutinize contextual multilevel inﬂuences
We noted earlier that many predictors capture contextual and social inﬂuences on employee turnover. Constructs
such as peer relations, climate, and leadership highlight how others’ behaviors and attitudes can inﬂuence how
employees interpret their own work situation. Plus, our moderation analyses provide support for the general notion
that context can attenuate or amplify certain variables, especially as it highlights personal ﬁt, the job market, and
workforce withdrawal tendencies. With advancements in multilevel analytical techniques, we are now in a posi-
tion to model the inﬂuence of social networks, other-rated perceptions, and meso/macro-contextual indicators of
turnover. A compelling example of this type of research comes from Liu et al. (2012), who not only examined
individual-level satisfaction trajectories but also unit-level satisfaction agreement effects. We advocate for future
research to take similar multilevel approaches and possibly even combine them with other recommendations we have
3.2.9 Recommendation #9: Test meta-analytic moderators within the scope of the data
Although we intended to be as comprehensive as possible, it was also important to only examine moderators with suf-
ﬁciently accumulated effects so as to draw reliable conclusions (i.e., to limit Type I and Type II error rates). Ultimately,
this issue speaks to broader concerns about moderator testing in meta-analysis, and for the turnover literature in par-
ticular: Due to the nature of meta-analytic data, moderation tests are limited in the types of questions that can be
answered. If moderation is an interaction among two or more variables in predicting an outcome, a primary study would
need only create a product between an independent variable and moderator (or moderators), and regress that outcome
on the main effects and product (Cohen, Cohen, West, & Aiken, 2013). This is relatively straightforward because data
from each respondent on each variable are usually available (barring missing responses). With meta-analysis, however,
one must rely on summary data (e.g., correlations, sample averages) not individual data points. As such, this regres-
sion approach cannot be performed in a meta-analysis unless correlations are provided for both main effects–outcome
and interactions–outcome—and they rarely are. Thus, meta-analytic interactions must be computed differently. One
alternative is to correlate each study’s moderator level with a respective predictor–outcome effect size; another is to
separate effect sizes into subgroups, such as with results from individualistic versus collectivistic cultures (Choi, Oh, &
Colbert, 2015), public versus private organizations (Jiang et al., 2012), military versus civilian samples (Griffeth et al.,
2000), or others. Yet, there are problems with these two methods, particularly with regard to not appropriately weigh-
ing studies by sample size, as well as statistical power concerns (Aguinis, Beaty, Boik, & Pierce, 2005; Aguinis et al.,
2013; Steel & Kammeyer-Mueller, 2002).
To remedy these issues, we recommend that researchers test for meta-analytic moderators using a WLS regres-
sion approach as done here and in other recent meta-analyses (e.g., Heavey, Holwerda, & Hausknecht, 2013), and to
limit moderation tests to only those variables on which a sufﬁcient number of primary studies are based. At the time
of this writing, the subgroup method seems to be most popular approach. Yet, many such subgroup tests are based
off of an already-limited number of studies, so dividing them further weakens our conﬁdence in true effect size dif-
ferences, because lower-k estimates have wider conﬁdence intervals that are more likely to overlap with the other
RUBENSTEIN ET AL.33
subgroup (Quiñones, Ford, & Teachout, 1995). Further, such tests make little sense if one subgroup is disproportion-
ately represented. Should Subgroup A have 14 studies and Subgroup B only 1 study, it becomes difﬁcult to conﬁ-
dently say that the two subgroup effect sizes truly differ, even if a Z-test is signiﬁcant. This realization led us to a ﬁnal
3.2.10 Recommendation #10: Further study underexplored cultures and occupations
Most turnover studies we reviewed came from Western countries (e.g., the United States and Canada), with fewer
explorations in other cultures (e.g., East Asia, Africa, Latin America and South America) that likely have different
norms and construct meanings. With more studies in such contexts, we might see different estimates. For instance,
in China and other paternalistic cultures, leadership often extends beyond consideration, including benevolent behav-
iors such as helping employees with problems at home (Chen, Eberly, Chiang, Farh, & Cheng, 2014). As such, lead-
ership might more strongly affect turnover in such cultures. Yet, with only 7 out of 42 leadership-turnover studies
from outside the United States, it is difﬁcult to determine just how much they differ. Similarly, only 4 out of 55 stud-
ies on rewards besides pay were conducted outside North America. Perhaps this effect is downwardly biased, and
would be stronger with more international samples, for instance, given the more generous beneﬁts mandates in Europe
compared to the United States, like paid maternity/paternity/sick leave and health insurance (Glassdoor Economic
Research, 2016). Even among these Western samples, though, we also noticed a high degree of occupational homo-
geneity, with most studies coming from civilians in private organizations performing white-collar or hospital work. For
the pan-occupational and cross-cultural generalizability of our results to hold, it is imperative that future work focuses
on turnover prediction in other cultures and occupations, to determine if, when, and how certain constructs operate
differently, but most importantly, why they do so.
3.3 Practical implications
Organizations are rightfully interested in curtailing unexpected or unwanted turnover to protect social capital and
organizational memory, and to reduce sizable expenses of onboarding newcomers (Allen et al., 2010). Our results offer
useful insights for the practitioner community.
Allen et al. (2010) discussed how managers commonly believe that employees quit in order to take higher paying
positions elsewhere. We ﬁnd that the relationship between pay and turnover has increased since Griffeth et al. (2000),
suggesting that the inﬂuence of pay on exit decisions is perhaps stronger today than it was two decades ago. Yet, as
with the Griffeth et al. ﬁndings, many other predictors more readily controlled by managers can be more important
than pay. One such factor includes rewards besides pay (e.g., training or promotional opportunities, bonuses, and non-
cash beneﬁts). Other predictors include job characteristics, leadership, climate, and organizational support. A prevail-
ing thought many managers hold about turnover is that it is mostly due to dissatisfaction with the work itself or low pay,
and because “the job is the job,” rarely can anything be done to remedy high quit rates (or, they simply accept them as
bearable). To the contrary, our results corroborate the notion that often, “employees quit bosses, not jobs,” and that at
least as much, turnover can be due to toxic work climates or feeling unsupported by the organization. In these respects,
we would argue that leadership development programs might be especially valuable to retention. Such efforts might
not just focus on how leaders can develop strong relationships with followers but also on how leaders can serve as
a bridge between subordinates and higher-up organizational stakeholders, as well as on how to build a climate that
leverages the idiosyncratic strengths of the company (Hackman & Wageman, 2007).
