HOW IMPORTANT ARE JOB ATTITUDES? META-ANALYTIC COMPARISONS OF INTEGRATIVE BEHAVIORAL OUTCOMES AND TIME SEQUENCES

Page created by Troy Neal
 
CONTINUE READING
娀 Academy of Management Journal
2006, Vol. 49, No. 2, 305–325.

            HOW IMPORTANT ARE JOB ATTITUDES? META-ANALYTIC
            COMPARISONS OF INTEGRATIVE BEHAVIORAL OUTCOMES
                          AND TIME SEQUENCES
                                                   DAVID A. HARRISON
                                                Pennsylvania State University

                                                    DANIEL A. NEWMAN
                                                    University of Maryland

                                                       PHILIP L. ROTH
                                                      Clemson University

                      Drawing on the compatibility principle in attitude theory, we propose that overall job
                      attitude (job satisfaction and organizational commitment) provides increasingly pow-
                      erful prediction of more integrative behavioral criteria (focal performance, contextual
                      performance, lateness, absence, and turnover combined). The principle was sustained
                      by a combination of meta-analysis and structural equations showing better fit of
                      unified versus diversified models of meta-analytic correlations between those criteria.
                      Overall job attitude strongly predicted a higher-order behavioral construct, defined as
                      desirable contributions made to one’s work role (r ⴝ .59). Time-lagged data also
                      supported this unified, attitude-engagement model.

   Job attitudes and job performance are perhaps the                 than they were to in-role performance (e.g., Herz-
two most central and enduring sets of constructs in                  berg, Mausner, Peterson, & Capwell, 1957; Vroom,
individual-level organizational research. Yet, a                     1964). Subsequent quantitative reviews also failed
longstanding debate persists about the nature and                    to show job attitudes as having strong predictive
the strength of relationships between these funda-                   utility. One meta-analysis reported a lackluster
mental predictors and criteria (Austin & Villanova,                  value (␳ˆ ⫽ .17) as the best estimate of the correla-
1992; Brief, 1998; Johns, 1998; Judge, Thoreson,                     tion between satisfaction and performance (Iaffal-
Bono, & Patton, 2001). An elemental question re-                     dano & Muchinsky, 1985). Another review showed
mains: How important are job attitudes for predict-                  organizational commitment bore a weaker relation-
ing and understanding job performance in particu-                    ship to job performance (␳ˆ ⫽ .14) than to at least
lar, and work role– directed behaviors in general?                   one withdrawal behavior, turnover (␳ˆ ⫽ ⫺.28;
   Authors of early qualitative reviews concluded                    Mathieu & Zajac, 1990). Consequently, the pendu-
that only weak support existed for the relationship                  lum of causal potency has swung away from job
between one principal attitude, job satisfaction,                    attitudes (at least until recently; see Judge et al.
and supervisor ratings or output measures of job                     [2001]). One widely held view is that attitudes are
performance (e.g., Brayfield & Crockett, 1955). A                    inconsistent or epiphenomenal forces in work be-
common inference in those reviews was that job
                                                                     havior (e.g., Locke & Latham, 1990): they explain
attitudes were more strongly related to absence,
                                                                     only 3–4 percent of performance variance and have
turnover, and other forms of work role withdrawal
                                                                     little practical importance for managers.
                                                                        The current article subjects that view to empiri-
                                                                     cal scrutiny via comprehensive and comparative
  The first two authors contributed equally to the writ-             tests. In doing so, we attempt to contribute to man-
ing of this article. We would like to acknowledge James              agement knowledge in five ways. First, we investi-
M. Conway for his help locating studies for the meta-                gate and more fully map the individual-level crite-
analysis of contextual and focal performance. Coding                 rion space (i.e., a set of work behaviors valued by
assistance was provided by Amanda Yancey, Ilana Goro-
                                                                     organizations [Austin & Villanova, 1992]) by bring-
dinskya, Maya Imberman, Rebecca Pedooem (all under a
work-study grant from Alliant International University),             ing four original meta-analyses to the literature,
and Crystal Wiggins. We are also grateful to Kevin Moss-             estimating the connections between contextual per-
holder, three anonymous reviewers, and Tom Lee for                   formance and (1) lateness, (2) absence, (3) turnover,
helpful and patient comments on drafts.                              and (4) focal (in-role) performance. Second, we cre-
                                                               305
306                                         Academy of Management Journal                                      April

