Predicting Employee Turnover from
Thomas Hugh Feeley, Jennie Hwang &
George A. Barnett
Employees (n40) at a fast-food restaurant were surveyed about characteristics of their
position and their level of satisfaction. Employees were then asked to report with whom
they regularly communicated inside and outside the workplace and to indicate how close
they were to employees with whom they were linked. Employee turnover was measured
after three months had elapsed. A goal of the research was to replicate a model of
employee turnover that predicts employees more central in their social network to be less
likely to leave, and to test a social support explanation of the centrality model. The results
indicated that employees who reported a greater number of out-degree links with friends
were less likely to leave. The number of in-degree links with friends did not significantly
predict turnover, and neither did network links with peers. Friendship prestige, measured
by the number of in-degree links, was strongly correlated with relational closeness and
amount of time spent with employees outside the workplace.
Keywords: Turnover; Centrality; Networks; Friendship; Peer; Social Support
Organizations may be considered webs or systems of relationships (Katz & Kahn,
1978; Sias & Perry, 2004). A large chunk of time in the workplace is spent interacting
with coworkers using various communication media (face-to-face, CMC, email, or
phone). Communication and interaction are hardly confined to the break room or
water cooler, and many individuals have many and varied relationships with other
employees that range from peer or coworker relations at one end of the continuum to
best friends or even intimate relations at the other (Sias & Cahill, 1998). The
workplace presents an interesting context for scholars of social relations in that
individuals can choose to engage and sustain relationships, but may also be forced into
Thomas Feeley is a faculty member in Communication and Family Medicine at the University at Buffalo (UB),
of the State University of New York. George Barnett is Professor of Communication and Jennie Hwang is a
doctoral candidate in Communication at UB. Correspondence to: Thomas Hugh Feeley, Department of
Communication, University at Buffalo, The State University of New York, 329 Baldy Hall, North Campus,
Amherst, NY 14261, USA. Email: email@example.com.
ISSN 0090-9882 (print)/ISSN 1479-5752 (online) #2008 National Communication Association
Journal of Applied Communication Research
Vol. 36, No. 1, February 2008, pp. 5673
coworker relationships due to interdependence or equivalence in job roles (Jablin,
2001; Shah, 1998). These involuntary relationships can be unpleasant and are often
sources of strain; very often, the employee must find a method to cope with the other
or elect to leave his or her position or organization (Hess, 2000; Sias & Cahill, 1998).
The current study examines the role of social networks in the employee retention
and turnover process. Social networks are defined as ‘‘interconnected individuals who
are linked by patterned communication flows’’ (Rogers & Kincaid, 1981, p. 82). Stated
more simply, social networks specify who speaks with whom in an organization
(Feeley & Barnett, 1997). Communication networks are defined by Monge and
Contractor (1999) as ‘‘patterns of contact between communication partners that are
created by transmitted and exchanging messages through time and space’’ (p. 440).
The current study examines individuals or actors as the unit of analysis, whereas other
studies use groups or organizations as the node (compared to actors) or unit of
analysis (e.g., Doerfel & Taylor, 2004).
Scholars in organizational communication (e.g., Feeley, 2003; Krackhardt & Porter,
1985; Pfeffer, 1991) suggest that examining both structural factors (e.g., social
networks) and individual factors (e.g., job satisfaction and pay) in the same study
design promises to add to our understanding of important antecedent (e.g., employee
strain and intentions to quit) and criterion factors (e.g., turnover and retention,
promotion, and disengagement) in the workplace assimilation process (see Jablin,
2001, for an excellent review on employee entry, assimilation, and disengagement).
Research into networks (see Monge & Contractor, 1999, for a review) examines the
type of information transmitted and the strength of the links between entities or
actors. In the case of social support networks, studies often focus on two mechanisms:
how these networks buffer the effects on stress and/or how resources are disseminated
(Monge & Contractor, 1999; Walker, Wasserman, & Wellman, 1994).
Seeking to understand the employee turnover process is a valuable research
initiative as employee turnover can be both helpful and harmful to an organization.
Turnover can be helpful in getting rid of ‘‘bad eggs’’ in the workplace and bringing in
‘‘new blood,’’ perhaps to stir up creativity and healthy competitiveness. As Staw
(1980) suggested, positive consequences of turnover, including increased perfor-
mance, reduction of entrenched conflict, increased mobility and morale, innovation,
and adaptation, show that turnover serves a beneficial function for the viability of
the organization. Turnover may be ‘‘very lucrative, a veritable windfall for the
organization’’ when a responsible cost/benefit analysis is examined (Dalton & Todor,
1982, p. 212). Constructive turnover also helps to increase overall organizational
effectiveness (Abelson & Baysinger, 1984), increase structural flexibility, and
stimulate changes in policy and practice (Johnson, Griffeth, & Griffin, 2000; Mobley,
1982). In addition, low turnover can be costly to employers in terms of dollars, as
companies often use periodic raises to reinforce employee loyalty (Dalton,
Krackhardt, & Porter, 1981). On the other hand, turnover can be harmful to an
organization in many ways. First, employee hiring and training can be costly and
labor intensive, and often serves to slow down production or operations. Also,
employee turnover can be harmful to employee morale and actually beget more
Turnover and Networks 57
turnover, leaving structural holes in an area of an organization (Krackhardt & Porter,
1986; Staw, 1980; Susskind, Miller, & Johnson, 1998). Understanding why employees
decide to stay or leave an organization is an important area of inquiry for scholars
and management practitioners alike.