Equally important are the relatively strong effects for withdrawal attitudes, job marketperceptions, and withdrawal
behaviors. In this regard, gauging ratings on such predictors would be highly informative. Problematically, though, many
employees might be reluctant to share such revealing information with their organizations, and any responses they
do share might be biased so as to appear socially desirable. Some viable options to obtain such data might include
ensuring response anonymity, obtaining other reports of these perceptions/behaviors, bringing in neutral consultants
to administer surveys, cultivating a work environment where employeesknow they will not be reprimanded for sharing
34 RUBENSTEIN ET AL.
opinions, or by using unobtrusive measures like absenteeism rates or monitoring how employees use company time
(Kerlinger & Lee, 2000).
Another major focus of retention efforts in today’s organizations is work engagement (Bersin, 2014; Graber, 2015).
Companies like Gallup and Deloitte continuously update managers about employee engagement levels: One survey
found that 79% of businesses are seriously worried about engagement and retention (Adkins, 2016; Deloitte Con-
sulting Group, 2014). Employee engagement is a relatively new phenomenon within the turnover literature, and we
could only identify four primary studies linking it to turnover behavior. Although we must be careful about overin-
terpreting this effect based on the limited evidence, engagement does appear to be a useful predictor. Thus, efforts
to increase engagement would likely be valuable. Speciﬁcally, the major psychological drivers of engagement include
experienced meaningfulness of work, psychological safety of the work environment, and availability to engage one’s
personal resources at any given moment (Crawford, Rich, Buckman, & Bergeron, 2014; Kahn, 1990). We identiﬁed
a host of antecedents that would align with these drivers, such as improving job characteristics and rewards/pay to
improve meaningfulness, monitoring leader behaviors, climate, justice, and job security to facilitate safety, and a (lack
of) work–family conﬂict, role conﬂict, or stress to sustain availability. To the degree that managers attend to one or
more of these drivers, we would expect engagement to improve, along with subsequent employee retention rates.
It is also prudent to discuss what our results mean for employee selection. Of course, due to equal employment
opportunity concerns, we cannot advise organizations to select individuals based on their age, marital status, or how
many children they have. However, other prehire predictors may be quite useful to curb turnover. For instance, man-
agers might evaluate job applicants’ perceived ﬁt with the organization in terms of values and personality. Disposi-
tionally, motivational metrics such as performance efﬁcacy or personal goal setting would also be promising, and we
would echo Zimmerman’s (2008, p. 335) claim to the utility of personality inventories focusing on emotional stabil-
ity and conscientiousness (as well as CSEs, as noted earlier). Other studies (e.g., Barrick & Zimmerman, 2005) have
also advocated for using biographical data for selection. We concur, speciﬁcally for things like past job tenure. Given
that companies often hire hundreds of new employees each year—even more for those with regularly high turnover
rates—differentiating applicants even by one point on these metrics could meaningfully reduce turnover, resulting in
substantial cost savings. There have also been efforts to predict commitment propensity and quit intentions during
selection (Lee, Ashford, Walsh, & Mowday, 1992), so similar steps could be taken to select individuals prone to higher
attitudes or embeddedness (Barrick & Zimmerman, 2005; Choi et al., 2015; Judge, Heller, & Mount, 2002).
There are limitations to this paper that must be acknowledged. First, treating dichotomous turnover behavior as con-
tinuous might be considered controversial, as correcting correlations can lead to an increase in sampling error variance
of the adjusted meta-analytic effect distribution (Hunter & Schmidt, 2004). As such, this procedure may have inﬂated
the results. We would advise readers to consider the uncorrected and corrected correlations when interpreting the
Second, as a meta-analysis, our results may be biased in favor of the most studied constructs, despite newer, equally
predictive variables being overshadowed. Although we did try to incorporate manyinto our model, this article is not the
ﬁnal word as to what employees consider when they decide to quit. Exciting developments in the areas of dyadic work
relationships (Chiaburu & Harrison, 2008), commitment reconceptualized as identiﬁcation or internalization (Klein,
Molloy, & Brinsﬁeld, 2012), and emotional labor (Kammeyer-Mueller et al., 2013) may prominently ﬁgure into future
discussions of employee turnover and empirical study.
Third, and as noted in Recommendations 9 and 10, although we aimed to be as comprehensive as possible in testing
moderation hypotheses, we were unable to examine every moderation relationship, and some tests still had low power,
despite inclusion restrictions. Thus, although many predictors showed signiﬁcant effect size variability, we were only
able to test hypotheses on a selection of them. Future studies should expand our moderation tests to the full range of
predictors and study how results might differ as a function of other contextual factors.
RUBENSTEIN ET AL.35
This paper reports the most comprehensive analysis of the individual-level voluntary turnover literature to date. In
surveying this dense forest, we revealed an array of distal and proximal factors that contribute to exit decisions,
while highlighting the context-sensitive nature of this phenomenon. Given the results, as well as the challenges in the
existing literature uncovered throughout this work, we advocate for a paradigm shift in turnover research that
embraces cutting-edge methodologies to capture the dynamic and multilevel ways by which turnover decisions unfold.
It is our intention that this study not be a conclusive statement to turnover study, but rather a checkpoint, to take stock
of where we have been and to offer a practical guide as to the most promising avenues for future inquiry.
We gratefully acknowledge the constructive feedback from editor Dr. John Hausknecht and two anonymous reviewers
on earlier versions of this manuscript.
1From this point forward, when we use the term turnover, we refer to individual-level voluntary turnover.
2In these analyses, we did not consider multiorganization samples, unless the article separately reported mean levels of a vari-
able for each organization studied. Employees are likelyonly inﬂuenced by their own organizational context, such as the mean
level of job satisfaction of their own organization. Mean job satisfaction across multiple organizational samples is therefore
less meaningful to examine.
3Due to the substantial number of studies (k=316) included in the meta-analysis, we only include those references that are
cited in text. The full coded article list is available upon request from the ﬁrst author.
Adkins, A. (2016). Employee engagement in U.S. stagnant in 2015. Retrieved from http://www.gallup.com/poll/188144/empl-
Aguinis, H., Beaty, J. C., Boik, R. J., & Pierce, C. A. (2005). Effect size and power in assessing moderating effects
of categorical variables using multiple regression: A 30-year review. Journal of Applied Psychology,90, 94–107.
Aguinis, H., Gottfredson, R. K., & Culpepper, S. A. (2013). Best-practice recommendations for estimating cross-level interaction
effects using multilevel modeling. JournalofManagement,39, 1490–1528. https://doi.org/10.1177/0149206313478188
Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Englewood Cliffs, NJ: Prentice-Hall.
Allen, D. G., & Griffeth, R. W. (2001). Test of a mediated performance-turnover relationship highlighting
the moderating roles of visibility and reward contingency. Journal of Applied Psychology,86, 1014–1021.
Allen, D. G., Bryant, P. C., & Vardaman, J. (2010). Retaining talent: Replacing misconceptions with evidence-based strategies.