ate a multivariate matrix of meta-analytic correla-              In addition to the evidence for a shared concep-
tions between pairs of these five behavioral criteria         tual domain, there is evidence of these constructs
and the two most commonly studied job attitudes:              having a great deal of shared variance. Mathieu and
job satisfaction and organizational commitment.               Zajac (1990) showed that measures of commitment
This effort involves combining our new meta-anal-             from the Organizational Commitment Question-
yses with clarified results of 17 existing meta-anal-         naire (OCQ) were more strongly connected to over-
yses, a process resulting in 21 estimates of bivariate        all satisfaction than to facet-specific (pay, co-
relationships in adult working populations. Third,            worker, supervision, etc.) satisfaction. Satisfaction
we use the ensuing meta-analytic matrix to com-               and affective commitment measures have a strong
pare the fit of competing theoretical models that             correlation (e.g., meta-analytic ␳ˆ ⫽.65 [Meyer, Stan-
specify relationships between attitudinal predic-             ley, Herscovitch, & Topolnytsky, 2002]). In fact, the
tors and (different structures among) behavioral cri-         correlation between overall job satisfaction and af-
teria. Fourth, we assess the time sequencing among            fective commitment is stronger than many of the
job attitudes and behaviors, comparing predictive             relationships between indicators typically taken as
with “postdictive” time-lagged designs. Fifth and             representing a single underlying construct. More-
most importantly, we attempt to answer the ques-              over, the correlation between affective commitment
tion posed in our title by estimating links between           and job satisfaction is stronger than the correlations
predictors and criteria defined at increasingly com-          between pairs of (affective, normative, and contin-
patible levels of generality, an effort culminating in        uance) facets of Meyer and Allen’s (1991) commit-
a broad attitude-engagement model that connects               ment construct (Meyer et al., 2002), and stronger
overall job attitude with overall individual                  than relationships between indicators of other gen-
effectiveness.                                                eral constructs (see du Toit & du Toit, 2001). Thus,
                                                              it is reasonable to treat job satisfaction and attitu-
                                                              dinal commitment as specific reflections of a gen-
        PREDICTORS: JOB SATISFACTION,
                                                              eral attitude, as each is a fundamental evaluation of
      ORGANIZATIONAL COMMITMENT, AND
                                                              one’s job experiences. Hence, we extend the work
            OVERALL JOB ATTITUDE
                                                              of Judge and coauthors (2001) and argue we can
   Job satisfaction, although defined in many ways,           conceptualize both job satisfaction and organiza-
has often been thought of as an emotional state               tional commitment as indicating an underlying
resulting from the evaluation or appraisal of one’s           overall job attitude.
job experiences (Locke, 1976), or as a psychological
state simultaneously represented by cognitive and
                                                                 CRITERIA: FOCAL VERSUS CONTEXTUAL
affective indicators (Brief & Weiss, 2002; cf. Schlei-
                                                                            PERFORMANCE
cher, Watt, & Greguras, 2004). The consensual por-
tion of organizational commitment’s definition is                Another major issue in a comprehensive test of
that it is a feeling of sharing beliefs and values with       attitude-behavior relationships at work is the
one’s entire organization—itself a positive emo-              breadth of the criterion space. For the past two
tional state (e.g., Meyer & Allen, 1991). That is,            decades, scholars have systematically expanded in-
despite conceptual and empirical distinctions (e.g.,          dividual-level behavioral criteria, responding in
Tett & Meyer, 1993), it is clear that job satisfaction        part to the early and fairly gloomy reviews of atti-
and organizational commitment have theoretical                tude-performance connections (e.g., Organ, 1977).
and empirical commonalities. Both satisfaction and            Organ and his colleagues have defined organiza-
commitment are nonspecific with regard to the ac-             tional citizenship behavior (OCB) using elements of
tions prescribed. In Meyer and Allen’s (1991) three-          work activity not fully captured by traditional (fo-
dimensional reconceptualization, affective com-               cal performance, task completion) concepts (Bate-
mitment is the most strongly overlapping in                   man & Organ, 1983; Organ, 1997; Smith, Organ, &
constitutive and operational definition with atti-            Near, 1983). Borman and Motowidlo further ab-
tude. Indeed, recently it has been termed “attitudi-          stracted these behaviors (1993) into contextual per-
nal commitment” (Riketta, 2002). Hulin (1991) also            formance, a more inclusive criterion dimension.
noted considerable theoretical overlap between af-            Such behaviors were seen as more interpersonally
fective commitment and overall job satisfaction,              oriented (Motowidlo, 2000; Van Scotter & Motow-
remarking that the only clear difference between              idlo, 1996), more discretionary, and more “extra-
the two is their conceptual target. The target of job         role” (e.g., helping coworkers, encouraging or im-
satisfaction is one’s position or work role; the target       proving morale, and endorsing, supporting, and
of affective commitment is the entire organization            defending organizational objectives), than what has
(Hulin, 1991: 489).                                           been characterized as “in-role” performance (Or-
2006                                       Harrison, Newman, and Roth                                         307