Social Networks and Coworker Relationships
Assimilation concerns how an individual becomes integrated into the culture or
system of an organization (Jablin, 1987; Van Maanen, 1978). Jablin (2001) suggests
that employee assimilation is composed of two interrelated processes: how the
organization socializes employees and how individual employees socialize or
individualize their role and relationships in the workplace. Recent research suggests
that employers tolerate and even promote employee relationships and integration
into the social network. Berman, West, and Richter (2002) surveyed senior managers
of companies located in cities of over 50,000 residents and found that managers not
only condone employee friendships, but also actively strive to foster an open and
nurturing climate in the organization in order to encourage friendliness. In another
interesting line of research, Friedman and Holtom (2002) found that several
organizations created network groups in an attempt to decrease attrition in
managerial-level minority employees. These authors found that employees who are
part of network groups report lower intentions to leave than employees who are not
affiliated with organization-sponsored network groups.
There is an underlying proposition in the organizational communication literature
that employees who are highly networked (i.e., employees who speak to more
individuals) are happier, more committed employees who will be less likely to ‘‘fall
off’’ or leave the organization. Workmate ties provide everyday social support
(Wellman & Frank, 2001). Research has provided empirical support for this
proposition. Employees more centrally located in social networks are more involved
in the workplace (Marshall & Stohl, 1993), more committed to their roles (Eisenberg,
Monge, & Miller, 1983; Feeley, 2000; Hartman & Johnson, 1989), more satisfied with
their jobs (Monge, Edwards, & Kirste, 1983), and identify more with the organization
(Bullis & Bach, 1991).
Centrality in one’s social network is traditionally measured as the number of direct
(or indirect) links one has in the organization compared to the number of possible
links one could have (Feeley & Barnett, 1997; Monge et al., 1983). Thus, any measure
of centrality is relative; highly integrated or dense networks feature many webs of
links (pathways) between members, while sparsely connected networks might have a
few highly central and connected individuals with many on the periphery of the social
network who communicate in silos or clusters within their own individual
58 T. H. Feeley et al.
The notion of centrality can be traced back to the seminal work of Bavelas (1950)
and Leavitt (1951) in group efficiency in decision-making. Those more central have
greater access to information and material resources, while those less central are more
dependent on others for information and require more links and effort to
communicate with focal others. The most important or prominent actors are usually
located in strategic locations within the network (Wasserman & Faust, 1994).
Freeman (1979) identified three primary measures of centrality in a communication
network: in and out degree, closeness, and betweenness. One with a high out-degree
speaks to more individuals in the group and is more connected. An actor with high
in-degree centrality is in direct contact or is adjacent to many other actors; he or she
is ‘‘where the action is’’ (Wasserman & Faust, 1994). Betweenness refers to the
frequency with which a position falls between pairs of positions in the network and
can be considered more influential (Monge & Contractor, 1999; Mullen, Johnson, &
Salas, 1991). Closeness refers to the extent to which one is close to all others in the
network. One with high closeness is traditionally located near the center of
the network and requires fewer links to communicate with all others in the network;
the actor with high closeness can quickly interact with all others and is theoretically
more productive in communicating information to others (Wasserman & Faust,
1994). Freeman, Roeder, and Mulholland (1979) use the terms ‘‘control,’’ ‘‘indepen-
dence,’’ and ‘‘activity’’ in the network to characterize betweenness, closeness, and
degree centrality, respectively.
An important consideration in network research is how a communication link or
connection is measured. One class of network studies examines egocentric networks
(Everett & Borgatti, 2004) that require individuals to self-report who they speak with,
and then uses matrix algebra (actor by actor matrix) to produce measures of
centrality. Also potentially relevant is to whom individuals speak on a regular basis,
and for what type of information. For example, Shah (2000) found that employees
rely on structurally equivalent others (i.e., those who hold a similar role or position in
the network) for job-related information and rely on cohesive referents (i.e., those
who have close ties with a focal individual) for general information and as social
referents. Thus, the strength of the tie (Granovetter, 1973) or link between employees
may be important in examining organizational retention and turnover. Other
network analyses have used organizations as the node or unit of analysis (Doerfel &
Taylor, 2004; Shumante, Fulk, & Monge, 2005) in lieu of the actor.
Peer-Relationships vs. Friendships in the Workplace
Individuals are more likely to have a greater number of peer relationships than
friendship networks in an organization (Kram & Isabella, 1985; Sias & Cahill, 1998).
Peer relationships can be important to an employee’s morale and level of loyalty to
the organization. A recent line of research by Sias and colleagues (Sias & Cahill, 1998;
Sias & Perry, 2004) examines how peer relationships over time can strengthen into
friendships or even ‘‘best friendships.’’ On the other hand, peer relationships can
Turnover and Networks 59
‘‘go sour,’’ and employees are faced with the unwelcome task of disengaging from
Research into social networks provides evidence that greater involvement in peer
relationships can lead to decreased turnover. Feeley and Barnett (1997); see also
Feeley, 2000) found individuals who were more central in their network to be less
likely to have left after six months; employees at a large supermarket were asked to
report with whom they spoke about work-related or social matters in the workplace.
Feeley and Barnett (1997) suggested that employees more centrally located in the
network may become more committed to the organization and may also have access
to job-related information that may be privileged to those ‘‘in the know.’’ These
authors dubbed this an ‘‘erosion model’’ of turnover, wherein those on the periphery
of the network turnover or, apparently, slide down the network edges.
A recent study by Mossholder, Settoon, and Henagan (2005) studied turnover in a
sample of healthcare employees from a large public medical center. Using survival
analysis over a five-year time frame, the authors found that network centrality
significantly predicted turnover (r.22). Data from Mossholder et al. (2005)
provided evidence that the erosion model may generalize to other types of
organizations that experience considerably lower turnover than the organizations
studied by Feeley (2000); Feeley & Barnett, 1997). The following prediction is made
based upon the current literature investigating the effect of centrality on employee
H1: Employees who are more centrally located in the peer relationship network
will be less likely to turnover.