Academy of Management Perspectives,24, 48–64. https://doi.org/10.5465/AMP.2010.51827775
Allen, D. G., Shore, L. M., & Griffeth, R. W. (2003). The role of perceivedorganizational support and supportive human resource
practices in the turnover process. JournalofManagement,29, 99–118. https://doi.org/10.1177/014920630302900107
Allen, D. G., Peltokorpi, V. M., & Rubenstein, A. L. (2016). When “embedded” means “stuck”: Moderating
effects of job embeddedness in adverse work environments. Journal of Applied Psychology,101, 1670–1686.
Allen, D. G., Weeks, K. P., & Mofﬁtt, K. R. (2005). Withdrawal cognitions and voluntary turnover: The moderating roles of
self-monitoring, locus of control, proactive personality, and risk aversion. Journal of Applied Psychology,90, 980–990.
American Psychological Association (2012). Stress in America. Retrieved from http://www.apa.org/news/press/releases/stress/
Arnold, H. J., & Feldman, D.C. (1982). A multivariate analysis of the determinants of job turnover. Journal of Applied Psychology,
67, 350–360. https://doi.org/10.1037/0021-9010.67.3.350
36 RUBENSTEIN ET AL.
Bakker, A. B., & Leiter, M. P. (Eds.). (2010). Work engagement: A handbook of essential theory and research. New York, NY: Psychol-
Bal, P. M., & Jansen, P. G. W. (2016). Work ﬂexibility across the lifespan. Research in Personnel and Human Resources Management,
Ballinger, G. A., Lehman, D. W., & Schoorman, F.D. (2010). Leader–member exchange and turnover before and after succession
events. Organizational Behavior and Human Decision Processes,113, 25–36. https://doi.org/10.1016/j.obhdp.2010.04.003
Ballinger, G. A., & Schoorman, F.D. (2007). Individual reactions to leadership succession in workgroups. Academy of Management
Review,32, 118–136. https://doi.org/10.5465/AMR.2007.23463887
Barrick, M. R., & Mount, M. K. (1996). Effects of impression management and self-deception on the predictive validity of per-
sonality constructs. Journal of Applied Psychology,81, 261–272. https://doi.org/10.1037/0021-9010.81.3.261
Barrick, M. R., & Zimmerman, R. D. (2005). Reducing voluntary, avoidable turnover through selection. Journal of Applied Psychol-
ogy,90, 159–166. https://doi.org/10.1037/0021-9010.90.1.159
Barrick, M. R., & Zimmerman, R. D. (2009). Hiring for retention and performance. Human Resource Management,48, 183–206.
Becker, T. E. (2005). Potential problems in the statistical control of variables in organizational research: A qualitative
analysis with recommendations. Organizational Research Methods,8, 274–289. https://doi.org/10.1177/109442810527-
Becker, T. E., Atinc, G., Breaugh, J. A., Carlson, K. D., Edwards, J. R., & Spector, P. E. (2016). Statistical control in correlational
studies: 10 essential recommendations for organizational researchers. Journal of Organizational Behavior,37, 157–167.
Benson, G. S., Finegold, D., & Mohrman, S. A. (2004). You paid for the skills, now keep them: Tuition reimbursement and volun-
tary turnover. Academy of Management Journal,47, 315–331.
Bentein, K., Vandenberghe, C., Vandenberg, R., & Stinglhamber, F. (2005). The role of change in the relationship
between commitment and turnover: a latent growth modeling approach. Journal of Applied Psychology,90, 468–482.
Bernerth, J. B., & Aguinis, H. (2016). A critical review and best-practice recommendations for control variable usage. Personnel
Psychology,69, 229–283. https://doi.org/10.1111/peps.12103
Berry, C. M., Lelchook, A. M., & Clark, M. A. (2012). A meta-analysis of the interrelationships between employee lateness,
absenteeism, and turnover: Implications for models of withdrawal behavior. Journal of Organizational Behavior,33, 678–
Bersin, J. (2014). It’s time to rethink the “employee engagement” issue. Retrieved from http://www.forbes.com/sites/
Blau, G. (1989). Testing the generalizability of a career commitment measure and its impact on employee turnover. Journal of
Vocational Behavior,35, 88–103.
Blau, G. (1994). Developing and testing a taxonomy of lateness behavior. Journal of Applied Psychology,79, 959–970.
Blau, G. (2007). Does a corresponding set of variables for explaining voluntary organizational turnover transfer to explain-
ing voluntary occupational turnover? JournalofVocationalBehavior,70, 135–148. https://doi.org/10.1016/j.jvb.2006.
Blau, G. J. (1987). Locus of control as potential moderator of the turnover process. Journal of Occupational Psychology,60, 21–
Blegen, M. A., Mueller, C. W., & Price, J. L. (1988). Measurement of kinship responsibility for organizational research. Journal of
Applied Psychology,73, 402–409. https://doi.org/10.1037/0021-9010.73.3.402
Bluedorn, A. C. (1982). A uniﬁed model of turnover from organizations. Human Relations,35, 135–153.
Borenstein, M., Hedges, L. V., Higgins, J.P. T., & Rothstein, H. R. (2009). Introduction to meta-analysis. Chichester, UK: Wiley.
Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2010). A basic introduction to ﬁxed-effect and random-effects
models for meta-analysis. Research Synthesis Methods,1, 97–111. https://doi.org/10.1002/jrsm.12
Boswell, W. R., Boudreau, J. W., & Dunford, B. B. (2004). The outcomes and correlates of job search objectives:
Searching to leave or searching for leverage? Journal of Applied Psychology,89, 1083–1091. https://doi.org/
RUBENSTEIN ET AL.37
Boswell, W. R., Boudreau, J. W., & Tichy, J. (2005). The relationship between employee job change and job satisfaction: The
honeymoon-hangover effect. Journal of Applied Psychology,90, 882–892. https://doi.org/10.1037/0021-9010.90.5.882
Bretz, R. D., Boudreau, J. W., & Judge, T. A. (1994). Job search behavior of employed managers. Personnel Psychology,47, 275–
Brief, A. P., Van Sell, M., & Aldag, R. J. (1979). Vocational decision making among women: Implications for organizational behav-
ior. Academy of Management Review,4,521–530. https://doi.org/10.5465/AMR.1979.4498329
Buchko, A. A. (1992). Employee ownership, attitudes, and turnover: An empirical assessment. Human Relations,45, 711–733.
Bunderson, J. S. (2001). How work ideologies shape the psychological contracts of professional employees: Doctor’s responses
to perceived breach. Journal of Organizational Behavior,22, 717–741. https://doi.org/10.1002/job.112
Cable, D. M., & DeRue, D. S. (2002). The convergent and discriminant validity of subjective ﬁt perceptions. Journal of Applied
Psychology,87, 875–884. https://doi.org/10.1037/0021-9010.87.5.875
Cappelli, P., & Sherer, P. D. (1991). The missing role of context in OB: The need for a mesolevel approach. In L. L. Cummings & B.