gan, 1988; Organ & Paine, 1999). We suggest con-            means through which employees can withhold in-
textual performance is now an important part of             puts from an organization. Many foundational the-
what Fisher (1980: 607) called the “total set of            ories of organizational behavior, including equity
work-related behaviors,” and examining this con-            theory (Adams, 1965), inducements-contributions
struct fulfills recommendations to study broader,           theory (March & Simon, 1958), and social exchange
more abstract criteria (see Hanisch, Hulin, &               theory (Thibault & Kelly, 1959) suggest straightfor-
Roznowski, 1998; Judge et al., 2001).                       ward reasons why individuals contribute or with-
   Past studies have focused on attitudinal predic-         hold such inputs. Under their auspices, we theorize
tors of contextual performance (Organ & Ryan,               that lateness and absence are often controllable
1995). Research on links between contextual per-            forms of input reduction, subject to the same moti-
formance and other criterion dimensions (e.g., late-        vations for withholding inputs as OCBs, helping
ness, absenteeism, and turnover) is more recent. Of         behaviors, and other elements of contextual perfor-
equal importance, the position of contextual per-           mance (cf. Harrison, Johns, & Martocchio, 2000).
formance in the temporal progression of behavioral          Those who are willing to expend the (extra-role)
responses to negative attitudes has not been made           effort to engage in contextual performance are less
explicit. Below, we review and develop formal hy-           apt to reduce their (in-role) effort to meet the focal
potheses supporting such links. Those hypotheses            demands of their work schedules. Additionally, ab-
serve as conceptual bases of four new meta-analy-           senteeism and lateness permit an employee to re-
ses, which themselves are necessary for completing          duce the costs of an aversive job by engaging in
the meta-analytic matrix of pairwise correlations           more pleasurable activities while still maintaining
between all commonly studied behavioral criteria            the job’s economic benefits. There are also fewer
and job attitudes.                                          opportunities to enact forms of contextual perfor-
                                                            mance when one spends less time at work (is late or
                                                            absent). Thus,
Contextual Performance and Turnover
                                                              Hypothesis 2. Contextual performance is neg-
  Chen, Hui, and Sego (1998) proposed that avoid-             atively related to absenteeism.
ance of citizenship behavior may be a discretionary
and primary means for employees to reduce work                Hypothesis 3. Contextual performance is neg-
role inclusion. If the morale-building or relation-           atively related to lateness.
ship-enhancing actions comprising contextual per-
formance (Van Scotter & Motowidlo, 1996) are con-           Contextual Performance and Focal Performance
sidered “prepayment” for eventual good treatment
by an employer, then avoidance of contextual per-              The connection between contextual and focal
formance may signal employees’ intentions to                (task) performance has been given more research
“write off” these investments in a firm they plan to        attention than the connection between contextual
leave. Likewise, in their job embeddedness model,           performance and withdrawal behaviors (e.g., Con-
Mitchell, Holtom, Lee, Sablynski, and Erez (2001)           way, 1999; Motowidlo & Van Scotter, 1994; Ro-
proposed that a major factor inhibiting turnover is         tundo & Sackett, 2002). Task performance is typi-
the depth and breadth of interpersonal relation-            cally defined as the degree to which an individual
ships developed through contextual performance              meets or exceeds expectations about focal role re-
behaviors. Mossholder, Settoon, and Henagan                 quirements. Recently, Hunt (2002) argued that
(2005) also showed evidence that workers with               when employees have a fixed pool of inputs or
fewer interpersonal ties were more likely to quit.          efforts, a negative relationship should be expected
Hence, contextual performance promotes the for-             between contextual and focal performance. He re-
mal and informal connections that reduce an em-             fers to these situations as “Taylorist jobs,” in which
ployee’s likelihood of quitting.                            strict adherence to routinized procedures is
                                                            advocated.
  Hypothesis 1. Contextual performance is neg-                 Most jobs, however, have become less routinized,
  atively related to turnover.                              less unidimensional, and less strictly defined (Cas-
                                                            cio, 1998), reducing the asserted trade-off between
                                                            contextual and focal performance. Additionally, for
Contextual Performance, Absenteeism, and
                                                            a variety of circumstances, individual difference
Lateness
                                                            variables have been found to produce relatively
  In formulating ideas about links between contex-          high levels of both task performance and citizen-
tual performance, absenteeism, and lateness, we             ship behavior. These individual difference vari-
also note the role of absenteeism and lateness as           ables include conscientiousness, emotional stabil-
308                                       Academy of Management Journal                                       April