Feeley (2000) replicated the earlier findings by Feeley and Barnett (1997) and found
that employee commitment did not explain the networkturnover relationship. In
fact, Feeley’s data indicated that leavers were slightly more committed to the
workplace than stayers.
Toward a Social Support Explanation
The cumulative findings of the studies by Feeley (2000), Feeley and Barnett (1997), and
Mossholder et al. (2005) indicate a moderately strong relationship (rbetween .22
and .48) between social network position and employee turnover. However, these
studies do not provide a clear and empirically tested explanation as to why those more
on the periphery of the network are leaving. Feeley (2000) predicted that the
relationship between network centrality and turnover would be mediated by employee
commitment to work and intentions to remain at work. Feeley’s data failed to support
this model. This paper argues that employees more central, more active, or more ‘‘in
the thick of things’’ have greater access to social support in the form of coworkers.
Social support in the workplace is an important predictor of employee strain and
dissatisfaction (Lee & Ashforth, 1996). Recent research by Brotheridge (2001) tested
five models of employee coping using a sample of 680 Canadian government
employees. Brotheridge’s data supported a coping-as-strain-deterrent model that
60 T. H. Feeley et al.
predicts that both work stressors and coping resources have independent and direct
effects on strain. Simply stated, having more coping resources that may include
coworker social support serves to reduce levels of strain regardless of the level of stress
that may be perceived.
Apker and Ray (2003) discuss the importance of stress and social support for
individuals who are employed in the healthcare industry. Summarizing the social
support literature amassed over several decades of research, Apker and Ray (2003)
conclude that communication of social support from coworkers, particularly peers
and immediate supervisors, is an important means for coping with job stressors by
empowering the individual to have better control over and understanding of his or
her stressors in the workplace. It is interesting to note that Metts, Geist, and Gray
(1994) found that advice and support provided during work, compared to support
received outside of work, were more effective in a sample of nursing professionals.
A social support theoretical model would suggest that not only should the number
of peer relationships matter to turnover (as the number of peers may be a proxy
measure of the availability of coping resources), but the type or closeness of the
employee relationship should also matter. A social support explanation would predict
the greatest coping resources to be available to employees who report many close
friends, and the fewest coping resources to be available to employees with few peer
relationships. The following hypotheses are advanced:
H2: Employees who are more centrally located in the friendship network will be
less likely to turnover.
H3: Centrality in one’s friendship network will explain more of the variance in
turnover than centrality in one’s peer relationship network.
Both Feeley and Barnett (1997) and Feeley (2000) failed to distinguish between peer
and social relationships and to account for this in their analyses. Moreover, both
studies neglected to measure the relative strength of the links perceived by employees.
Feeley (2000) asked employees to indicate by checkmark ‘‘the individuals you
communicate with at work on a regular basis about work-related or social-related
topics’’ (p. 269). Feeley and Barnett (1997) asked employees to indicate if an
employee was unknown to them, a coworker with whom they spoke, or a personal
friend. However, the authors combined the latter two categories for their measure of
a network link. The current study attempts to provide a distinction between peer
social networks and friendship networks while also measuring the closeness of each
relation. It is expected that employees with close others to turn to about work (or
personal) stressors will be more likely to remain in their position, as individuals with
close network ties, it is predicted, will have greater coping resources. Thus:
H4: Employees reporting greater relational closeness with network ties will be less
likely to turnover.
The current study examines employee turnover at a fast-food restaurant. The
decision to study such an organization was made in an effort to examine turnover at
Turnover and Networks 61
an organization that is frequently challenged by employee retention. Studying a fast-
food restaurant also provides a more direct comparison to the organizations in the
Feeley studies, which included a supermarket (Feeley & Barnett, 1997), a fast-food
restaurant, a pizza parlor, and a drug store (Feeley, 2000).
In November of 2004, employees took a survey that took approximately 20 minutes
to complete during their scheduled shift. After three months, the roster of employees
who completed the survey was submitted to the store manager who noted which
employees had left, when, and whether the turnover was voluntary or involuntary.
The project was approved by the Social and Behavioral Sciences Institutional Review
Board at University at Buffalo, of the State University of New York. Employees under
18 years of age were required to have parental consent to participate in the survey.
Participants and Organization
A census of employees (n40) agreed to participate in the survey, and two of these
were randomly selected to win a gift certificate of $25.00 for a local media/electronics
store. The store manager, who was apprised of the nature of the study, was not asked
to participate. The organization is a fast-food restaurant located in a suburb
approximately 10 miles from the city, and the store manager assured the research staff
of the turnover problem. The majority of employees were part-time, and many
employees were full-time high school or college students.
Individual and workplace factors. Employees were asked to report their age, sex, job
position (four categories), full- or part-time status, education level, race, and if they
were currently a full-time student. Respondents were asked to report their level of job
satisfaction on a single-item, seven-point scale anchored by ‘‘not at all satisfied’’ and
‘‘most satisfied.’’ Each respondent also reported how close s/he was to each of the
other 39 employees, and how much time s/he spent, in an average week, with each
other employee. Relational closeness was measured on a single-item, seven-point
scale anchored by ‘‘not at all close’’ and ‘‘very close’’ while time was measured on a
single-item, seven-point scale ranging from ‘‘never’’ to ‘‘always.’’ Higher values
indicate a greater amount of a given factor. Turnover was measured as a dichotomous
factor; stayers were coded as ‘‘0,’’ and leavers were coded as ‘‘1.’’
