Predicting Exit, Voice, Loyalty, and Neglect

Author(s): Michael J. Withey and William H. Cooper

Source: Administrative Science Quarterly, Vol. 34, No. 4 (Dec., 1989), pp. 521-539
Published by: Johnson Graduate School of Management, Cornell University
Stable URL: http://www.jstor.org/stable/2393565
Accessed: 04/08/2009 09:05

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=cjohn.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.

Johnson Graduate School of Management, Cornell University is collaborating with JSTOR to digitize, preserve
and extend access to Administrative Science Quarterly.

http://www.jstor.org
PredictingExit,Voice, This paper reports the results of two longitudinal studies
Loyalty,and Neglect we conducted to find out when dissatisfied employees
will respond to their dissatisfaction with exit, voice, loy-
alty, or neglect. We found consistent evidence that exiters
Michael J. Withey were affected by the costs and the efficacy of their re-
MemorialUniversityof sponses as well as the attractiveness of their employing
Newfoundland organization. Loyalists were primarily affected by the effi-
cacy of their responses, although, unexpectedly, loyalty
William H. Cooper resembled entrapment in the organization more than it
Queen's University did supportive allegiance to the organization. Neglecters
were primarily affected by the costs and the efficacy of
their responses, and voicers were very difficult to predict.
We conclude by arguing that we need a much better un-
derstanding of voice and loyalty in order to predict better
how employees will respond to dissatisfaction.'
INTRODUCTION
People who are unhappy at work can do a nqmber of things.
They can focus attention on their nonwork interests, doing
nothing about their work situation (neglect). They can work to
improve the situation, through voice. They can find a better
job and quit (exit). Or they can stay and support the organiza-
tion, responding with loyalty. This study is about when dis-
satisfied employees will make these different choices.
Systematic thinking about exit, voice, and loyalty as re-
sponses to dissatisfaction had its start in Hirschman's (1970)
seductive treatment of them. He argued that firms, organiza-
tions, and states recover from declines in their fortunes to the
extent that exit-permanent movement away from the orga-
nization-and voice-attempting to improve it-bring about
change. Hirschman described exit and voice as the main eco-
nomic and political alternatives when firms, organizations, and
states are in decline. Suppliers of consumer goods can re-
ceive signals that there is a problem in two ways: dissatisfied
customers can either exit by switching to a different supplier
or voice their complaints. Messages are sent to governments
by citizens who either exit by emigrating or voice their dis-
content by working to elect a new government.
While Hirschman's arguments were developed to explain the
responses of organizations to decline, they can prove useful
in understanding how individuals may act when things are not
going well. Thus, in Rusbult, Zembrodt, and Gunn's (1982)
extension of Hirschman's categories to romantic involve-
ments, people who are unhappy can leave their lover or talk
? with their partner about problems with their relationship. In
1989 by CornellUniversity.
0001-8392/89/3404-0521/$1.00. our study dissatisfied employees could look for another job
0 and quit or they could talk with their supervisor and organize
This paperis based on a dissertation others who were unhappy.
completedby the firstauthorunderthe
supervisionof the second. We thankHugh Exit and voice are two conceptually distinguishable responses
Arnold,JulianBarling,CarolBeatty,John to dissatisfaction in settings where individuals and collections
Dowling,Vic MacDonald,and Don Night- of individuals don't like the way things are going. Exit and
ingalefor theirhelp at the thesis stage,
two very thoughtfulASO reviewers,Dan voice can be independent, as when an unhappy person just
Farrelland Janet Nearfor theircomments leaves, or sequential, as when someone follows through on a
on an earlierversionof the paper,Tom
Mahaffeyfor a timelyquestion,andTed threat to quit when the offending parties' response to voice is
Neave for gettingthingsstarted.Financial inadequate. Responses can also co-occur, as in "noisy" exits
supportto the second authorwas pro- (Barry, 1974; Birch, 1975). Hirschman is clear about the exit
vided by the SocialSciences and Human-
ities ResearchCouncilof Canada and voice responses, and they have been readily operational-
(SSHRCC494-88-1015). ized by others, such as Freeman and Medoff (1984), on voice
521/AdministrativeScience Quarterly,34 (1989): 521-539
as union membership, and Spencer (1986), on the relationship
between voice and exit.
Less clear is what Hirschman (or anyone else) means by loy-
alty. He introduces the construct as a way of explaining why
anyone would use voice when exit is available (Hirschman,
1970: 77). He describes loyalty as the product of (mostly
economic) factors that tie the individualto the organization,
making exit costly and undermining voice. Sometimes loyalty
is portrayed in his work as an attitude that moderates or con-
ditions the use of exit and voice. At other times he describes
loyalty as a behavior in which the individualdoes things that
support the organization. These multiple meanings led Barry
(1974) to argue that Hirschman's loyalty is a filler concept that
is poorly developed.
Treatments of loyalty by others, following Hirschman, have
been similarly varied. For example, treatments of loyalty
range from Rusbult et al's (1988) operationalization of loyalty
as a passive constructive behavior (using items such as being
quietly supportive and being patient) to Kolarskaand Aldrich's
(1980) work, which, rather than referring to the response as
loyalty, talks about "silence." These understandings of loyalty
reflect the ambiguity of the construct. "Loyalty" in ordinary
language includes attitude-like and behavior-like components,
and within the behavior-like component there are both active
elements (doing things that are supportive) and passive ele-
ments (being quiet).
Rusbult and her colleagues have also added a fourth response
to dissatisfaction: neglect. The addition of neglect may help
to cover more adequately the range of responses to dissatis-
faction. Neglect differs from loyalty in that it is not directed at
recovery of the relationship. Rather, the individualresponding
with neglect implicitly accepts that recovery is not going to
happen. Neglect may be shown by putting in less effort, not
working at a relationship, and letting it fall apart (Rusbult,
Zembrodt, and Gunn, 1982). In a work setting it may include
spending less time at work and expending less effort when
there.
Thus exit, voice, loyalty, and neglect (the EVLNresponses)
are four ways of responding to dissatisfying situations. They
are both conceptually and empirically distinguishable (Farrell,
1983; Rusbult and Zembrodt, 1983), even though the bound-
aries between them may be imprecise. They are meant to
provide a useful set of categories for thinking about how
people respond to dissatisfaction in many domains. We set
out to understand what predicts each of them.