ity, and agreeableness (Hurtz & Donovan, 2000;                 In contrast, for the compensatory forms and al-
LePine, Erez, & Johnson, 2002). Further, Taylorist          ternative forms models, single withdrawal behav-
jobs are most likely to produce negative within-            iors are assumed to be substitutable in specific
person correlations of contextual with task perfor-         ways for one another. Rosse and Miller (1984) de-
mance, while our current focus is on between-per-           scribed them in terms of “water under pressure”;
son correlations. Given this pattern of evidence, we        the metaphoric flowing water is the urge to with-
expect that some individuals bring higher levels of         draw from a dissatisfying work environment (see
personal resources (time, energy, human capital) to         also Johns, 1997). Under the alternate forms model,
their jobs, fostering higher levels of focal and con-       external constraints on one behavior (the turnover
textual performance.                                        faucet is closed) mean that the urge will be expressed
  Hypothesis 4. Contextual performance is posi-             in another behavior (the absenteeism faucet is open).
  tively related to focal (task) performance.               Under compensatory forms, enacting one form of
                                                            withdrawal will have a tempering (relief valve) effect
                                                            on dissatisfaction, and therefore lessen the probabil-
      CRITERIA: WITHDRAWAL BEHAVIORS
                                                            ity of enacting another form of withdrawal. Both
   Alongside contextual and focal performance,              models are taken to predict negative within-person
withdrawal behavior is arguably a third major di-           covariance between individual withdrawal behaviors
mension of the individual-level criterion space. Ac-        over short periods of time (Martocchio & Harrison,
tions such as lateness, absenteeism, and turnover           1993). The spillover model connects withdrawal be-
have a long history of study in management, and             haviors in a positive way (Rosse & Miller, 1984).
direct bottom-line implications for firms. Although         Engaging in lateness, absence, or turnover is a reflec-
researchers have meta-analyzed connections be-              tion of a general, underlying propensity to withdraw,
tween pairs of withdrawal behaviors (e.g., Mitra,           which itself is determined by an overall, negative job
Jenkins, & Gupta, 1992), and between each of the            attitude. What differentiates the three behaviors is
major withdrawal behaviors and job attitudes (e.g.,         merely the threshold that the underlying attitudinal
Hackett, 1989), they have not been examined si-             propensity must breach to reveal itself in a particular
multaneously or as key components of a broader              action (lateness has the lowest threshold [Hulin,
criterion space.                                            1991]). This model would be operationalized with all
   Just as there are debates about the connections of       three behaviors serving as congeneric reflections of a
job attitudes with performance, there are decades-          single withdrawal construct. The compensatory, al-
old sets of opposing ideas about the nomological            ternative forms, and spillover models also mandate a
networks of single- and multiple-behavior forms of          separate dimension of withdrawal from task and con-
withdrawal (see the summary by Johns [1998]). Hu-           textual performance in our meta-analytic model-fit-
lin (1984, 1991) suggested that the meanings of             ting, as all three models specify a single underlying
lateness, absence, and turnover can be found in             urge to withdraw that is variously manifested through
their patterns of covariation. Rosse and Miller             lateness, absenteeism, and turnover.
(1984) identified five sets of those patterns, or no-          Finally, according to the progression of with-
mological networks, as underlying theories of rela-         drawal model, positive covariances occur between
tionships among withdrawal behaviors themselves,            pairs of withdrawal behaviors in a specific, cascad-
and between withdrawal behaviors and their pro-             ing order (Benson & Pond, 1987; Krausz, Ko-
posed antecedents and consequences (also see their          slowsky, & Eiser, 1998; Mobley, 1982; Rosse, 1988).
reinterpretation by Harrison and Martocchio                 Under the progression conceptualization, all three
[1998]). According to the independent forms model           withdrawal behaviors are presumed to be re-
of withdrawal, lateness, absenteeism, and turnover          sponses to negative job attitudes. But an additional
each have a unique etiology. In its extreme form,           requirement is that they be connected in a causal
this model is taken to predict near-zero covariances        chain, generating a simplex pattern of behavior-
among uniquely determined withdrawal behaviors              specific correlations from lateness to absence, and
(Rosse & Miller, 1984). However, a more precise             then absence to turnover.
characterization of the independent forms model
might be that it predicts differential connections of
job attitudes to each type of withdrawal behavior.           CONNECTING ATTITUDINAL PREDICTORS TO
Under an independent forms model of withdrawal,                      BEHAVIORAL CRITERIA
a model fitted to attitude-behavior correlations that
                                                            The Compatibility Principle
keeps lateness, absenteeism, and turnover distinct
(and therefore includes no underlying withdrawal               Over the past two decades, some researchers
construct) should fit best.                                 (e.g., Fisher, 1980; Hulin, 1991) have argued that
2006                                          Harrison, Newman, and Roth                                        309