Centrality. Network centrality was measured by the degree measure: the number of
in-degree and out-degree links one has. UCINET [VI] software (Borgatti, Everett, &
Freeman, 2002) was used to compute all network measures. In-degree centrality was
measured by counting the number of employees who reported a relationship with a
focal employee; out-degree was measured by counting the number of out-links from
62 T. H. Feeley et al.
that employee. As a hypothetical example, if Tim reports speaking on a regular basis
with six persons about work-related topics, he would have a peer out-degree measure
of six (or this measure could be normed by dividing by the total possible number of
links, i.e., by 39). If eight employees report speaking to Tim about work, his in-degree
would be 8.
Degree is traditionally measured by the total number of positions in
direct contact with an individual. Separate analyses were computed for work or peer
network, and friendship network.
Each individual was asked to indicate (on a checklist) if they communicated with
each other employee by marking one of three choices: (1) the actor speaks with the
person rarely or never; (2) the actor speaks with the person about work-related
topics; or (3) the actor speaks to the person about work- and non-work-related
topics. For each entry in the actor-by-actor network matrix (peer vs. friendship), ‘‘1’’
represents a reported relation, and ‘‘0’’ represents a non-reported relationship.
Research Design and Analysis
Descriptive statistics (i.e., means, frequencies, and standard deviations) were used to
characterize the employees and the social network. Pearson correlations were used to
measure bivariate relations between study factors. Separate hierarchical logistic
regression analyses were computed for each hypothesized independent factor (peer
in-degree/out-degree and friendship in-degree/out-degree) on turnover. SPSS 13.0
was used to perform regression analyses and all other descriptive statistics. Separate
analyses were computed for two reasons: first, the sample size of 40 is considered too
small for regressing one factor onto many independent variables, and second, there
are concerns about collinearity among independent variables that are non-
Dare reported for effect size estimates. Also reported are
odds ratios, significance levels, and 95% confidence intervals for statistically
significant odds ratios.
The median age was 20 years, and 23 employees were female. The majority of
employees were Caucasian (n30); nine employees were African-American,
Hispanic, or Native American (one respondent failed to report race). Of the 32
employees who reported student status, 50% reported that they were currently full-
time students. All but one employee was part-time. Many employees were either high
school students or had failed to finish high school (n27); two employees were high
school graduates; two employees were attending college or had graduated from
college; and nine individuals failed to report education level. In terms of job position,
20 employees held entry-level positions, seven were trainers, one worked in
maintenance, and the remaining 12 were assistant managers.
The mean relational closeness was 2.59, somewhat more distant than average, and
the mean time was 1.71, indicating that the amount of time spent together was
Turnover and Networks 63
relatively low. Means and standard deviations are reported in Table 1. Each employee
reported communicating about work to eight employees and about work and non-
work to 10 employees. Job satisfaction was near the middle of the scale at 4.69 (of a
maximum score of 7). After three months, there were 28 stayers and 12 leavers; nine
of these were involuntary leavers and three were voluntary. Table 1 reports zero-order
correlation coefficients between each pair of non-categorical factors and turnover.
Work or peer out-degree was significantly correlated with age of respondent (r
.38), time (r.36), and perceived closeness (r.38). Friendship in-degree was
associated (pB.01) with time (r.73) and closeness (r.85), while friendship out-
degree was correlated with age (r.37), turnover (r.38), and time (r.40).
All these correlations are significant beyond the .05 level.
Test of Hypotheses
Peer networks. Hypothesis 1 predicted that centrality in one’s peer network would
predict employee turnover. Separate regression equations were conducted for peer in-
degree and peer out-degree.
The correlation between peer in-degree and out-degree
was only .32, indicating that these two variables are not collinear.
Peer out-degree failed to predict a statistically significant amount of the variance in
turnover when controlling for the first block (age, race, job satisfaction, and work
position). The two independent blocks explained 18% of the variance in turnover.
Block 2 (out-degree) only accounted for 2% of the variance in above Block 1 (b.06,
The regression equation for peer in-degree explained more of the total variance
compared to peer out-degree, but the model with peer in-degree still failed to reach
statistical significance (R
.23). Peer in-degree explained 7% of the variance in
turnover over and above employee personal characteristics (b.30, p.09). Thus,
the data failed to support Hypothesis 1. Statistics for all four logistic regression
equations are reported in Table 2.
Friendship networks. Partial support was found for Hypothesis 2: centrality in one’s
friendship out-degree network significantly predicted turnover and explained almost
20% of the unique variance (R
D.20, b.19, pB.02). Individuals who reported
having more friends in the workplace were less likely to have left after three months.
The total model that includes friendship out-degree centrality predicted 35% of the
variance in turnover. The total model for friendship in-degree networks did not
predict a significant amount of turnover (R
D.03, b.12, p.31). Again, the
correlation between the two measures of centrality in the friendship network was low,
at only .39. These two variables are not collinear.
Comparing peer and friendship networks indicates support for Hypothesis 3:
friendship networks predicted more of the variance in turnover than did peer
Hypothesis 4 predicted that greater relational closeness to others in the social
network would predict employee turnover. Specifically, it was argued that greater
64 T. H. Feeley et al.
Table 1 Descriptive Statistics and Zero-Order Correlations Between Pairs of Factors
Factor Mean (SD) 23456789
1. Peer in-degree 8.33 (3.14) .32* .04 .13 .08 .05 .17 .04 .17
2. Peer out-degree 8.33 (6.71) .40* .03 .18 .38* .17 .36* .38*
3. Friendship in-degree 10.33 (4.28) .39* .10 .37* .17 .73* .85*
4. Friendship out-degree 10.36 (9.66) .10 .31 .38* .40* .31*
5. Job satisfaction 4.69 (1.48) .02 .24 .27 .20
6. Age 21.85 (7.49) .08 .39* .22
7. Turnover .30 .02 .18
8. Time 1.71 (.40) .79*
9. Relational closeness 2.59 (0.57)
Turnover and Networks 65
closeness would engender a lower likelihood of turnover. There was no support for
Hypothesis 4: perceived relational closeness was unrelated to turnover (R
This study extends the research into social networks and organizational turnover to
non-egocentric networks and provides a more nuanced view of the type of social
support that is predictive of turnover. Results from the current study amassed with
results from two previous studies (Feeley, 2000; Feeley & Barnett, 1997) provide a
reliable estimate of the relationship between social network position and employee
turnover. In general, employees who are more active in their social networks are
significantly more likely to remain in the workplace.