LIKELYPREDICTORSOF EVLN
The theory of -action that guides this study is that dissatisfied
employees are sensitive to three things when they think
about what to do: the cost of the action, the efficacy of the
action, and the attractiveness of the setting in which the ac-
tion occurs. Most generally, people will perform an action
when the cost of the act is low for the individual, the action
seems likely to be helpful to the individual,and the setting is
an attractive one in which to incur the costs. Conversely,
people will avoid an act when it is costly and pointless and
they thinkthat the setting is unattractive.Inthis theory,
522/ASQ, December 1989
Predicting EVLN

people are depicted as implicitly weighing possible payoffs

against likely costs, factoring in how positive they feel toward
their current employer (or supplier, country, lover, etc.).
We have translated cost, efficacy, and setting into more spe-
cific terms. We examined the impact on the four responses
of two indirect costs of action (exit costs and voice costs),
how happy people were before they became dissatisfied,
whether they think improvement is likely, whether they think
there is much connection between what they do and what
happens to them, how committed they are to the organiza-
tion, and what alternative jobs are available to them.

Cost of the Responses

Behaviors have both direct and indirect costs. The direct
costs are the time and energy expended in them. The indirect
costs are the unpleasant things that flow from the actions.
The indirect costs of exit include lost income and pension
benefits, lost psychological investment, and skills that atrophy
-when they aren't transferable. The indirect costs of voice in-
clude the possibility of retaliation, loss of reputation, and the
emotional costs of confronting people with power. The indi-
rect costs of loyalty and neglect may be more subtle. For ne-
glect, they include the costs of possible discipline, blocked
opportunities, loss of reputation, and a sense of failure. For
some forms of loyalty there are the emotional costs of feeling
that something more active may be called for. All of these
seem likely to affect what action is taken. In the interest of
doing something both helpful and feasible, we chose to ex-
amine two of these indirect costs: exit costs and voice costs.
We expect that the higher the costs of the responses, the
more likely individuals would be to choose the alternative re-
sponses.

Efficacy of the Responses

Prior satisfaction. Using prior level of satisfaction as a mea-
sure is one way in which individuals can judge the possible
efficacy of action. If there was a time when people were
happier in their job, they have at least one reason to think that
recovery is possible. We expect that individuals who re-
member being happier in their current job will be more likely
to stay and be supportive or work to improve things than
would people who don't have this basis for hope. When no
such basis for hope exists, we expect that individuals won't
work to improve things, choosing instead to withdraw into
neglect and/or exit. This is identical to the predictions made
by the investment model tested by Rusbult et al. (1988).

Possibility of improvement. We expect that people who be-

lieve improvement is possible will be more likely to work at
voice and/or loyalty. Such optimism may be based on the be-
lief that the culture is supportive of change, rather than op-
poses it, and that the person has good relations, rather than
bad ones, with his or her supervisor. Optimism may also be
based on the experience of an earlier period of satisfaction
that may be recovered, although prior satisfaction alone is not
sufficient for belief that improvement is possible. People who
don't share this optimism may be more likelyto choose ne-
glect and/orexit.
523/ASQ, December 1989
I
Many interaction effects can be imagined.
While some were found in exploratory
testing (Withey, 1986), we think the rudi-
mentary nature of the model makes the
specification of interactions premature.
Predicting EVLN

Predictions
Exit. We expect that thinking about leaving, organizing a
search, looking for another job, and quitting will be most likely
when the costs of exit are low and the costs of voice are
high, prior satisfaction is low and improvement is not likely,
organizational commitment is low, an attractive alternative job
is available, and the person's locus of control is internal. This
is a sketch of exiters as men and women of action who have
few ties to their current organization and who see leaving as
easier and more efficacious than staying.
Voice. We expect that people will talk with their supervisors
and others about working to improve dissatisfying conditions
when the costs of exit are high and the costs of voice are
low, prior satisfaction is high and improvement seems likely,
organizational commitment is high, and the person's locus of
control is internal. Because the prediction about the effects of
alternatives is equivocal-it could go either way-we're
treating the relationship between alternatives and voice as an
empirical question. Hence, our initialpicture of voicers is that
of people of action for whom exercising voice is cheaper than
leaving, who are tied to the organization, and who have
reason to believe that things will get better if they voice their
dissatisfaction.
Loyalty. The high costs of exit and voice, a high level of prior
satisfaction, the high likelihood of improvement, and high or-
ganizational commitment are all likely to promote support of
the organization, as will the lack of alternative jobs and an ex-
ternal locus of control. This is a portraitof the loyalist as a
passive person with strong ties to the current setting who
thinks that acting is costly and believes that things will get
better on their own.
Neglect. We expect that neglecting work is most likely when
the costs of exit and voice are both high, prior satisfaction
and the likelihood of improvement are both low, organiza-
tional commitment is low, alternative jobs are available, and
people have an external locus of control. This outline of the
neglecter is one of a passive person who thinks that action is
costly and useless and who thinks things are better else-
where.
We tested these predictions using longitudinal data collected
from two samples in 1984. We first tested the predictions
using data from respondents in the larger of the two samples
and then used data from the smaller sample to assess the
extent to which the results could be replicated.

METHOD
Samples
Data from the larger sample were collected in 1984 through
a survey mailed to 1,000 randomly selected graduates of the
Queen's University Bachelor of Commerce program who
lived in North America. One hundred were chosen from each
of the previous ten graduating years. Eighty-five graduates
were excluded because they had moved and left no for-
warding address or were no longer employed full time. Of the
915 potentialrespondents, 460 returnedtheir questionnaires
at Time 1 (Ti). This is an effective response rate of 50.3 per-
525/ASQ, December 1989
cent. The Time 2 (T2) questionnaires were mailed six months
later. Nine were returned undelivered. Of the remaining 451
people, 368 returned their questionnaires, representing an ef-
fective response rate of 80 percent of the 460 T1 respon-
dents and 40.6 percent of the net T1 mailing. Thirteen of the
368 responses received at T2 were deleted because the indi-
viduals were no longer working full time, reducing the T2
graduate sample to 355. The sample was further reduced to
303 when we excluded respondents who had spent less than
six months on their jobs at T1 and/or who described them-
selves as owners, partners, or company presidents.
There was no significant difference between T1 respondents
and nonrespondents on sex (X2 = .07, n.s.), but respondents
did differ from nonrespondents on years since graduation (X2
= 21.42, p < .01). More recent graduates were overrepre-
sented among the respondents. The 92 T2 dropouts and the
368 T2 respondents did not differ on any of the T1 measures
(p > .05). Demographic characteristics for this sample are re-
ported in Table 1. The respondents were educated, young,
mostly male, and nonmanagers/professionals in the early
years of their careers.