the apparently meager connections between job at-              Judge and colleagues (2001) disregarded effects of
titudes and job performance (typically measured                job facet satisfaction and concluded that overall
only as focal performance) are, in fact, consistent            satisfaction had a much stronger meta-analytic re-
with fundamental principles of attitudes proposed              lationship with overall job performance than pre-
by social psychologists. These authors contend that            viously believed (␳ˆ ⫽ .30). Their conclusion was
job attitudes do not predict job behavior well be-             followed by a call for research on relationships of
cause behavioral criteria are defined and treated at           job attitudes to even broader behavioral criteria:
a different level of abstraction than attitudinal pre-         “Issues of construct generality and correspondence
dictors. These arguments (Fisher, 1980; Fisher &               have fundamental effects on the nature and magni-
Locke, 1992; Hulin, 1991; Roznowski & Hulin,                   tude of the relationships between attitudes and be-
1992) are based largely on the theories of Fishbein            haviors . . . but have rarely been considered in the
and Ajzen (e.g., Ajzen & Fishbein, 1977, 1980; Fish-           satisfaction-performance literature” (Judge et al.,
bein & Ajzen, 1974, 1975), who developed the com-              2001: 392). Responding to this suggestion and tak-
patibility principle for attitudes to account for their        ing their work a step further, we explicitly consid-
apparently inconsistent relationships to behavior              ered those issues in the present research. Using
(their original term was “congruence,” later re-               structural equation modeling of meta-analytic cor-
named “compatibility” by Ajzen [1988]). They re-               relations between pairs of job attitudes and behav-
sponded to extensive disputes in social psychology             ioral criteria (Viswesvaran & Ones, 1995), we tested
over the role of attitudes in determining and pre-             the fit of models of the consequences of overall job
dicting behaviors (e.g., Mischel, 1968; Wicker,                attitude. As those models move from theorizing
1969), proposing that an attitude impels behavior              more specific to more general behavioral criteria,
only when the two constructs are compatible in                 we expected they would show better fit to existing
their action, target, context, and time (see Epstein           data. In addition to assessing model fit, we exam-
[1980] for a parallel explanation with dispositional           ined the connection between overall job attitude
constructs). Ajzen and Fishbein (1977) theorized               and behavioral criteria. Following the compatibil-
and showed that attitude-behavior connections                  ity principle, we expected the attitude-behavior
were strongest when the attitude was matched in                connection to grow progressively stronger as be-
specificity or generality to behavior. More recently,          havioral criteria were defined in broader, more in-
Kraus (1995) also found support for the compatibil-            clusive ways. We termed the most general form of
ity principle in nonwork contexts, with meta-ana-              these criterion dimensions individual effective-
lytic correlations (r’s) of .29 versus .62 under low           ness, the tendency to contribute desirable inputs
and high compatibility, respectively.                          toward one’s work role.
   In applying the compatibility principle to the
present question, it should be noted that job satis-
                                                               Attitude-Behavior Relationships: Specific Models
faction and organizational commitment are atti-
tudes that connote a broad target, but not an action,            As we have noted, withdrawal, contextual per-
context, or time. According to attitude theory, such           formance, and focal (task) performance have not
attitudes should kindle a general, undifferentiated            been drawn together in a comprehensive empirical
force to engage in (positive or negative) behaviors            analysis involving attitudinal predictors. However,
that express or manifest the attitude. Such attitudes          several theoretical models of the structure of these
should therefore predict wide sets or aggregates of            behavioral criteria exist. Many comprehensive
behaviors directed toward one’s role that are not              models (Campbell, 1990; Murphy, 1990; Van Scot-
limited to a specific task dimension, social environ-          ter & Motowidlo, 1996) do not explicitly address
ment, or type of activity at work. This idea, also             the possibility that correlations among criterion di-
found in Fisher’s assertion about attitude-behavior            mensions could reflect a higher-order or more gen-
compatibility, that “general satisfaction measures             eral effectiveness construct. However, Viswesvaran
should be related to the favorableness or unfavor-             and Ones (2000) noted that meta-analytic evidence
ableness of an individual’s total set of work-related          indicated a positive manifold, or sharedness,
behaviors” (1980: 607), has not been fully tested. If          among the various conceptions of performance di-
this idea were correct, as researchers define behav-           mensions: task performance, contextual perfor-
ioral criteria at increasingly higher levels of ab-            mance, counterproductivity, and so on. They inter-
straction (e.g., actions contributing to aspects of            preted this overlap as a higher-order performance
one’s job or work role), the empirical connection              or “p-factor” (Arvey, 1986; Viswesvaran, 1993), a
between overall job attitudes and such criteria                substantively meaningful construct not simply due
should become stronger.                                        to idiosyncratic rater halo error (Viswesvaran &
   In keeping with the compatibility principle,                Ones, 2000: 223).
310                                        Academy of Management Journal                                       April