Two important elements in the current research design warrant attention. First,
friendship networks predicted a significant amount of the turnover variance while
peer networks failed to predict an appreciable amount of the variance when
controlling for other factors. Second, it appears that the number of friends is
more important than the closeness of the friendship relations when predicting
This last observation is important to consider when revisiting a theoretical
model of why friendship networks account for turnover. It has been theorized that
friends in the workplace provide coping resources that serve to reduce the amount
of strain felt by an employee. The current findings indicate that having a number
of different support lines is more important than having one or two close friends
to lean on when the vicissitudes of part-time employment stress an employee.
With different shifts, different hours, and different responsibilities in the
workplace, it makes sense that a more flexible and more available network of
social support is important.
Table 2 Results of Hierarchical Logistic Regression Analyses Predicting Turnover
Independent blocks Peer-in Peer-out Friend-in Friend-out
Block 1: Controls
Age 1.03 1.04 0.96 0.89
Job satisfaction 0.56 0.73 0.63 0.54
Work position 4.03 3.21 1.68 0.35
Race 0.59 1.80 1.87 1.68
Block 2: Centrality 0.74* 1.06 0.89 0.83**
DBlock 2 .07 .02 .03 .19
total .23 .18 .19 .35
All statistics are reported as odds ratios except R
statistics. *pB.10; **pB.05. Peer-inPeer
in-degree; Peer-outPeer out-degree; Friend-inFriend in-degree; Friend-outFriend out-
for Block 1.164.
66 T. H. Feeley et al.
Determinants of Network Position
Actors who are recalled and reported as friends by fellow coworkers (i.e., individuals
with higher social in-degree scores) are more likely to report spending more time
with colleagues (r.73) and have greater self-reported relational closeness with
colleagues (r.85). The magnitudes of the two relationships were remarkable,
and these data were admittedly serendipitous. Given these findings, it would appear
inconsistent with the proposed model that employees with greater prestige would not
be more likely to stay at their job, as these prestigious employees would seem to have
a strong support network to rely on inside and outside the network. At a minimum,
the finding that those individuals who are more often attributed as friends by others
also self-report spending more time and investing greater personal involvement in the
other is an interesting validity concern for network scholars who study ego networks
(i.e., networks measured at the individual level) such as Feeley (2000, 2003) and Parks
(1995). The current study identified distinct social networks when comparing in-
degree and out-degree peer and friendship networks (r.39). Examination of the
standard deviations of the network measures indicates more variance or measure-
ment error in out-degree links than in in-degree links.
The current data also suggest that relational ties forged in the workplace may often
carry over to other contexts. Certainly, it could be the case that employees may have
been friends or acquaintances before working together (Wellman & Frank, 2001).
However, the current study failed to measure the possibility that coworkers knew one
another before working together (perhaps through school or previous work).
Those more central in the social network were younger in age. The correlation
between age and social in-degree (r.37, pB.05) and between age and social
out-degree (r.31, p.051) was significant or near-significant. Perhaps those
who are younger have more time to socialize and develop new friendship relations,
while those who are older may be more likely to be involved in a committed
relationship or have family responsibilities. Another possibility is that older
employees may have established friendships outside of work that may or may not
provide social support; recall that age, by itself, was unrelated to turnover.
There are several other factors that may determine social or friendship network
position, such as individual characteristics, network size, possession and delivery of
social resources, homophily with others, and the formal organizational structure
(Wellman & Frank, 2001). Further, because the network data were gathered at a single
point in time, it cannot be determined whether employees proactively become
involved in their networks or whether they were ‘‘pushed’’ to the periphery by the in-
group. Two research studies (Hess, 2000; Sias & Perry, 2004) suggest that employees
on the periphery may choose not to participate or may be ostracized to the outskirts
by employees who seek to disengage from them. Borrowing from Heider’s (1958)
seminal work on balance theory, Hess found that individuals involved in involuntary
relationships may often disengage, withdraw, or avoid the other individual in the unit
relation. Avoidance becomes a strategy to cope with the imbalanced social relation
represented in the actor, the other, and the workplace. A second strategy may be to
Turnover and Networks 67
become even more involved in one’s social network (that does not include the
It is interesting to note that the current study replicates findings in Feeley (2000)
that indicate a near-zero relationship between attitudes toward work and network
centrality. Feeley and Barnett (1997) suggested that those more central are more likely
to be committed to the workplace or position. Two studies have now failed to bear
out this proposed relation; it may be more likely that those more central are more
loyal or committed to the relationships and friends who are all ‘‘in it together’’ in the
workplace. Many of the worst jobs are made bearable and often quite memorable
when the workplace is filled with some friendly and like-minded employees.