M. Staw (Eds.), Research in organizational behavior (Vol. 13, pp. 55–110). Greenwich, CT: JAI Press.
Carr, J. Z., Schmidt, A. M., Ford, K., & DeShon, R. P. (2003). Climate perceptions matter: A meta-analytic path analysis relating
molar climate, cognitive and affective states, and individual level work outcomes. Journal of Applied Psychology,88, 605–
Cavanaugh, M. A., Boswell, W. R., Roehling, M. V., & Boudreau, J. W. (2000). An empirical examination of self-reported work
stress among U.S. managers. Journal of Applied Psychology,85, 65–74. https://doi.org/10.1037/0021-9010.85.1.65
Chan, D. (1998). Functional relations among constructs in the same content domain at different levels of analysis: A typology
of composition models. Journal of Applied Psychology,83, 234–242. https://doi.org/10.1037/0021-9010.83.2.234
Chang, C.-H., Rosen, C. C., & Levy, P. E. (2009). The relationship between perceptions of organizational politics and
employee attitudes, strain, and behavior: A meta-analytic examination. Academy of Management Journal,52, 779–801.
Chen, G., Ployhart, R. E., Thomas, H. C., Anderson, N., & Bliese, P. D. (2011). The power of momentum: A new model of dynamic
relationships between job satisfaction change and turnover intentions. Academy of Management Journal,54, 159–181.
Chen, X., Hui, C., & Sego, D. J. (1998). The role of organizational citizenship behavior in turnover: Conceptualization and pre-
liminary tests of key hypotheses. Journal of Applied Psychology,83, 922–931. https://doi.org/10.1037/0021-9010.83.6.922
Chen, X. P., Eberly, M. B., Chiang, T. J., Farh, J. L., & Cheng, B. S. (2014). Affective trust in Chinese leaders linking paternalistic
leadership to employee performance. JournalofManagement,40, 796–819. https://doi.org/10.1177/0149206311410604
Chiaburu, D. S., & Harrison, D. A. (2008). Do peers make the place? Conceptual synthesis and meta-analysis of
coworker effects on perceptions, attitudes, OCBs, and performance. Journal of Applied Psychology,93, 1082–1103.
Choi, D., Oh, I.-S., & Colbert, A. E. (2015). Understanding organizational commitment: A meta-analytic examination
of the roles of the ﬁve-factor model of personality and culture. Journal of Applied Psychology,100, 1542–1567.
Clausen, T., & Borg, V. (2010). Do positive work-related states mediate the association between psychosocial work
characteristics and turnover? A longitudinal analysis. International Journal of Stress Management,17, 308–324.
Clegg, C. W.(1983). Psychology of employee lateness, absence, and turnover: A methodological critique and an empirical study.
Journal of Applied Psychology,68, 88–101. https://doi.org/10.1037/0021-9010.68.1.88
Cohen, A. (2000). The relationship between commitment forms and work outcomes: A comparison of three models. Human
Relations,53, 387–417. https://doi.org/10.1177/0018726700533005
Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2013). Applied multiple regression/correlation analysis for the behavioral sciences.
New York, NY: Routledge.
Colarelli, S. M., Dean, R. A., & Konstans C. (1987). Comparative effects of personal and situational inﬂuences on job outcomes
of new professionals. Journal of Applied Psychology,72, 558–566. https://doi.org/10.1037/0021-9010.72.4.558
Crawford, E. R., Rich, B. L., Buckman, B., & Bergeron, J. (2014). The antecedents and drivers of employee engagement. In C.
Truss, R. Delbridge, K. Alfes, A. Shantz, & E. Soane (Eds.), Employee engagement in theory and practice (pp. 57–81). New York,
38 RUBENSTEIN ET AL.
de Croon, E. M., Sluiter, J. K., Blonk, R. W. B., Broersen, J. P. J., & Frings-Dresen, M. H. W. (2004). Stressful work, psychological
job strain, and turnover: A 2-year prospective cohort study of truck drivers. Journal of Applied Psychology,89, 442–454.
Deloitte Consulting. (2014). Global human capital trends 2014. In Engaging the 21st Century workforce. Deloitte Uni-
versity Press Accessed January 3, 2017 at http://d2mtr37y39tpbu.cloudfront.net/wp-content/uploads/2014/03/
Dess, G. G., & Shaw, J. D. (2001). Voluntary turnover, social capital, and organizational performance. Academy of Management
Review,26, 446–456. https://doi.org/10.5465/AMR.2001.4845830
Edwards, J. R. (2007). Polynomial regression and response surface methodology. In C. Ostroff & T. A. Judge (Eds.), Perspectives
on organizational ﬁt (pp. 361–372). San Francisco, CA: Jossey-Bass.
Eisenbeiss, S. A., van Knippenberg, D., & Boerner, S. (2008). Transformational leadership and team innovation: Integrating team
climate principles. Journal of Applied Psychology,93, 1438–1446. https://doi.org/10.1037/a0012716
Elvira, M. M., & Cohen, L. E. (2001). Location matters: A cross-level analysis of the effects of organizational sex composition on
turnover. Academy of Management Journal,44, 591–605. https://doi.org/10.2307/3069373
Erdogan, B., & Bauer, T. N. (2010). Differentiated leader-member exchanges: The buffering role of justice climate. Journal of
Applied Psychology,95, 1104–1120. https://doi.org/10.1037/a0020578
Felps, W., Mitchell, T. R., Hekman, D. R., Lee, T. W., Holtom, B. C., & Harman, W. S. (2009). Turnover contagion: How
coworkers’ job embeddedness and job search behaviors inﬂuence quitting. Academy of Management Journal,52, 545–561.
Glassdoor Economic Research (2016). Where are the best social beneﬁts in Europe? New report from Glassdoor Economic Research.
Unpublished research report.
Graber, S. (2015). The two sides of employee engagement. Retrieved from https://hbr.org/2015/12/the-two-sides-of-
Graen, G. B., Liden, R. C., & Hoel, W.(1982). Role of leadership in the employee withdrawal process. Journal of Applied Psychology,
67, 868–872. https://doi.org/10.1037/0021-9010.67.6.868
Griffeth, R. W., Hom, P. W., & Gaertner, S. (2000). A meta-analysis of antecedents and correlates of employee turnover:
Update, moderator tests, and research implications for the next millennium. Journal of Management,26, 463–488.