Table 1
Demographic Characteristicsof Respondents Who Completed Usable
Questionnaires at both T1 and T2*
Sample
Demographic characteristics Graduates Accounting Firm
Numberof cases 303 134
% Male/female 66/34 53/47
Averageage 27.5 30.7
(3.3) (9.0)
Averagetenure in months 38 67
(31) (72)
% Manager/professional 33.0 10.4
% Completedan
undergraduatedegree 99.7 49.6
* Alldata are for T1. Standarddeviationsare in parentheses.

The smaller sample was drawn from a mid-size accounting

firm with offices throughout Ontario, Canada. Employees in
18 offices completed the questionnaire twice, with six
months between administrations. With a few exceptions, all
employees who were available completed the questionnaires.
One hundred and fifty-one individuals voluntarily completed
the questionnaire at T1. Ten quit before T2. Of the remaining
141 employees, 124 completed the questionnaire at T2,
yielding an effective T2 response rate of 87.9 percent. Time
2 respondents and nonrespondents differed on one T1 mea-
sure: the use of voice mechanisms (t = 2.16, p < .05);
nonrespondents used voice mechanisms less than did re-
spondents. Demographic characteristics for the 124 indi-
viduals who completed the questionnaire at T2 and the ten
who quit before T2 are also shown in Table 1. The ac-
counting-firm sample was less educated, older and longer
tenured, more nearly equally male and female, and less man-
agerial/professional than the sample of graduates. It comes
closer to resembling the population than does the sample of
graduates.
526/ASQ, December 1989
Predicting EVLN

Measures
Existing scales were used when they matched the constructs
in which we were interested and had respectable psychomet-
ric properties. When we couldn't find satisfactory scales, we
generated suitable items and pretested them on a sample of
58 accounting-firm respondents priorto T1. Problematic items
were then revised to increase the clarity of the items and the
internal consistency of the scales. Unless otherwise stated,
all measures employed 5-point Likert-type scales. Factor anal-
yses were used to clarify the structure of scales when new
scales or combined scales were used. Table 2 provides de-
scriptive statistics for the T1 measures of the independent
variables and T2 measures of the dependent variables for the
sample of graduates. Internal consistencies were above .70
(Nunnally, 1978) except in the three cases noted below.
Table2
Descriptive Statistics for T1 Independent Variables and T2 Dependent
Variables for Sample of Graduates
N of Cronbach
Variable items Mean S.D. alpha
Exitcosts
Skillspecificity 2 1.52 .68 .71
Sunkcosts 4 2.25 .72 .52
Investment 2 4.05 .84 .81
Voice costs 5 2.76 .90 .84
Priorsatisfaction 18 3.85 .74 .94
Possibilityof improvement 9 3.11 .74 .89
Betteralternatives 2 2.56 1.02 .90
Organizational commitment 15 3.54 .71 .90
Locus of control 23 13.91 3.93 .74
Exit 3 1.95 .96 .74
Voice 3 2.34 .63 .41
Loyalty 3 2.56 .92 .71
Neglect 3 1.34 .55 .55

Independent variables. Three measures of exit costs were

used: skill specificity and sunk costs, which are both aspects
of Becker's (1960) side bets, and investments. Skill specificity
refers to the extent to which skills learned on the job lack
transfer value to other jobs. It was measured by items such
as "My present job involves skills which would be useful in
many other organizations" (reverse-scored). Sunk costs refer
to the economic investments in the job that would be lost if
the person quit, measured by, for example, "If I left this job,
my pension plan losses would be significant." This scale had
low internal consistency (at = .52). The measure of invest-
ment was a modification of the scale developed by Farrelland
Rusbult (1981). Investment refers to the extent to which
people feel they have put something of themselves into the
job, for example, "In general, I have invested a great deal in
this job." Because factor analysis of data in the larger of the
two T1 samples showed that each of the three scales loaded
on its own factor, the three separate scales were retained.
Items used to measure voice costs asked about the effort re-
quired to bring about change and the likelihood of punitive re-
sponses to such efforts, for example, "In this office, there
are penalties for speaking up too loudlyabout work-related
problems."Factoranalysis identifieda unitaryfactorstructure.
527/ASQ, December 1989
 Job satisfaction was measured using Brayfield and Rothe's
(1951) 18-item scale. Representative items included "I con-
sider my job rather unpleasant" (reverse-scored) and "Most
days I am enthusiastic about my job."
Two aspects of belief in the possibility of improvement were
measured. One was original and one was borrowed. The new
scale was created to assess receptiveness to change. This
was done by asking about how supportive and open the
workplace was to change, for example, "This office is orga-
nized so as to make change as easy as possible." The bor-
rowed scale was Graen, Liden, and Hoel's (1982) measure of
leader/member relations, which uses 4-point response scales.
This scale assesses the quality of the employee's relationship
with his or her superior, which is a likely ingredient in beliefs
about the possibility of improvement. An example is "To what
extent can you count on your supervisor to 'bail you out' at
his/her expense when you really need help?" The nine items
that loaded on the first factor were combined to form a single
measure of belief in the possibility of improvement.
Locus of control was measured using Rotter's (1966) stan-
dard instrument. Items are forced-choice and ask which item
in a pair is most representative of the individual's thinking. For
example, respondents chose between "What happens to me
is my own doing" and "Sometimes I feel that I don't have
enough control over the direction my life is taking." Higher
scores indicate that the internal-control response was chosen
more often. Organizational commitment was measured using
Porter et al.'s (1974) scale, for example, by items such as "I
really care about the fate of this organization."2
The availabilityof alternative jobs was measured using a
subset of Price and Bluedorn's (1979) scale. The items asked
about the ease of finding a job in the same geographic area
that is better than or much better than the present one.