Study Limitations and Future Research
The current data are taken from an organization that employs mostly part-time
employees (usually students) and is accustomed to high turnover. Turnover in the
organization is a function both of the job and of the opportunities outside the
workplace. The owner of the organization suggested that there are many opportu-
nities for part-time employment in the community at the wage his/her organization
offers; many employees find it easy to change jobs even if fired. Thus, the current
organization may not be a typical organization. The turnover studies to date that
have considered turnover as a function of network position (e.g., Feeley & Barnett,
1997; Krackhardt & Porter, 1986; Mossholder et al., 2005) have investigated
organizations that feature a high amount of turnover as there is more variance to
explain and perhaps a greater opportunity to get buy-in and cooperation from
management for a research study. It may be that betweenness and closeness are more
appropriate measures of network centrality at an organization with less turnover and
a greater proportion of professionals.
There may also have been a certain amount of measurement error with satisfaction
with work and relational closeness; both were measured with single-item instruments.
The decision to use single-item measures was based upon time. It was imperative that
employees take no longer than 15 minutes to complete the survey, as employees
completed the survey in work time*which can add up in terms of cost to the
employer. Thus, the balance between more reliable, multi-item measures and
respondent fatigue and attrition should be considered in future research.
The current study identified that 12 individuals left the organization and that nine
of these individuals had been fired from their position. However, conversations with
management clarified that identifying leavers as involuntary turnover may be
inaccurate as many of the nine simply stopped coming to work. Thus, future
research could benefit from a more fine-grained measure of employee turnover and
include this in the research design as a dependent factor. Exit interviews might also
uncover from the individual why s/he chose to leave (or why s/he was terminated);
questions could be posed about the employee’s relationships, both peer and
friendship, within the workplace.
68 T. H. Feeley et al.
The length of time between network measures and the measurement of turnover
was three months; other studies have examined turnover at six months (Feeley &
Barnett, 1997) or even five years (Mossholder et al., 2005). Direct comparison of any
two studies is not possible without considering the type of organization and other
possible moderating factors. For example, the current study found that 30% of
employees left after data collection and before 90 days when turnover was measured.
Compare these data to Mossholder et al. (2005) who found 35% of employees to
leave before measuring turnover 1,825 days after initial data collection. Also,
Mossholder et al. studied healthcare employees not fast-food employees, many of
whom are not ‘‘career’’ employees.
Managers seeking to reduce turnover should pay attention to the current study and
the corpus of research in this area of inquiry. The results reported here are of special
importance to retail service companies that employ young, part-time workers, such as
fast food outlets, supermarkets, or landscaping businesses. First, managers should
attempt to create a denser network of social relations in the workplace; that is, more
people need to speak to more people about work- and non-work-related topics.
Moreover, increasing network links among employees should not come at the price of
decreased employee and organizational productivity. Employers may choose to create
network groups (Friedman & Holtom, 2002) or use other formal techniques to
increase employees’ assimilation into the workplace (Jablin, 2001). Formal and
informal mentoring are possibilities, as is appropriate long-term training of new
employees; the mentor can direct messages to the ‘‘rookies’’ to increase their in-
degree centrality in the network. Less formally, managers can be more vigilant
observers of communication networks in the workplace and pay special attention to
employees who may be socially isolated from the social network, while understanding
that many individuals may choose this isolation. If the less central individuals are
desired as long-term employees, managers should take the initiative to talk to the
relative isolates about network-related topics (social support) to increase their in-
degree centrality in the network.
In the early stages of organizational assimilation, there are two opportunities to
integrate the employee into the organizational network. First, during employee
orientation, a new employee can be made to feel comfortable and have an
opportunity to interact with current employees in a non-work setting, such as
during the orientation session, which might be located off-site. Also, the organization
should use employees who are at the same level to orient and introduce new
employees to the workplace. As an analogy, many rookies on professional sports
teams are paired with one another to ‘‘learn the ropes’’ of the team and league.
Newcomers may be better able to relate and connect with someone at their level of
A second opportunity may arise during employee training, which may occur in
parallel with orientation or after orientation. Jablin (2001) estimates that one-third of
Turnover and Networks 69
all new employees undergo formal training programs. The content of training may be
considered in light of the current results. Clear and realistic expectations for job
duties and responsibilities are important considerations. Training efforts need not
end upon employee entry into the organization; research suggests that senior
executives are now just as likely as entry-level employees to be the target of training
initiatives (Martocchio & Baldwin, 1997). The importance of training throughout
one’s lifespan in an organization is related to an employee’s attempt to individualize
or change/adapt his or her roles based upon the work environment for a better fit. It
is thought that the better the fit between work roles and work environment, the more
likely an employee is to satisfy his or her values, attitudes, and needs related to
employment (Jablin, 2001).
It should be noted that employees who are more integrated, active, or socialized in
the workplace not only stay longer, but may also perform better (Baldwin, Bedell, &
Johnson, 1997; Marshall & Stohl, 1993). Future research and managers could not only
examine turnover and retention, but also examine employee productivity and
success, using conventional organizational indicators (e.g., sales, productivity,
absenteeism, and citations).
Finally, employers might strategically use central actors or emergent leaders in the
network to communicate news and critical information rather than using formal
networks such as memoranda or postings. Perhaps news from close or respected
others at the same level is received more favorably than information coming directly
from management. Of course, there is a concern that central actors will inevitably
interpret information when communicating it to colleagues and friends.
This study adds to our understanding of the role of communication in organizational
turnover. It examined employee turnover at a fast-food restaurant. Twelve (i.e., 30%
of roster) individuals had left their position by the time three months had elapsed,
and employees who reported a higher level of out-degree network centrality were
more likely to remain at their position. A social support model was proposed to
account for the data, and posits that employees with a greater number of friendship
ties have greater coping resources to buffer the strain of workplace stressors.