   In the present article, we organize these ideas           is considered, overall job attitude is predicted to
about criterion structure in ways that correspond            have unique effects on each criterion dimension
with increasing fidelity to the compatibility princi-        (see Figure 1, top panel). The independent forms
ple reviewed above. Specifically, we move from               model of job withdrawal (Rosse & Miller, 1984)
conceptualizations of the criterion space that might         would fall under this rubric, as it rests on different
be characterized as diversified, treating multiple           strengths of predictors for lateness, absence, and
responses to job attitudes as unique behaviors or            turnover (including a version that supposes one
sets, to those that might be characterized as unified,       behavior is a function of unfavorable job attitude
treating all behavioral dimensions as parts of an            while the others are not [Johns, 1998]).
overall effectiveness construct. The former mod-                Model B: Diversified criteria, plus progression
els—including those that mandate a distinct crite-           of withdrawal. Within the diversified criterion
rion dimension for each form of withdrawal behav-            model, relationships among single withdrawal be-
ior—treat elements of the criterion space as more            haviors can be structured to be consistent with one
behaviorally specific, and the latter treat them as          or more of the withdrawal theories reviewed ear-
more general. According to the compatibility prin-           lier. In particular, overall job attitude can relate in
ciple, the latter models should show stronger con-           a unique way to each behavior within the criterion
nections between overall job attitude and the                space, while the withdrawal behaviors inside that
(shared) variance in behavioral criteria. Differences        space are interrelated in a predefined way. In keep-
between diversified and unified theoretical models           ing with the progression of withdrawal hypothesis,
stem mainly from how they arrange elements of                we propose a model of a diversified criterion space
task performance, contextual performance, and                that overlays an ordered sequence among with-
withdrawal behaviors relative to one another. We             drawal behaviors, moving from lateness to absen-
describe those models below and show them in                 teeism to turnover (see the dotted arrows in Figure
Figure 1.                                                    1, top panel).
   Model A: Diversified criteria. Job performance               Model C: Diversified criteria, but unitary with-
has been defined as behaviors that are under indi-           drawal. As suggested in reviews of the conse-
vidual control and that affect the goals of the em-          quences of job satisfaction and organizational com-
ploying organization (Campbell, 1990). As we men-            mitment (Herzberg et al., 1957; Mathieu & Zajac,
tioned above, a diversified model of the criterion           1990; Vroom, 1964), we specified a model in which
space would specify no general, higher-order factor          task and contextual performance are distinguished
underlying the various dimensions of work behav-             from withdrawal as criteria (see Figure 1, middle
ior or performance. According to this model, sets of         panel). In terms of specific versus general ap-
actions such as lateness, absenteeism, turnover,             proaches to the criterion space, this intermediate,
and contextual performance are determined in dif-            three- rather than five-dimensional model, groups
ferent ways and in different strengths by job atti-          lateness, absenteeism, and turnover together as out-
tudes. The criteria do not share a single etiology.          croppings of an underlying withdrawal construct
Instead, performance-related behaviors reflect how           (see Hanisch & Hulin, 1991; Hanisch, Hulin, & Ros-
much individual control or discretion each one               nowski, 1998; Rosse & Hulin, 1985). This model is
entails. Contextual performance is associated with           also consistent with withdrawal as spillover (Rosse
the level of effort or persistence that an individual        & Miller, 1984).
exerts beyond what is required. Absenteeism and                 Model D: Unified criterion. Moving from three
lateness, as the reduction of effort, are somewhat           dimensions to a single, general effectiveness di-
less discretionary, with controls on their expres-           mension involves moving from a diversified to a
sion that vary across jobs (Johns, 1991). Focal per-         unified behavioral criterion. Such a model is con-
formance is the least discretionary. Typically in-           sistent with the hierarchical perspective of a
role or expected, it serves as the basis for the             p-factor in the criterion space (Viswesvaran and
distribution of formal organizational rewards (Bor-          Ones [2000]; see Figure 1, bottom panel). Hulin
man & Motowidlo, 1993). The discretionary com-               (1982) implied a similar structure decades earlier.
ponent of turnover depends upon the external labor           The unified model implies that overall job attitude
market and information about alternative job                 is associated with the shared or empirically over-
opportunities.                                               lapping portions of behavioral criteria at work. In
   Such a conceptualization also implies that over-          terms of attitude theory, this model represents the
all job attitude has a stronger connection to contex-        greatest attitude-behavior compatibility (Fishbein
tual performance, lateness, and absence, than to             & Ajzen, 1997). Both constructs are treated at the
focal performance and (perhaps) turnover (e.g.,              highest level of generality or abstraction. Overall
Chen et al., 1998). That is, when individual control         job attitude is generic with respect to actions, con-
2006                             Harrison, Newman, and Roth                         311

                                       FIGURE 1
       Models of Relationships between Individual Job Attitudes and Work Behavior
312                                        Academy of Management Journal                                     April