Dependent variables. The measures of the EVLNresponses

used in the analysis were obtained at T2 and were based on
Farrell's (1983) measures of the four responses, adapted to
refer to their incidence in the work setting. All used response
options from "have never considered this action" to "have
engaged in this action frequently." Three items were included
for each response. Examples include "Getting into action and
looking for another job," for exit, "Talkingto supervisor to try
and make things better," for voice, "Quietly doing my job and
letting higher-ups make the decisions," for loyalty, and
"Calling in sick and not dealing with what is happening," for
neglect. Problems exist with the internal consistency of the
voice measure (a( = .41), and, to a lesser extent, with the
2 neglect measure (a = .55). Such low internal consistency
We examinedthe possibilitythat overlap would be expected to attenuate the correlation coefficients
between our independentand dependent
variablesproducedartifactualrelationships involving these variables.
between them. Inthe case of locus of
controland neglect, there was no overlap An important precondition for the analysis is that the mea-
in the two measures. Inthe case of com-
mitmentand turnoverintentions,there sures of the four responses be empirically separable. The
was some overlap,but removalof the of- factor analysis shown in Table 3 indicates that the items yield
fendingitems fromthe commitmentscale a clean four-factor structure, with each of the twelve items
did not lead to any meaningfuldifference
in the results.We thereforedid not ad- loading on the appropriate factor. Thus, the four responses
dress the issue furtherin this paper. represent separate constructs.
528/ASQ, December 1989
 Predicting EVLN

Table 3

Varimax Rotation Factor Analysis of Items Used to Measure Dependent Variables at T2

Factor
Item 1 2 3 4

Exit:
Getting into action and looking for another job -.02 .89 .08 .08
Deciding to quit the company -.07 .86 .14 -.08
Getting myself transferred to another job .26 .60 .07 -.01
Voice:
Talking to supervisor to try and make things better -.09 - .04 -.20 .52
Putting a note in the suggestion box, attempting to correct a problem .11 .11 .04 .77
Writing a letter to a government agency to find out what can be done
about a problem .07 -.06 .11 .72
Loyalty:
Waiting patiently and hoping any problems will solve themselves .73 .02 .08 -.19
Quietly doing my job and letting higher-ups make the decisions .80 .08 .09 .07
Saying nothing to others and assuming things will work out .81 .02 .01 -.05
Neglect:
Calling in sick and not dealing with what is happening .05 .17 .77 -.02
Coming in late to avoid problems -.03 -.03 .82 -.02
Becoming less interested and making more errors .34 .27 .58 -.01

Additional Measures of EVLN

We obtained supervisory ratings and access to company
records for the employees in the accounting firm in order to
assess the construct validity of the measures of the EVLNre-
sponses. For exit, we obtained data on actual turnover at T2.
For voice, supervisors were asked at T2 to what extent each
of their subordinates had engaged in voicing activities. Be-
cause supervisors generally answered that most employees
were not active voicers, their responses were simply coded
as yes (14 percent) or no (86 percent) for each employee. For
loyalty and neglect, immediate supervisors provided a global
performance rating and access to company records on
number of hours of sick time used. We expected that loy-
alists would demonstrate their loyalty by staying away from
work less and that neglecters would have lower performance
ratings.
The T1 measure of exit was modestly correlated with actual
turnover at T2 (r = .25, p < .01). Voice at T2 was modestly
correlated with the supervisory classification of voice at T2 (r
= .18, p < .05). Loyalty at T2 was modestly negatively corre-
lated with the T2 measure of sick time used (r = -.19, p <
.05). Neglect at T2 was modestly negatively correlated with
the supervisor's rating of performance at T2 (r = -.21, p <
.01). Among the factors that depressed the correlations are
low base rates (for exit and voice), lack of time for intentions
to lead to behavior (for exit), the possibility that supervisors
are poor listeners (for voice), and low internal consistencies
(for voice and neglect). Taken together, the correlations be-
tween the measures of the four responses and the supervi-
sory ratings and archival data provide evidence of the
moderate construct validity of the measures.
Interviews
Semistructured interviews were conducted with 40 em-
ployees of the accounting firm at T1, 21 of whom were rein-
terviewed at T2. The T1 interviews averaged 45 minutes in
529/ASQ, December 1989
3
Because manytests were conducted,the
probability
of makingType 1 errorsis in-
flated.Accordingly,the Bonferonniproce-
durewas used throughout(Larzelereand
Mulaik,1977). Ingeneral,an unadjusted
alphalevel of .001 is equivalentto a Bon-
feronniadjustedalphalevel of .05.
Predicting EVLN

consistency of the voice measure no doubt contributed to this

weakness, as did conceptual inadequacies.
Loyalty. The model explained loyalty somewhat better. The
effects of exit costs were mixed, with skill specificity favoring
loyalty and investment discouraging it, while the high cost of
voice increased loyalty. Loyalty was also increased by low
prior satisfaction, believing that improvement was unlikely,
feeling externally controlled, and being uncommitted to the
organization. The results for prior satisfaction, believing in the
possibility of improvement, and organizational commitment
were opposite to predictions.
Neglect. Neglecters had mixed beliefs about the costs of
exit, with skill specificity favoring neglect and investment dis-
couraging it. They also had high voice costs, low prior satis-
faction, didn't believe improvement was likely, felt externally
controlled, were not committed to the organization, and had
a better alternative. The only difference between loyalists'and
neglecters was that neglecters had a better alternative and
loyalists were unaffected by alternatives.
Group analysis. To conclude the analysis of the data, we as-
sessed how well the independent variables of the model dif-
ferentiated among individuals who were unequivocal exiters,
voicers, loyalists, or neglecters. Individualswere classified as
heavy users of one response if they had a T2 standardized
score greater than or equal to one on the response and had
standardized scores of less than one on each of the other
three responses. This procedure identified 81 heavy users: 18
exiters, 30 voicers, 20 loyalists, and 13 neglecters.
Table 5 reports the results of one-way ANOVAs that were run
to test for differences among the four groups on each of the
T1 independent variables. Duncan's range test was applied to
determine which of the groups differed for the nine indepen-
dent variables when significant F-ratios were found.