Griffeth, R. W., Steel, R. P., Allen, D. G., & Bryan, N. (2005). The development of a multidimensional measure
of job market cognitions: The Employment Opportunity Index (EOI). Journal of Applied Psychology,90, 335–349.
Hackman, J. R., & Wageman, R. (2007). Asking the right questions about leadership. American Psychologist,62, 43–47.
Hale, D., Ployhart, R., & Shepherd, W. (2015). A two-phase longitudinal model of a turnover event: Disruption,
recovery rates, and moderators of collective performance. Academy of Management Journal,59, 906–929.
Hancock, J. I., Allen, D. G., Bosco, F. A., McDaniel, K. R., & Pierce, C. A. (2013). Meta-analytic review of employee turnover as a
predictor of ﬁrm performance. Journal of Management,39, 573–603. https://doi.org/10.1177/014920631142-4943
Harris, K. J., Wheeler, A. R., & Kacmar, M. (2011). The mediating role of organizational job embeddedness in the LMX-outcomes
relationships. The Leadership Quarterly,22, 271–281. https://doi.org/10.1016/j.leaqua.2011.02.003
Harrison, D. A., Newman, D. A., & Roth, P. L. (2006). How important are job attitudes? Meta-analytic compar-
isons of integrative behavioral outcomes and time sequences. Academy of Management Journal,49, 305–325.
Hausknecht, J. P., Sturman, M. C., & Roberson, Q. M. (2011). Justice as a dynamic construct: Effects of individual trajectories on
distal work outcomes. Journal of Applied Psychology,96, 872–880. https://doi.org/10.1037/a0022991
Hausknecht, J. P., Trevor, C. O., & Farr, J. L. (2002). Retaking ability tests in a selection setting: Implica-
tions for practice effects, training performance, and turnover. Journal of Applied Psychology,87, 243–254.
Heavey, A. L., Holwerda, J. A., & Hausknecht, J. P. (2013). Causes and consequences of collective turnover: A meta-analytic
review. Journal of Applied Psychology,98, 412–453. https://doi.org/10.1037/a0032380
Higgins, J. P., Thompson, S. G., Deeks, J. J., & Altman, D. G. (2003). Measuring inconsistency in meta-analyses. British Medical
Journal,327, 557–560. https://doi.org/10.1136/bmj.327.7414.557
RUBENSTEIN ET AL.39
Hollenbeck, J. R., & Williams, C. R. (1986). Turnover functionality versus turnover frequency: A note on work attitudes and
organizational effectiveness. Journal of Applied Psychology,71, 606–611. https://doi.org/10.1037/0021-9010.71.4.606
Holtom, B. C., & Inderrieden, E. J. (2006). Integrating the unfolding model and job embeddedness model to better understand
voluntary turnover. Journal of Managerial Issues,18, 435–452.
Holtom, B. C., Mitchell, T. R., Lee, T. W., & Eberly, M. B. (2008). Turnover and retention research: A glance at the
past, a closer review of the present, and a venture into the future. Academy of Management Annals,2, 231–274.
Holtom, B. C., Tidd, S. T., Mitchell, T. R., & Lee, T. W. (2013). A demonstration of the importance of temporal considerations in
the prediction of newcomer turnover. Human Relations,66, 1337–1352. https://doi.org/10.1177/0018726713477459
Hom, P. W., & Hulin, C. L. (1981). A competitive test of the prediction of reenlistment by several models. Journal of Applied
Psychology,66, 23–39. https://doi.org/10.1037/0021-9010.66.1.23
Hom, P. W., & Kinicki, A. J. (2001). Toward a greater understanding of how dissatisfaction drives employee turnover. Academy
of Management Journal,44, 975–987. https://doi.org/10.2307/3069441
Hom, P. W., & Griffeth, R. W. (1995). Employee turnover. Cincinnati, OH: South-West College.
Hom, P. W., Griffeth, R. W., Palich, L. E., & Bracker,J. S. (1998). An exploratory investigation into theoretical mechanisms under-
lying realistic job previews. Personnel Psychology,51, 421–451. https://doi.org/10.1111/j.1744-6570.1998.tb00732.x
Hom, P. W., Griffeth, R. W., & Sellaro, C. L. (1984). The validity of Mobley’s (1977) model of employee turnover.Organizational
Behavior and Human Performance,34, 141–174. https://doi.org/10.1016/0030-5073(84)90001-1
Hom, P. W., Lee, T. W., Shaw, J. D., & Hausknecht, J. P. (2017). One hundred years of employee turnover theory and research.
Journal of Applied Psychology. http://psycnet.apa.org/doi/10.1037/apl0000114
Hom, P. W., Mitchell, T. R., Lee, T. W., & Griffeth, R. W. (2012). Reviewing employee turnover: Focusing on proximal withdrawal
states and an expanded criterion. Psychological Bulletin,138, 831–858. https://doi.org/10.1037/a0027983
Hom, P. W., Roberson, L., & Ellis, A. D. (2008). Challenging conventional wisdom about employee turnover: Revelations from
corporate America. Journal of Applied Psychology,93, 1–34. https://doi.org/10.5465/AMBPP.2006.27161825
Huffman, A. H., Adler, A. B., Dolan, C. A., & Castro, C. A. (2005). The impact of operations tempo on turnover intentions of Army
personnel. Military Psychology,17, 175–202. https://doi.org/10.1207/s15327876mp1703_4
Huffman, A. H., Casper, W. J., & Payne, S. C. (2013). How does spouse career support relate to employee turnover?
Work interfering with family and job satisfaction as mediators. Journal of Organizational Behavior,35, 194–212.
Humphrey, S. E. (2011). What does a great meta-analysis look like? Organizational Psychology Review,1, 99–103.
Humphrey, S. E., Nahrgang, J. D.,& Morgeson, F. P. (2007). Integrating motivational, social, and contextual work design features:
A meta-analytic summary and theoretical extension of the work design literature. Journal of Applied Psychology,92, 1332–
Hunter, J. E., & Schmidt, F. L . (2004). Methods of meta-analysis (2nd ed). Thousand Oaks: Sage.
Ilies, R., Wilson, K. S., & Wagner, D. T. (2009). The spillover of daily job satisfaction onto employees’ fam-
ily lives: The facilitating role of work-family integration. Academy of Management Journal,52, 87–102.
Iverson, R. D. (1999). An event history analysis of employee turnover: The case of hospital employees in Australia. Human
Resource Management Review,9, 397–418. https://doi.org/10.1016/S1053-4822(99)00027-3
Iverson, R. D., & Pullman, J. A. (2000). Determinants of voluntary turnover and layoffs in an environment of
repeated downsizing following a merger: An event history analysis. JournalofManagement,26, 977–1003.