 Research has shown (e.g., Feeley & Barnett, 1997; Mossholder et al., 2005) that measures of
centrality, while conceptually distinct (Freeman, 1979), are highly correlated. In the current
study, friendship out-degree was correlated signiﬁcantly with betweenness (r.74) and
closeness (r.45); friendship in-degree was correlated signiﬁcantly with betweenness (r
.65), closeness (r.99), and friendship out-degree (r.39). Betweenness was measured
using Borgatti, Everett, and Freeman’s (1992) operationalization, which considers the
number of pairs of links an individual goes between. Borgatti et al.’s measure of closeness was
also used; closeness measures the minimum number of links required to get to all others in
the network (see Freeman, 1979, for conceptualization; and Freeman et al., 1979, for
70 T. H. Feeley et al.
 The reciprocity rate for the current study was .30, as determined by UCINET. This indicates
a low level of reciprocity between out-degree and in-degree links. Thus, in almost 70% of
reported links, the focal other failed to report the same level of relationship as the ego. It is
necessary to examine in- and out-degree networks separately, as overall degree typically
‘‘forces’’ reciprocation (e.g., Feeley, 2000; Monge et al., 1983). The authors acknowledge an
anonymous reviewer for this methodological improvement.
 Two additional logistic regressions were run using closeness and betweenness measures of
centrality in friendship networks. High betweenness suggests that an individual has a high
level of control of communication in the workplace, while closeness represents efﬁciency in
communication, as one who has greater closeness can get and send information more quickly
to others. A person with high betweenness may be considered a liaison, a gatekeeper, or a
connector or linchpin. This individual ﬁlls a structural hole (Burt, 1992); that is, s/he
connects two or more clusters or regions of density in the network. In this case, s/he has
friends in different groups of people. Thus, it would be expected that individuals with high
betweenness would be likely to give and receive social support and, therefore, remain an
organizational member. Closeness as thus deﬁned is unrelated to the measure of relational
closeness used in the employee survey; network closeness examines how close one is to all
others in the social network. Both analyses used hierarchical logistic regression with
individual-level factors included as the ﬁrst block (age, job satisfaction, job position, and
race) and betweenness or closeness as the second block. Closeness centrality did not predict
D.02, n.s. Betweenness centrality in friendship networks signiﬁcantly
predicted turnover, R
D.22, pB.05, OR.90 (95% CI .82, .99).
Abelson, M. A., & Baysinger, B. D. (1984). Optimal and dysfunctional turnover: Toward an
organizational level model. The Academy of Management Review,9, 331341.
Apker, J., & Ray, E. B. (2003). Stress and social support in health care organizations. In
T. Thompson, A. Dorsey, K. Miller, & R. A. Parrott (Eds.), Handbook of health communication
(pp. 347368). Mahwah, NJ: Lawrence Erlbaum.
Baldwin, T. T., Bedell, M. D., & Johnson, J. L. (1997). The social fabric of a team-based M.B.A.
program: Network effects on student satisfaction and performance. Academy of Management
Bavelas, A. (1950). Communication patterns in task oriented groups. Journal of the Acoustical
Society of America,22, 272283.
Berman, E. M., West, J. P., & Richter, Jr., M. N. (2002). Workplace relations: Friendship patterns and
consequences (according to managers). Public Administration Review,62, 217230.
Borgatti, S., Everett, M., & Freeman, L. C. (1992). UCINET IV Version 1.0. Columbia, SC: Analytic
Borgatti, S., Everett, M., & Freeman, L. C. (2002). UCINET VI: Network analysis software. Columbia,
SC: Analytic Technologies.
Brotheridge, C. M. (2001). A comparison of alternative models of coping: Identifying relationships
among coworker support, workload, and emotional exhaustion in the workplace. Interna-
tional Journal of Stress Management,8,114.
Bullis, C., & Bach, B. W. (1991). An explication and test of communication network content and
multiplexity as predictors of organizational identiﬁcation. Western Journal of Speech
Burt, R. S. (1992). Structural holes, the social structure of competition. Cambridge, MA: Harvard
Dalton, D. R., Krackhardt, D. M., & Porter, L. W. (1981). Functional turnover: An empirical
assessment. Journal of Applied Psychology,66, 716721.
Turnover and Networks 71
Dalton, D. R., & Todor, W. D. (1982). Turnover: A lucrative hard holler phenomenon. The Academy
of Management Review,7, 117123.
Doerfel, M. L., & Taylor, M. (2004). Network dynamics of interorganizational cooperation: The
Croatian civil society movement. Communication Monographs,71, 373394.
Eisenberg, E. M., Monge, P. R., & Miller, K. I. (1983). Involvement in communication networks as a
predictor of organizational commitment. Human Communication Research,10, 179201.
Everett, M. G., & Borgatti, S. P. (2004). Ego network betweenness. Social Network,27,3138.
Feeley, T. H. (2000). Testing a communication network model of employee turnover based on
centrality. Journal of Applied Communication Research,28, 262277.
Feeley, T. H. (2003). Using the theory of reasoned action to model retention in rural primary care
physicians. The Journal of Rural Health,19, 245251.
Feeley, T. H., & Barnett, G. A. (1997). Predicting turnover from communication networks. Human
Communication Research,23, 370387.
Freeman, L. C. (1979). Centrality in social networks: Conceptual clariﬁcation. Social Networks,1,
Freeman, L. C., Roeder, D., & Mulholland, R. R. (1979). Centrality in social networks: II.
Experimental results. Social Networks,2, 119141.
Friedman, R. A., & Holtom, B. (2002). The effects of network groups on minority employee
turnover intentions. Human Resource Management,41, 405421.
Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology,78, 13601380.
Hartman, R. L., & Johnson, D. J. (1989). Social contagion and multiplexity: Communication
networks as predictors of commitment and role ambiguity. Human Communication Research,
Heider, F. (1958). The psychology of interpersonal relations. New York: John Wiley and Sons.