texts, and times. We contend that positive job atti-         ples of employed adults. This procedure was con-
tude creates a tendency to engage or contribute              sistent with previous meta-analyses represented in
desirable inputs to one’s work role, rather than             our attitude-behavior matrix (Judge et al., 2001).
withhold them. Each behavioral criterion is a re-            Primary studies sampled people working in natural
flection of this general tendency.                           settings and estimated individual-level effects. Ef-
   Model E: Unified criterion, plus progression of           fect sizes for relationships with turnover measures
withdrawal. Finally, it is possible that both the            were restricted to those articles and papers report-
compatibility principle and the progression of               ing actual separations from an organization, rather
withdrawal model operate simultaneously (Rosse,              than turnover intentions or withdrawal cognitions.
1988). In this specification, individual withdrawal             Identification of studies. Studies for the present
behaviors owe a major portion of their covariation           meta-analyses were located through electronic and
to the general effectiveness criterion. At the same          manual searches of databases, bibliographies from
time, there is a dependency structure between pairs          quantitative and qualitative reviews, and confer-
of withdrawal behaviors (see Figure 1, bottom                ence proceedings. Initially, we searched the
panel, dashed arrows).                                       PsycINFO, ERIC, and ABI/Inform databases from
                                                             1983 (the first year in which a paper on OCB was
                                                             published) through October 2004, using the subject
                     METHODS
                                                             terms “citizenship behavior,” “contextual perfor-
   To test the compatibility principle, and therefore        mance,” “prosocial behavior,” and “extra-role be-
estimate the importance of overall job attitude for          havior.” The focus of the literature review was on
predicting a higher-order job behavior construct,            published articles and chapters, unpublished doc-
we applied the models described above to a meta-             toral dissertations, conference papers, and cited but
analytic matrix of relationships among specific job          unpublished manuscripts. The abstracts from all
attitudes and behaviors that have frequently ap-             studies identified by the databases were searched
peared in past research. These behaviors included            manually for references to absenteeism, sick leave
focal performance (task or in-role performance, typ-         use, attendance, turnover, retention, quitting, late-
ically measured by supervisor ratings), contextual           ness, promptness, tardiness, performance, and in-
performance (typically measured as OCB), lateness,           role behavior. We also posted messages on listservs
absenteeism, and turnover. Although published                (e.g., RMNet, HRDivNet) asking for unpublished or
meta-analytic estimates were available for bivariate         forthcoming studies. Even with this broad search,
relationships between attitudes (job satisfaction            there were relatively few studies to cumulate that
and organizational commitment) and each specific             correlated contextual performance with each of the
criterion dimension, one of the contributions of our         three withdrawal behaviors. However, we were
study is to review and estimate meta-analytic rela-          able to obtain at least five independent samples for
tionships between contextual performance and other           each new estimate, a number that compares favor-
criteria. We derived meta-analytic correlations be-          ably to those in other recent meta-analyses (e.g.,
tween contextual performance and turnover (Hypoth-           Martocchio, Harrison, & Berkson, 2000). Table 1
esis 1), absenteeism (Hypothesis 2), lateness (Hypoth-       reports the number of studies (k) and total number
esis 3), and focal performance (Hypothesis 4).               of individuals (N) for each new meta-analytic esti-
   In many of the primary studies included in our            mate. Because one of the prior meta-analytic esti-
search, contextual and task performance ratings              mates (lateness with focal performance, from Ko-
were taken from the same source (e.g., supervisors).         slowski, Sagie, Krausz, and Singer [1997]) was
Therefore, to be commensurate with the other                 based on fewer than five original studies, we also
meta-analytic values that were not subject to bias           report an updated estimate for it in Table 1.
by common method variance or percept-percept                    Meta-analytic procedures. Following the
inflation (cf. Organ & Ryan, 1995), we separated             method used to derive the prior correlations in-
original studies on the basis of whether data for the        cluded in our eventual meta-analytic matrix of at-
two variables came from a common source. Non-                titude-behavior connections, we employed Hunter
common source estimates were used in our tests of            and Schmidt’s (2004) corrections for attenuation
competing models.                                            due to unreliability. Such corrections are often
                                                             viewed as conservative (Hunter & Schmidt, 2004).
                                                             Many of the primary studies did not report corre-
Meta-Analyses of Links between Contextual
                                                             lations with contextual performance, but only
Performance and Other Job Behaviors
                                                             facet-specific correlations (e.g., with the “sports-
  Rules for inclusion. We selected studies for con-          manship” facet of OCB). We combined correla-
textual performance meta-analyses that used sam-             tions from these dimensional measures to form a
2006                                          Harrison, Newman, and Roth                                            313

                                                  TABLE 1
                             Results of Meta-Analyses for Contextual Performance
                                                                                                  Percentage of Variance
                         Uncorrected            Total                               Corrected        Accounted for by
                             r            k      N        95% Confidence Interval       r             Sampling Error

Contextual performance
  Turnover                   ⫺.20         5     1,619          ⫺.12   to ⫺.36          ⫺.22                21%
  Lateness                   ⫺.11         5       578          ⫺.03   to ⫺.28          ⫺.15                70
  Absenteeism                ⫺.22         8       957          ⫺.16   to ⫺.27          ⫺.26               100
  Focal performance           .20        24     9,912           .16   to .24            .23                20