Jackofsky, E. F., Ferris, K. R., & Breckenridge, B. G. (1986). Evidence for a curvilinear relationship between job performance and
turnover. JournalofManagement,12, 105–111. https://doi.org/10.1177/014920638601200109
Jiang, K., Liu, D., McKay, P. F., Lee, T. W., & Mitchell, T. R. (2012). When and how is job embeddedness predictive of turnover? A
meta-analytic investigation. Journal of Applied Psychology,97, 1077–1096. https://doi.org/10.1037/a0028610
Johns, G. (2006). The essential impact of context on organizational behavior. Academy of Management Review,31, 386–408.
Joseph, D., Ang, S., & Slaughter, S. A. (2015). Turnover or turnaway? Competing risks analysis of male and
female IT professionals’ job mobility and relative pay gap. Information Systems Research,26, 145–164.
40 RUBENSTEIN ET AL.
Judge, T. A. (1993). Does affective disposition moderate the relationship between job satisfaction and voluntary turnover?
Journal of Applied Psychology,78, 395–401. https://doi.org/10.1037/0021-9010.78.3.395
Judge, T. A., & Bono, J. E. (2001). Relationship of core self-evaluations traits—self-esteem, generalized self-efﬁcacy, locus of
control, and emotional stability—with job satisfaction and job performance: A meta-analysis. Journal of Applied Psychology,
86, 80–92. https://doi.org/10.1037/0021- 9010.86.1.80
Judge, T. A., Heller, D., & Mount, M. K. (2002). Five-factor model of personality and job satisfaction: A meta-analysis. Journal of
Applied Psychology,87, 530–541. https://doi.org/10.1037/0021-9010.87.3.530
Judge, T. A., Simon, L. S., Hurst, C., & Kelley, K. (2014). What I experienced yesterday is who I am today: Relationship of work
motivations and behaviors to within-individual variation in the ﬁve-factor of personality. Journal of Applied Psychology,9,
Kahn, W. A. (1990). Psychological conditions of personal engagement and disengagement at work. Academy of Management
Journal,33, 692–724. https://doi.org/10.2307/256287
Kammeyer-Mueller, J. D., Rubenstein, A. L., Long, D. M., Odio, M. A., Buckman, B. R., Zhang, Y., & Halvorsen-Ganepola, M. D. K.
(2013). A meta-analytic structural model of dispositional affectivity and emotional labor. Personnel Psychology,66, 47–90.
Kammeyer-Mueller, J. D., Wanberg, C. R., Glomb, T. M., & Ahlburg, D. (2005). The role of temporal shifts in turnover processes:
it’s about time. Journal of Applied Psychology,90, 644–658. https://doi.org/10.1037/0021-9010.90.4.644
Karagonlar, G., Eisenberger, R., & Aselage, J. (2016). Reciprocation wary employeesdiscount psychological contract fulﬁllment.
Journal of Organizational Behavior,37, 23–40.
Kemery, E. R., Dunlap, W. P., & Griffeth, R. W. (1988). Correction for variance restriction in point-biserial correlations. Journal
of Applied Psychology,73, 688–691. https://doi.org/10.1037/0021-9010.73.4.688
Kerlinger, F. N., & Lee, H. B. (2000). Foundations of behavioral research. Fort Worth, TX: Harcourt.
Kinicki, A. J., Lockwood, C. A., Hom, P. W., & Griffeth, R. W. (1990). Interviewer predictions of applicant qualiﬁ-
cations and interviewer validity: Aggregate and individual analyses. Journal of Applied Psychology,75, 477–486.
Klein, H. J., Molloy, J. C., & Brinsﬁeld, C. T. (2012). Reconceptualizing workplace commitment to redress a stretched con-
struct: Revisiting assumptions and removing confounds. Academy of Management Review,37, 130–151. https://doi.org/
Koslowsky, M., Sagie, A., Krausz, M., & Singer, A. D. (1997). Correlates of employee lateness: Some theoretical considerations.
Journal of Applied Psychology,82, 79–88. https://doi.org/10.1037/0021-9010.82.1.79
Kraimer, M. L., Seibert, S. E., Wayne, S. J., Liden, R. C., & Bravo, J. (2011). Antecedents and outcomes of organiza-
tional support for development: The critical role of career opportunities. Journal of Applied Psychology,96, 485–500.
Kristof, A. L. (1996). Person-organization ﬁt: An integrative review of its conceptualizations, measurement, and implications.
Personnel Psychology,49, 1–49. https://doi.org/10.1111/j.1744-6570.1996.tb01790.x
Kristof-Brown, A. L., Zimmerman, R. D., & Johnson, E. C. (2005). Consequences of individuals’ ﬁt at work: A meta-
analysis of person–job, person–organization, person–group, and person–supervisor ﬁt. Personnel Psychology,58, 281–342.
Lee, T. W., Ashford, S. J., Walsh, J. P., & Mowday, R. T. (1992). Commitment propensity, organizational commitment,
and voluntary turnover: A longitudinal study of organizational entry processes. Journal of Management,18, 15–32.
Lee, T. W., Burch, T. C., & Mitchell, T. R. (2014). The story of why we stay: A review of job embedded-
ness. Annual Review of Organizational Psychology and Organizational Behavior,1, 199–216. https://doi.org/
Lee, T. W., & Maurer, S. D. (1999). The effects of family structure on organizational commitment, intention to leave and volun-
tary turnover. Journal of Managerial Issues,11, 493–513.
Lee, T. W., & Mitchell, T. R. (1994). An alternative approach: The unfolding model of voluntary employee turnover. Academy of
Management Review,19, 51–89. https://doi.org/10.5465/AMR.1994.9410122008
Lee, T. W., Mitchell, T. R., Holtom, B. C., McDaneil, L. S., & Hill, J. W. (1999). The unfolding model of voluntary turnover: A
replication and extension. Academy of Management Journal,42, 450–462. https://doi.org/10.2307/257015
Lee, T. W., Mitchell, T. R., Sablynski, C. J.,Burton, J. P., & Holtom,B. C. (2004). The effects of job embeddedness on organizational
citizenship, job performance, volitional absences, and voluntary turnover. Academy of Management Journal,47, 711–722.
RUBENSTEIN ET AL.41
Lewin, K. (1945). The research center for group dynamics at Massachusetts Institute of Technology. Sociometry,8, 126–136.
Li, J. I., Lee, T. W., Mitchell, T. R., Hom, P. W., & Griffeth, R. G. (2016). The effects of proximal withdrawal states on job atti-
tudes, job search, intention to leave and voluntary employee turnover. Journal of Applied Psychology,101, 1436–1456.