Table5
F-ratios and Mean Scores on the PredictorVariables for Heavy Users of
Each of the Four Responses in Sample of Graduates
T2 Behavioral Responses*
Ti Independent Exit Voice Loyalty Neglect
variables (N = 18) (N = 30) (N = 20) (N = 13) F

Exitcosts
Skillspecificity 1.42 1.45 1.55 1.69 0.65
Sunkcosts 1 .76a 2.46b 2.24b 2.21ab 3.79@
Investment 4.14a 4.18a 3.75ab 3.46b 3.26-
Voice costs 2.93a 2.34b 3.20a 2.65ab 3.87
Priorsatisfaction 3,47a 4.15b 3,65a 3.74ab 4.62w
Possibilityof
improvement 2.80a 3.44b 2.96a 3.01ab 3.58-
Better alternatives 3.1 la 2.23b 2.55ab 2.81b 2.93-
Organizationalcommitment 2.97a 3.38b 3.55b 3.49b 5.980
Locus of control 13.73 15.64 14.50 12.08 2.42

p < .05; 6p < .01; Op < .001.

* Two means sharinga superscriptedletterare statisticallyequal.

Two of the exit costs-sunk costs and investment-had sig-

nificantlydifferent mean scores across the four groups. For
sunk costs, the exiter group had among the lowest scores,
531/ASQ, December 1989
while for investment, neglecters had among the lowest
scores. Sunk costs apparently reduced exit, while investment
did not, consistent with the difference between these two
exit costs that was noted in the previous analysis. Voice costs
also differed across the four groups, with among the lowest
scores found within the voice group, as expected.
Voicers had among the highest scores for prior satisfaction,
while exiters and loyalists had among the lowest scores.
Similarly, voicers were among those who saw the greatest
likelihood of improvement, and exiters and loyalists were
among those who saw the least chance of improvement. Ap-
parently, high prior satisfaction and the belief that improve-
ment is possible promote voice, while exit and loyalty are
most likely in their absence.
Two other variables served to differentiate the groups. Exiters
were among those most likely and voicers were among those
least likely to believe that better alternatives were available.
Finally, exiters were the least organizationally committed.
Overall, these results are consistent with the earlier analysis
of the full sample of graduates. In addition, they indicate that
the four behavioral responses are empirically separable and
that the model distinguishes among them.
Many of the predictions of the model held for the sample of
graduates, but the model was not a powerful predictor. Exit
was the most strongly predicted, while voice proved particu-
larly elusive. Loyalists and neglecters were predicted with in-
termediate strength, and, with one exception, by the same
independent variables. Differences in the model's ability to
predict the four responses may be due to differences in the
care with which each was measured, differences in how
much the behaviors varied in the samples, and the inclusion
of independent variables that may be more relevant to some
of the behaviors than to others (cf. Cooper and Richardson,
1986).
Accounting-Firm Sample
Analysis of the data from the accounting firm allowed us to
assess the replicability of the results of the survey of grad-
uates, albeit with a smaller sample. Table 6 presents a sum-
mary of results for the accounting-firm sample. The sample
size was reduced to 124 when we omitted the ten cases of
turnover.
Exit was decreased by one of the costs of exit (low sunk
costs) and increased by the high costs of voice, low job satis-
faction, not believing that improvement is likely, low organiza-
tional commitment, and having a better alternative. With the
exception of skill specificity and investment, these results
provide a replication of the sample of graduates. The model
accounted for 23 percent of the variance in exit. As in the
sample of graduates, the analysis of actual turnover (that is,
comparison of the 10 leavers with the 124 stayers) produced
relationships with the independent variables that were similar
(and, again, weaker) to those presented in Table 6.
The results for voice explained no variance. Only the positive
relationship with job satisfaction was replicated. None of the
variables satisfied the criteria for entry into the regression
equation
532/ASQ, December 1989
Predicting EVLN

Table 6

Zero-order Correlation Coefficients and Significant Betas in Regression

Analyses for Accounting-Firm Sample

T2 Dependent variables
T1 Independent variables Exit Voice Loyalty Neglect

Exit costs
Skill specificity .03 - .05 .10 .17-
Sunk costs -.20- .09 .10 -.08
Investment -.01 .15 - .15 - .26w
Voice costs .27" - .04 .15 .15
Priorsatisfaction -.47 .21- -.22- -.38
Possibility of improvement - .29" .13 - .12 - .03
Locus of control -.13 .00 - .1 8- -.30w
Organizational commitment -.45* .04 -.08 -.26"
Better alternatives .29" -.01 -.04 .05
Adjusted R2 .23 .00 .05 .28

*p < .05; Up < .01; Up < .001.

* Variables with betas that were significant at a < .05.