Hess, J. A. (2000). Maintaining nonvoluntary relationships with disliked partners: An investigation
into the use of distancing behavior. Human Communication Research,26, 458488.
Jablin, F. M. (1987). Organizational entry, assimilation, and exit. In F. M. Jablin, L. L. Putnam, K. H.
Roberts, & L. M. Porter (Eds.), Handbook of organizational communication: An interdisci-
plinary perspective (pp. 679740). Newbury Park, CA: Sage.
Jablin, F. M. (2001). Organization entry, assimilation, and disengagement/exit. In F. M. Jablin & L.
L. Putman (Eds.), The new handbook of organizational communication: Advances in theory,
research, and methods (pp. 732818). Thousand Oaks, CA: Sage.
Johnson, J. T., Griffeth, R. W., & Grifﬁn, M. (2000). Factors discriminating functional and
dysfunctional salesforce turnover. Journal of Business and Industrial Marketing,15, 399415.
Katz, D., & Kahn, R. L. (1978). The social psychology of organizations. New York: Wiley.
Krackhardt, D. M., & Porter, L. W. (1985). When friends leave: A structural analysis of the
relationship between turnover and stayers’ attitudes. Administrative Science Quarterly,30,
Krackhardt, D. M., & Porter, L. W. (1986). The snowball effect: Turnover embedded in
communication networks. Journal of Applied Psychology,71,5055.
Kram, K. E., & Isabella, L. A. (1985). Mentoring alternatives: The role of peer relationships in career
development. Academy of Management Journal,28, 110132.
Leavitt, H. J. (1951). Some effects of certain communication patterns on group performance.
Journal of Abnormal and Social Psychology,46,3850.
Lee, R. T., & Ashforth, B. E. (1996). A meta-analytic examination of the correlates of the three
dimensions of job burnout. Journal of Applied Psychology,81, 123133.
Marshall, A. A., & Stohl, C. (1993). Participating as participation: A network approach.
Communication Monographs,60, 137157.
Martocchio, J. J., & Baldwin, T. T. (1997). The evolution of strategic organizational training: New
objectives and research agenda. In G.R. Ferris (Ed.), Research in personnel and human
resources management (pp. 146). Greenwich, CT: Elsevier Science.
72 T. H. Feeley et al.
Metts, S., Geist, P., & Gray, J. L. (1994). The role of relationship characteristics in the provision and
effectiveness of supportive messages among nursing professionals. In B. R. Burleson,
T. L. Albrecht, & I. G. Sarason (Eds.), Communication of social support: Messages, interactions,
relationships, and community (pp. 229245). Thousand Oaks, CA: Sage.
Mobley, W. H. (1982). Some unanswered questions in turnover and withdrawal research. The
Academy of Management Review,7, 111116.
Monge, P. R., & Contractor, N. S. (1999). Emergence of communication networks. In F. M. Jablin &
L. L. Putnam (Eds.), Handbook of organizational communication (2nd ed., pp. 441501).
Thousand Oaks, CA: Sage.
Monge, P. R., Edwards, J. A., & Kirste, K. K. (1983). Determinants of communication network
involvement: Connectedness and integration. Group and Organization Studies,8,83111.
Mossholder, K. W., Settoon, R. P., & Henagan, S. C. (2005). A relational perspective on turnover:
Examining structural, attitudinal, and behavioral predictors. Academy of Management
Mullen, B., Johnson, C., & Salas, E. (1991). Effects of communication networks structure:
Components of positional centrality. Social Networks,13, 169186.
Parks, M. R. (1995). Webs of inﬂuence in personal relationships. In C. R. Berger & M. Burgoon
(Eds.), Communication and social inﬂuence processes (pp. 155178). East Lansing, MI:
Michigan State University Press.
Pfeffer, J. (1991). Organization theory and structural perspectives on management. Journal of
Rogers, E. M., & Kincaid, D. L. (1981). Communication networks: Toward a new paradigm for
research. New York: Free Press.
Shah, P. P. (1998). Who are employees’ social referents? Using a network perspective to determine
referent others. Academy of Management Journal,41, 249268.
Shah, P. P. (2000). Network destruction: The structural implications of downsizing. Academy of
Management Journal,43, 101112.
Shumante, M., Fulk, J., & Monge, P. (2005). Predictors of the HIV-AIDS INGO network over time.
Human Communication Research,31, 482510.
Sias, P. M., & Cahill, D. J. (1998). From coworkers to friends: The development of peer friendships
in the workplace. Western Journal of Communication,62, 273299.
Sias, P. M., & Perry, T. (2004). Disengaging from workplace relationships. Human Communication
Staw, B. M. (1980). The consequences of turnover. Journal of Occupational Behaviour,1, 253273.
Susskind, A. M., Miller, V. D., & Johnson, J. D. (1998). Downsizing and structural holes: Their
impact on layoff survivors’ perceptions of organizational chaos and openness to change.
Van Maanen, J. (1978). People processing: Strategies of organizational socialization. Organizational
Walker, M. E., Wasserman, S., & Wellman, B. (1994). Statistical models for social support networks.
In S. Wasserman & J. Galaskiewicz (Eds.), Advances in social network analysis: Research in the
social and behavioral sciences (pp. 5378). Thousand Oaks, CA: Sage.
Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications. Cambridge,
UK: Cambridge University Press.
Wellman, B., & Frank, K. (2001). Network capital in a multi-level world: Getting support from
personal communities. In N. Lin, K. Cook, & R. Burt (Eds.), Social capital: Theory and
research (pp. 233273). Chicago: Aldine De Gruyter.
Turnover and Networks 73