Li, N., Liang, J., & Crant, J.M. (2010). The role of proactive personality in job satisfaction and organizational citizenship behavior:
A relational perspective. Journal of Applied Psychology,95, 395–404. https://doi.org/10.1037/a001-8079
Liao, H., Chuang, A., & Joshi, A. (2008). Perceived deep-level dissimilarity: Personality antecedents and impact on overall job
attitude, helping, work withdrawal, and turnover. Organizational Behavior and Human Decision Processes,106, 106–124.
Linnehan, F., & Blau, G. (2003). Testing the impact of job search and recruitment source on new hire turnover in a maquiladora.
Applied Psychology: An International Review,52, 253–271.
Liu, D., Mitchell, T. R., Lee, T. W., Holtom, B. C., & Hinkin, T. (2012). When employees are out of step with coworkers: How
job satisfaction trajectory and dispersion inﬂuence individual- and unit-level voluntary turnover. Academy of Management
Journal,55, 1360–1380. https://doi.org/10.5465/amj.2010.0920
Lyness, K. S., & Judiesch, M. K. (2001). Are female managers quitters? The relationships of gender, promo-
tions, and family leaves of absence to voluntary turnover. Journal of Applied Psychology,86, 1167–1178.
Maertz, C. P., & Campion, M. A. (2004). Proﬁles in quitting: Integrating process and content turnover theory. Academy of Man-
agement Journal,47, 566–582. https://doi.org/10.2307/20159602
March, J. G., & Simon, H. A. (1958). Organizations. New York, NY: Wiley.
Martin, T. N. (1980). Modelling the turnover process. Journal of Management Studies,17, 261–274.
Matta, F.K., Scott, B., Colquitt, J., Koopman, J., & Passantino, L. (in press). Is consistently unfair better than sporadically fair? An
investigation of justice variability and stress. Academy of Management Journal. https://doi.org/10.5465/amj.2014.0455
McElroy, J. C., Morrow, P. C., & Fenton, J. B. (1995). Absenteeism and performance as predictors of voluntary turnover. Journal
of Managerial Issues,7, 91–98. http://www.jstor.org/stable/40604052
McElroy, J. C., Morrow, P. C., & Rude, S. N. (2001). Turnover and organizational performance: A comparative analysis of
the effects of voluntary, involuntary, and reduction-in-force turnover. Journal of Applied Psychology,86, 1294–1299.
McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociol-
ogy,27, 415–444. https://doi.org/10.1146/annurev.soc.27.1.415
Mirvis, P. H., & Lawler, E. E. (1977). Measuring the ﬁnancial impact of employee attitudes. Journal of Applied Psychology,62,1–8.
Mitchell, T. R., Holtom,B. C., Lee, T. W., Sablynski, C. J.,& Erez, M. (2001). Why people stay: Using job embeddedness to predict
voluntary turnover. Academy of Management Journal,44, 1102–1121. https://doi.org/10.2307/3069391
Mobley, W. H. (1977). Intermediate linkages in the relationship between job satisfaction and employee turnover. Journal of
Applied Psychology,62, 237–240. https://doi.org/10.1037/0021-9010.62.2.237
Morgeson, F.P.,Mitchell, T. R., & Liu, D. (2015). Event system theory: An event-orientedapproach to the organizational sciences.
Academy of Management Review,40, 515–537. https://doi.org/10.5465/amr.2012.0099
Motowidlo, S. J. (1983). Predicting sales turnover from pay satisfaction and expectation. Journal of Applied Psychology,68, 484–
Ostroff, C. (1993). The effects of climate and personal inﬂuences on individual behavior and attitudes in organizations. Organi-
zational Behavior and Human Decision Processes,56, 56–90. https://doi.org/10.1006/obhd.1993.1045
Owens, B. P., Johnson, M. D., & Mitchell, T. R. (2013). Expressed humility in organizations: Implications for performance, teams,
and leadership. Organization Science,24, 1517–1538. https://doi.org/10.1287/orsc.1120.0795
Parasuraman, S., & Alutto, J.A. (1984). Sources and outcomes of stress in organizational settings: Toward the development of a
structural model. Academy of Management Journal,27, 330–350. https://doi.org/10.2307/255928
Park, T. Y., & Shaw, J. D. (2013). Turnover rates and organizational performance: A meta-analysis. Journal of Applied Psychology,
98, 268–309. https://doi.org/10.1037/a0030723
Pfeffer, J. (1983). Organizational demography. In L. L. Cummings, & B. Staw (Eds.), Research on organizational behavior (vol. 5,
pp. 299–359). Greenwich, Conn: JAI Press.
42 RUBENSTEIN ET AL.
Podsakoff, N. P., LePine, J. A., & LePine, M. A. (2007). Differential challenge stressor-hindrance stressor relationships with job
attitudes, withdrawal cognitions, turnover, and withdrawal behavior: A meta-analysis. Journal of Applied Psychology,92,
Podsakoff, N. P., Whiting, S. W., Podsakoff, P. M., & Blume, B. D. (2009). Individual-and organizational-level con-
sequences of organizational citizenship behaviors: A meta-analysis. Journal of Applied Psychology,94, 122–141.
Podsakoff, P. M., MacKenzie, S. B., Lee, J. Y., & Podsakoff, N. P. (2003). Common method biases in behavioral
research: A critical review of the literature and recommended remedies. Journal of Applied Psychology,88, 879–903.
Price, J. L., & Mueller, C. W. (1981). A causal model of turnover for nurses. Academy of Management Journal,24, 543–565.
Quiñones, M. A., Ford, J. K., & Teachout, M. S. (1995). The relationship between work experience and job
performance: A conceptual and meta-analytic review. Personnel Psychology,48, 887–910. https://doi.org/
Ramesh, A., & Gelfand, M. J. (2010). Will they stay or will they go? The role of job embeddedness in predicting turnover in
individualistic and collectivistic cultures. Journal of Applied Psychology,95, 807–823.
Rhoades, L., Eisenberger, R., & Armeli, S. (2001). Affective commitment to the organization: The contribution of perceivedorga-
nizational support. Journal of Applied Psychology,86, 825–836. https://doi.org/10.1037/0021-9010.86.5.825
Riolli, L., & Savicki, V. (2006). Impact of fairness, leadership, and coping on strain, burnout, and turnover in organizational
change. International Journal of Stress Management,13, 351–377. https://doi.org/10.1037/1072-5245.13.3.351
Rosse, J. G. (1988). Relations among lateness, absence, and turnover: Is there a progression of withdrawal? Human Relations,
41, 517–531. https://doi.org/10.1177/001872678804100702
Schmidt, K., & Daume, B. (1993). Job characteristics and voluntary employee turnover: Direct and moderated relationships.