Two of the findings for loyalty from the sample of graduates

were replicated in the accounting-firm sample. Higher levels
of loyalty were found among those with low satisfaction and
an external locus of control. The model accounted for 5 per-
cent of the variance in loyalty.
Finally, neglect was more likely for individuals who had little
invested in the job but high skill specificity, low prior satisfac-
tion and low organizational commitment, and an external
locus of control. The remaining three findings from the
sample of graduates failed to replicate. Twenty-eight percent
of the variance in neglect was accounted for.
Several of the findings were thus replicated across samples,
while others failed to replicate. The smaller sample size in the
accounting firm may explain some of the failure to replicate.
For example, investment was related to loyalty in both
samples at about the same level (r = - .15 and - .14), but
the relationship was only significant in the larger sample. Dif-
ferences in the demographics and work settings of the two
samples may also have contributed to the failures to replicate.
Nevertheless, the lack of consistency across samples sug-
gests that some of the findings from the mailed survey of
graduates are unreliable.
DISCUSSION AND CONCLUSIONS
As with all work, this study has its limitations. We relied on
self-report data, thereby increasing the likelihood that the re-
sults are artifacts of respondents' implicit theories about how
the model operates. While the use of a longitudinal design
reduces the effects of these artifacts, it does not eliminate
the problem. Despite these misgivings, the results are en-
couraging.
Exit
Exit was the most consistently predicted of the four re-
sponses, as was true in Rusbult et al. (1988). People thought
more about leaving, began searches, took steps to leave, or
actually left when they had low costs of leaving but thought
voicingwas too costly, were dissatisfied and lackedhope that
the currentsituationwould improve,didn'tcare about the or-
5331ASQ,December 1989
ganization, and had attractive alternatives awaiting them. Exit
thus appears to be the result of two sets of forces: push and
pull (March and Simon, 1958). People leave more readily
when they are pulled out of the organization by better alter-
natives. They also leave when they are pushed out by dissat-
isfaction, low commitment, high voice costs, and the belief
that improvement is unlikely. This provides a coherent portrait
of the exiter that closely resembles our predictions and is
consistent with the literature on turnover (e.g., Mowday,
Porter, and Steers, 1982).
Among the predictors, two that merit attention are the costs
of exit and the availabilityof alternatives. The three costs of
exit had differential effects on the EVLNresponses. Invest-
ment seemed to operate as a psychological cost of exit,
whereas skill specificity and sunk costs were primarilyeco-
nomic costs. The economic costs of exit reduced exit across
samples and increased loyalty and neglect in the larger
sample. Psychological costs also reduced exit in the larger
sample but decreased loyalty and neglect. From these
findings it appears that economic costs entrap people in their
current setting while psychological costs involve people in
their current setting. These differential effects suggest that it
is worth distinguishing between the economic and psycho-
logical costs of exit and separately testing the distinction in
future work on the model.
The other variable of interest was alternatives. Having a
better alternative increased exit and neglect but had no effect
on the other responses. In partial contrast, Rusbult et al.
(1988) found that better alternatives increased exit and voice
and had no effect on neglect. Because these results are in-
consistent, future research is needed to provide a more fine-
grained assessment of alternatives and their effects. A recent
longitudinal study of the stability and dissolution of romances
by Simpson (1987) provides a model for such an expansion.
He assessed alternative romantic partners by asking respon-
dents to evaluate their best alternative and the best imagined
alternative partner on eleven dimensions, as well as the ease
of finding an alternative partner (as in our study). Simpson
found that romances tended to be more stable when the in-
dividual had a less desirable actual or imagined alternative
partner as well as when individuals thought it was hard to find
a suitable alternative. We think a more extensive assessment
of alternative jobs would similarly help in better understanding
what prompts choices among the EVLNresponses.

Voice
The results for voice were the weakest of the study. The
findings from the survey mailed to graduates were not strong,
and only one of them replicated. We offer two explanations
for these weak results, one conceptual and one psychomet-
ric. A measurement problem that we had, and that Rusbult et
al. (1988) also had, is the low internal consistency of the
measure of voice. One explanation for this is that voice is a
complex category that may have several subcomponents. A
recent act-frequency study of voice by Kay (1989) suggests
that this may be true. Using the act-frequency method devel-
oped by Buss and Craik(1983), Kay asked graduates with
more than five years work experience to identifyacts that
534/ASQ, December 1989
Predicting EVLN

they thought exemplified exit, voice, loyalty, and neglect,

which, following Buss and Craik (1983), were left undefined
so as to pull in a wide range of acts. A second set of grad-
uates with more than five years work experience were then
asked to rate the prototypicality of 100 acts in one of these
categories, after which a third, experienced sample sorted the
most prototypical acts into the appropriate categories. Kay
found that among the most prototypical acts of voice-which
was defined in the prototypicality rating stage as "taking ac-
tion to improve the situation, including working with others to
make things better"-were "encourage discussion of issues
and problems," "propose new ways of doing things," "take
actions on problems," and "make suggestions on how to im-
prove things." These acts suggest that this category is com-
plex, with acts differing on whether voicers take action or talk
about action and whether they work individuallyor collec-
tively. If voice is a more complex category than imagined in
existing research, then its internal consistency and predict-
ability will suffer, which is what we found.
In addition to this measurement problem, voice has a prop-
erty not shared by the other behaviors. Specifically, the effec-
tiveness of some forms of voice is dependent on someone
else responding to the behavior. Making suggestions and
acting as a spokesperson implies that someone is expected
to respond. Two implications of this are that people may be
reluctant to voice dissatisfaction if they don't have someone
to protect them from the consequences of voice (Near and
Miceli, 1986) and, since dependencies on others render voice
less controllable by the individual, the effectiveness of voice
may be inherently less predictable than the other behaviors.
This latter property of voice showed up in the interviews.
Several employees of the accounting firm described them-
selves as quite unhappy with their jobs, yet they didn't voice
their dissatisfaction because they didn't think any good would
come of it. Supervisors were perceived to be too hard to find,
too busy to listen, and not particularlygood at helping with
work problems when they were available. Thus conditions
that should produce a voice response didn't. These interppr-
sonal barriers to the use of voice were not captured by the
voice-cost construct. These barriers should be assessed in
future research.
There is a final (and somewhat ironic) sense in which voice
may depend on others. Some individuals think somebody
should do something but are not willing to do it themselves.
These people may think that exit costs are high and voice
costs are low, and so forth, but may do nothing because they
think someone else will do it. A recent study by Oliver (1984)
showed that community activists were more pessimistic than
were token activists that their neighbors would be willing to
contribute actively to solving community problems. An impli-
cation of Oliver's work is that voicers see a problem and be-
lieve that if they don't work on it no one else will. This work
on community activists suggests that future predictive work
on the EVLNmodel should assess employees' beliefs about
their coworkers' willingness to solve work problems. The
prediction is that voicers should be optimistic that improve-
ment is possible but pessimistic that others will take the ini-
tiative.
535/ASQ, December 1989
There are thus both measurement and conceptual problems
to be solved for voice. Our suggestion is to think of voice as
any activity that individuals direct toward improving the situa-
tion at work and to use the most prototypical acts identified
by Kay (1989) to operationalize the construct. Future work
should also include measures of the extent to which indi-
viduals are protected by sponsors, individuals' beliefs about
whether their coworkers are going to do anything to solve
work problems, and interpersonal barriersto the use of voice.
These factors need to be assessed if we are to understand
and predict voice.
Loyalty and Neglect
We began this study thinking that loyalty is both a behavior
and an attitude and accordingly operationalized it in both
ways. In retrospect we think we erred by including the attitu-
dinal and passive behavioral aspects of loyalty while omitting
the more active aspects of behavioral loyalty. The picture of
loyalty that emerged from the study did not closely resemble
the one with which we began. We thought of loyalty as a
way for employees to quietly support the organization
through hard times. But rather than quietly waiting for the or-
ganization to recover, loyal employees in our samples
seemed to be just biding their time, much like neglecters. In
fact, the variables that predicted loyalty were, by and large,
the same variables that predicted neglect.
The interviews provided a picture of the loyalist as an en-
trapped person. When employees referred to loyalty they did
not describe it as constructive support of the organization;
rather, they referred to bonds at the interpersonal level and
not to loyalty to the organization. Some resented the need to
adjust their expectations to fit what was possible in the orga-
nization and did not see why the workplace was so inflexible.
For many, the choice was viewed as adjusting or leaving.
Perhaps not surprisingly, some individuals described the
virtues of passively accepting unavoidable workplace charac-
teristics as their way of making peace with the workplace.
This sounds less like loyalty as support of the organization
than like resignation and entrapment.
Perhaps the choices for those who remain are two, and not
three, as the model indicates: people can actively work to
change the situation (voice) or they can become silent (loyalty
shading to neglect), using the apt phrase of Kolarskaand Al-
drich (1980). In this view, voice is the only option for people
who are concerned about the organization. An alternate view
is that constructive loyalty exists, but we failed to detect it
because of the way we asked people about it. The measures
we used asked about quiet, patient, waiting activity. In retro-
spect, it is not surprising that this passive activity was in-
creased by low psychological investments, low prior
satisfaction, pessimism about improvement, and low organi-
zational commitment. It is not surprising, because Farrell's
(1983) measure of loyalty identifies people who wait patiently,
quietly do their job, and say nothing to others.
In contrast, the acts that emerged from Kay's act-frequency
study of loyalty (Kay, 1989; Cooper et al., 1989)-defined as
"supporting the organization, including both active and pas-
sive support"-were very differentfrom Farrell's(1983)
536/ASQ, December 1989
Predicting EVLN

items, which are based on a passive, constructive model of

the loyalty construct. Among the most prototypical acts in
Kay's study were "give something extra when the organiza-
tion needs it," "do things above and beyond the call of duty
without being asked," "work hard to get the job done," and
"actively promote the organization's business." Among the
least prototypical acts were "perform illegal acts," "wait pa-
tiently and hope any problems will solve themselves," and
"say nothing to others and assume things will work out," the
latter two being items from Farrell's (1983) scale.
These results suggest something is seriously amiss with the
concept of loyalty. Lay descriptions of the most prototypical
acts portray loyalty as active support of the organization. It is
hard to imagine that people who do these things would be
characterized as having low psychological investment in the
organization, low satisfaction, being pessimistic about im-
provement, and not being committed to the organization. We
conclude that we got the loyalty results we deserved but,
that the loyalty we found does not fit with what lay people
mean by loyalty. Future research should operationalize loyalty
as active support for the organization, using the most proto-
typical acts to measure the construct. We would not expect
to see these loyalists look like entrapped individuals.

EVLN
We have treated the use of exit, voice, loyalty, and neglect as
independent responses to dissatisfaction at work, but they
may be related. Some possible sequences of the behaviors
can be detected by examining behavioral consistency over
time.
People behaved fairly consistently over the six-month period
examined in this study. The within-behavior correlations from
T1 to T2 in the sample of graduates were .61, .47, .56 and
.57 for the measures of exit, voice, loyalty, and neglect, re-
spectively. Thus people who were doing one thing at T1
tended to be doing much the same thing at T2. Nevertheless,
not everyone remained fixed at one response.
When people changed their responses they had a number of
possibilities from which to choose. Two possible sequences
were suggested by the interviews. Both are highly rational
accounts of what people say they will do if they are dissatis-
fied. The first sequence starts with voicing dissatisfaction to
the immediate supervisor. If voice doesn't solve the problem,
people may take their dissatisfaction to a higher level, and if
that is unsuccessful, they either become silent and put up
with it, or they quit. The availabilityof another job was viewed
as a key factor in the choice of which to choose if voice failed.
If people couldn't find a better job, they would most likely re-
evaluate the problem, adjust their beliefs, and stay.
The second sequence begins with loyalty. If nothing changes,
and enough time passes, the next response is voice. If voice
is unsuccessful, the employee will then resort to exit or ne-
glect, the choice among these depending largely on the avail-
ability of alternatives. Interviewees saw the magnitude of the
problem as the main factor in determining whether to start
with loyalty or voice. The bigger the problem, the noisier the
response.
537/ASQ, December 1989
 To assess the likelihood of these and other sequences in fu-
ture research, it would.be necessary to have observations
over a longer time than the six months used in the present
study. We see the question of sequencing as one of the next
steps in understanding the process of behavioral responses
to dissatisfaction. Such research should be sensitive to
threshold effects and abrupt changes in behavior (Guastello,
1987). The purpose of such work would be to learn more
about individuals' work lives through time.
The present study also showed that the responses were re-
lated, particularlythe exit and neglect responses. Some pre-
vious research has suggested that two forms of what we
regard as neglect (lateness and absenteeism) and exit (de-
fined as permanent departure from the organization) are
aspects of the same withdrawal phenomenon (e.g., Beehr
and Gupta, 1978). The results of the present study provide
some support for this view. There was a positive correlation
between neglect at T1 and actual turnover at T2 (ra= .22, p <
.001). People who had left by T2 reported more neglect at T1
than did stayers (means = 1.71 vs. 1.33; t = 3.86, p < .001).
Thus neglect at T1 could be seen as a precursor of exit by T2.
The relationship disappeared when neglect at T2 was corre-
lated with actual turnover at T2 (r = .06, n.s.). That is, people
who left were no more neglectful at their new job than was
the rest of the sample. Three explanations are offered for the
decrease in neglect after turnover has occurred: (1) neglect
and turnover may be consecutive phases in the withdrawal
process, with a new job suppressing neglect that may occur
again later if dissatisfaction sets in or the novelty of the new
job wears off; (2) neglect may be a substitute for exit when
exit is blocked; or (3) employees who have stayed, but plan to
leave, may be neglectful while they search for another job or
do other work (cf. Haccoun and Dupont, 1987). Each of these
fits some of the data we have on exit and neglect. We offer
them as lines for follow-up.
These analyses suggest that the behaviors were both related
to themselves and related to each other over time. Individuals
tended to stick with the same response over the six-month
period we observed, but some movement between re-
sponses did take place. Identifying what causes individuals to
shift among responses is an important focus for future re-
search. Rather than concentrating on one behavior (e.g., exit),
it will prove useful to examine a wider range of behaviors that
are available to individuals who are unhappy with their work.
The model provides one promising framework for studying
the choices made by individuals who face work situations that
they don't like.
REFERENCES
Barry,Brian Beehr, TerryA., and Nina Gupta Brayfield,ArthurH., and HaroldF.
1974 "Exit,voice and loyalty." 1978 "A note on the structureof Rothe
BritishJournalof Political employee withdrawal."Orga- 1951 "Anindexof job satisfaction."
Science, 4: 79-107. nizationalBehaviorand Human Journalof AppliedPsychology,
Becker, Howard S. Performance,21: 73-79. 35: 307-311.
1960 "Notes on the concept of Birch,A. H. Buss, David M., and Kenneth H.
commitment."American 1975 "Economicmodels in political Craik
Journalof Sociology,66: science: The case of 'exit, 1983 "Theact of frequencyap-
32-40. voice and loyalty'."British proachto personality."Psy-
Journalof PoliticalScience, 5: chologicalReview, 90:
69-82. 105-126.
538/ASQ, December 1989
 Predicting EVLN