European Work and Organizational Psychologist,3,29–42. https://doi.org/10.1080/09602009308408574
Schneider, B. (1987). The people make the place. Personnel Psychology,40, 437–453. https://doi.org/
Schneider, B., Ehrhart, M. G., & Macey, W. H. (2013). Organizational climate and culture. Annual Review of Psychology,64, 361–
Shapiro, D. L., Hom, P., Shen, W., & Agarwal, A. (2016). Howe do leader departures affect subordinates’ orga-
nizational attachment? A 360-degree relational perspective. Academy of Management Review,41, 479–502.
Sheridan, J. E. (1985). A catastrophe model of employee withdrawal leading to low job performance, high absen-
teeism, and job turnover during the ﬁrst year of employment. Academy of Management Journal,28, 88–109.
Sheridan, J. E., & Vredenburgh, D. J. (1978). Usefulness of leadership behavior and social power variables in pre-
dicting job tension, performance, and turnover of nursing employees. Journal of Applied Psychology,63, 89–95.
Smith-Crowe, K., Burke, M. J., & Landis, R. S. (2003). Organizational climate as a moderator of safety knowledge-safety perfor-
mance relationships. Journal of Organizational Behavior,24, 861–876. https://doi.org/10.1002/job.217
Spector, P. E. (1991). Conﬁrmatory test of a turnover model utilizing multiple data sources. Human Performance,4, 221–230.
Spell, H. B., Eby, L. T., & Vandenberg, R. J. (2014). Developmental climate: A cross-level analysis of voluntary turnover and job
performance. Journal of Vocational Behavior,84, 283–292. https://doi.org/10.1016/j.jvb.2014.02.001
Spreitzer, G. M., & Mishra, A. K. (2002). To stay or to go: Voluntary survivor turnover following an organizational downsizing.
Journal of Organizational Behavior,23, 707–729. https://doi.org/10.1002/job.166
Stanley, L., Vandenberghe, C., Vandenberg, R., & Bentein, K. (2013). Commitment proﬁles and employee turnover. Journal of
Vocational Behavior,82, 176–187. https://doi.org/10.1016/j.jvb.2013.01.011
Steel, P. D., & Kammeyer-Mueller, J. D. (2002). Comparing meta-analytic moderator estimation techniques under realistic con-
ditions. Journal of Applied Psychology,87, 96–111. https://doi.org/10.1037/0021-9010.87.1.96
Steel, R. P. (2002). Turnover theory at the empirical interface: Problems of ﬁt and function. Academy of Management Review,27,
RUBENSTEIN ET AL.43
Steiger, J. H. (1980). Tests for comparing elements of a correlation matrix. Psychological Bulletin,87, 245–251.
Stumpf, S. A., & Dawley, P. K. (1981). Predicting voluntary and involuntary turnover using absenteeism and performance indices.
Academy of Management Journal,24, 148–163. http://doi.org/10.2307/255830
Suszko, M. K., & Breaugh, J. A. (1986). The effects of realistic job previews on applicant self-selection and employee turnover,
satisfaction, and coping ability. JournalofManagement,12, 513–523. https://doi.org/10.1177/014920638601200406
Swider, B. W., Boswell, W. R., & Zimmerman, R. D. (2011). Examining the job search–turnover relationship: The
role of embeddedness, job satisfaction, and available alternatives. Journal of Applied Psychology,96, 432–441.
Tekleab, A. G., Bartol, K. M., & Liu, W. (2005). Is it pay levels or payrises that matter to fairness and turnover? Journal of Organi-
zational Behavior,26, 899–921. https://doi./org/10.1002/job.352
Timmerman, T. A. (2006). Predicting turnover with broad and narrow personality traits. International Journal of Selection and
Assessment,14, 392–399. https://doi.org/10.1111/j.1468-2389.2006.00361.x
Trevor, C. O. (2001). Interactions among actual ease-of-movement determinants and job satisfaction in the prediction of vol-
untary turnover. Academy of Management Journal,44, 621–638. https://doi.org/10.2307/3069407
Tse, H. H. M., Huang, X., & Lam, W. (2013). Why does transformational leadership matter for employee turnover? A multi-foci
social exchange perspective. The Leadership Quarterly,24, 763–776. https://doi.org/10.1016/j.leaqua.2013.07.005
Vandenberge, C., & Bentein, K. (2009). A closer look at the relationship between affective commitment to super-
visors and organizations and turnover. Journal of Occupational and Organizational Psychology,82, 331–348.
Vardaman, J.M., Taylor, S. G., Allen, D. G., Gondo, M. B., & Amis, J.M. (2015). Translating intentions to behavior: The interaction
of network structure and behavioral intentions in understanding employee turnover.Organization Science,26, 1177–1191.
Vidyarthi, P. R., Erdogan, B., Anand, S., Liden, R. C., & Chaudhry, A. (2014). One member, two leaders: Extend-
ing leader–member exchange theory to a dual leadership context. Journal of Applied Psychology,99, 468–483.
Wefald, A. J., & Downey, R. G. (2009). Job engagement in organizations: Fad, fashion, or folderol? Journal of Organizational
Behavior,30, 141–145. https://doi.org/10.1002/job.560
Wille, B., Hofmans, J., Feys, M., & De Fruyt, F. (2014). Maturation of work attitudes: Correlated change with Big
Five personality traits and reciprocal effects over 15 years. Journal of Organizational Behavior,35, 507–529.
Williams, C. R. (1999). Reward contingency, unemployment, and functional turnover. Human Resource Management Review,9,
Wright, T. A., & Bonett, D. G. (1993). Role of employee coping and performance in voluntary employee withdrawal: A research
reﬁnement and elaboration. Journal of Management,19, 147–161. https://doi.org/10.1016/0149-2063(93)90050-W
Zatzick, C. D., Elvira, M. M., & Cohen, L. E. (2003). When Is more better? The effects of racial composition on voluntary turnover.
Organization Science,14, 483–496. https://doi.org/10.1287/orsc.14.5.483.16768
Zimmerman, R. D. (2008). Understanding the impact of personality traits on individuals’ turnover decisions: A meta-analytic
path model. Personnel Psychology,61, 309–348. https://doi.org/10.1111/j.1744-6570.2008.00115.x
Zimmerman, R. D., Swider, B. W., Woo, S. E., & Allen, D. G. (2016). Who withdraws? Psychological individual differences and
employee withdrawal behaviors. Journal of Applied Psychology,101, 498–519. https://doi.org/10.1037/apl0000068
Zou, G. Y. (2007). Toward using conﬁdence intervals to compare correlations. Psychological Methods,12, 399–413.
How to cite this article: Rubenstein AL, Eberly MB, Lee TW, Mitchell TR. Surveying the forest: A meta-analysis,
moderator investigation, and future-oriented discussion of the antecedents of voluntary employee turnover.
Personnel Psychology. 2017;00:1–43. https://doi.org/10.1111/peps.12226