Cooper, William H., Lorraine Dyke, Kay, Patricia Rotter, Julian B.

Patricia Kay, and William Graham 1989 "Anact frequencystudy of 1966 "Generalized expectancies for
1989 "The act frequency approach exit, voice, loyaltyand ne- internal versus external control
to organizational behavior." glect." Unpublishedhonours of reinforcement." Psycholog-
Unpublished manuscript, thesis, Departmentof Psy- ical Monographs, 80, whole
School of Business, Queen's chology, Queen's University, No. 609.
University, Kingston, Ontario, Kingston,Ontario,Canada. Rusbult, Caryl E., Dan Farrell, Glen
Canada. Kolarska,Lena, and Howard Rogers, and Arch G. Mainous Ill
Cooper, William H., and Alan J. Aldrich 1988 "Impact of exchange variables
Richardson 1980 "Exit,voice and silence: Con- on exit, voice, loyalty, and ne-
1986 "Unfair comparisons." Journal sumers' and managers're- glect: An integrative model of
of Applied Psychology, 71: sponses to organizational responses to declining job sat-
179-184. decline."Organization Studies, isfaction." Academy of Man-
1: 41-58. agement Journal, 31:
Farrell, Dan 599-627.
1983 "Exit, voice, loyalty, and ne- Larzelere,Robert E., and Stanley
glect as responses to job A. Mulaik Rusbult, Caryl E., and Isabella M.
dissatisfaction: A multi- 1977 "Single-sampletests for many Zembrodt
dimensional scaling study." correlations."Psychological 1983 "Responses to dissatisfaction
Academy of Management Bulletin,84: 557-569. in romantic involvements: A
Journal, 26: 596-607. multi-dimensional scaling anal-
March,James G., and HerbertA. ysis." Joyrnal of Experimental
Farrell, Dan, and Caryl E. Rusbult Simon Social Psychology, 19:
1981 "Exchange variables as pre- 1958 Organizations.New York: 274-293.
dictors of job satisfaction, job Wiley.
commitment, and turnover: Rusbult, Caryl E., Isabella M.
Mowday, RichardT., LymanW.
The impact of rewards, costs, Zembrodt, and Lawanna K. Gunn
Porter,and RichardM. Steers
alternatives, and investments." 1982 "Exit, voice, loyalty, and
1982 Employee-Organization
Organizational Behavior and neglect: Responses to dissat-
Linkages.New York:Academic
Human Performance, 28: isfaction in romantic involve-
Press.
79-95. ments." Journal of Personality
Near, Janet P., and MarciaP. Miceli and Social Psychology, 43:
Freeman, Richard B., and James L. 1986 "Retaliationagainstwhistle 1230-1242.
Medoff blowers: Predictorsand ef-
1984 What Do Unions Do? New Simpson, Jeffry A.
fects." Journalof AppliedPsy- 1987 "The dissolution of romantic
York: Basic Books. chology, 71: 137-145. relationships: Factors involved
Graen, George B., Robert C. Liden, Nunnally, Jum in relationship and emotional
and William Hoel 1978 PsychometricTheory.New distress." Journal of Person-
1982 "Role of leadership in the em- York:McGraw-Hill. ality and Social Psychology,
ployee withdrawal process." 53: 683-692.
Journal of Applied Psychology, Oliver, Pamela
67: 868-872. 1984 " 'Ifyou don't do it, nobody Spencer, Daniel G.
else will':Active and token 1986 "Employee voice and em-
Guastello, Stephen J. contributorsto localcollective ployee retention." Academy of
1987 "A butterfly catastrophe model action."AmericanSociological Management Journal, 29:
of motivation in organizations: Review,49: 601-610. 488-502.
Academic performance."
Journal of Applied Psychology, Porter, LymanW., RichardM. Withey, Michael J.
72: 165-182. Steers, RichardT. Mowday, and 1986 "Exit, voice, loyalty and ne-
Paul V. Boulian glect: Employee responses to
Haccoun, Robert R., and Serge 1974 "Organizational commitment, declining satisfaction." Unpub-
Dupont job satisfaction,and turnover lished doctoral dissertation,
1987 "Absence research: A critique among psychiatrictechni- School of Business, Queen's
of previous approaches and an cians."Journalof AppliedPsy- University, Kingston, Ontario,
example for a new direction." chology, 59: 603-609. Canada.
Canadian Journal of Adminis-
trative Sciences, 4: 143-156. Price, James L., and Allen C.
Bluedorn
Hirschman, Albert 0. 1979 "Test of a causal model of or-
1970 Exit, Voice, and Loyalty: Re- ganizationalturnover."In David
sponses to Decline in Firms, Dunkerleyand Graeme
Organizations, and States. Salaman(eds.), International
Cambridge, MA: HarvardUni- Yearbookof Organizational
versity Press. Studies: 217-236. London:
Routledge& KeganPaul.

539/ASQ, December 1989