Lateness
  Focal performance          ⫺.20         7     1,879          ⫺.13 to ⫺.50            ⫺.26                12

unit-weighted composite correlation between all                correlation matrix can produce incorrect standard
dimensions of contextual performance (e.g.,                    errors when the standard deviation varies across
OCB) and the respective criterion.                             input variables (Cudeck, 1989). Fortunately, these
   In contrast to the other job behaviors, turnover            standard errors are often overestimated (Cudeck,
was treated as having a reliability of unity (Griffeth,        1989: 323), making significance tests of individual
Hom, & Gaertner, 2000). Interrater or internal con-            parameters conservative.
sistency estimates of reliability were often not re-
ported for lateness, absenteeism, and in-role perfor-
                                                               Other Adjustments to the Meta-Analytic
mance. We substituted the average reliability
                                                               Correlation Matrix
estimate from the other studies that did provide it,
which is a form of imputation.                                    One limitation we discovered in previous meta-
                                                               analyses involving job attitudes and behavioral crite-
                                                               ria was failure to appropriately specify the level of
Analytic Framework for Fitting and Comparing
                                                               analysis of primary studies involved in the calcula-
Diversified versus Unified Models
                                                               tion of the final meta-analytic estimate (Ostroff & Har-
   Data for estimating the fit of models A through E           rison, 1999). For example, one of the prior meta-
were obtained by bringing together the four new                analytic estimates in the literature was based on a
meta-analytic estimates described above with 17                total N of approximately 3,000, but 1,244 of the data
other meta-analyses estimating bivariate relation-             points were from Angle and Perry’s (1981) unit-level
ships between job attitudes and work behaviors. Use            rather than individual-level analysis. In several cases,
of meta-analytically derived matrices for structural           inclusion of this correlation created significant bias in
equation models was advised by Viswesvaran and                 the published estimate. Thus, we removed Angle and
Ones (1995) and Shadish (1996) and has been re-                Perry’s (1981) result from the meta-analytic estimates
ported many times in the human resources and or-               that included it, permitting our final model to reflect
ganizational behavior literatures (e.g., Bhaskar-              individual-level relationships. Additionally, for the
Shrinivas, Harrison, Shaffer, & Lau, 2005). A series of        correlation between organizational commitment and
meta-analyses published since 1989 provided esti-              job satisfaction, we combined the results of Mathieu
mates for the input correlation matrix (see the foot-          and Zajac (1990) with those of Meyer and colleagues
notes for Table 2). In this approach, the structural           (2002), as those meta-analyses were based on inde-
equation model uses manifest indicators without                pendent sets of original studies.
correction for measurement error, as these correc-
tions have already been accomplished through
                                                               Time Sequencing
meta-analysis.
   We acknowledge that comparisons between non-                  Our use of directional arrows in Figure 1 carries
nested models are somewhat arbitrary, and their                with it a set of implicit and often disregarded hy-
use prohibited us from pinpointing specific paths              potheses about the temporal ordering among vari-
that accounted for between-model differences in                ables (Mitchell & James, 2001). All of the models
fit. Further, these models are limited by our neces-           implicitly specify that job attitudes are temporally
sary reliance on correlation (rather than covari-              prior to behavioral criteria, and the progression of
ance) estimates as input for the modeling proce-               withdrawal models (models B and E) implies a
dure. Structural equation models based on a                    temporal sequence from lateness to absence to turn-
314                                                     Academy of Management Journal                                                           April

                                                       TABLE 2
                      Meta-Analytic Correlations between Job Attitudes and Job-Related Behaviorsa
           Construct                             1                   2                   3                  4                  5                  6

1. Job satisfaction

2. Organizational commitment                    .60 b
  k studies                                     112
  N total observations                       39,187

3. Focal performance                            .30 c              .18 d
  k studies                                     312                 87
  N total observations                       54,471             20,973

4. Contextual performance                       .28 e              .25 f                .23 g
  k studies                                      32                 42                   24
  N total observations                       16,348             10,747                9,912

5. Turnover                                    ⫺.19 h             ⫺.22 i               ⫺.15 j             ⫺.22 k
  k studies                                      67                 66                   72                  5
  N total observations                       24,566             26,296               25,234              1,619

6. Lateness                                    ⫺.11 l              ⫺.15 m              ⫺.26 n             ⫺.15 o              .09 p
  k studies                                      15                   7                   7                  5                  5
  N total observations                        3,767               1,896               1,879                578              1,310

7. Absenteeism                                 ⫺.17 q              ⫺.16 r              ⫺.29 s             ⫺.26 t              .30 u             .38 v
  k studies                                      25                  30                  49                  8                 33                24
  N total observations                        4,741               5,748              15,764                957              5,316             6,769

   a
     All correlations are disattenuated for unreliability. Turnover correlations are not corrected for base rate. If more than one meta-analysis
reported on the same relationship, we used the estimate reflecting the greatest amount of data (in all cases, this was the most recent
estimate). When more than one set of estimates was given, correlations with absence frequency were used.
   The letter superscripts in the body of the table indicate the sources of the meta-analytic correlations as follows: “s,” Bycio (1992); “h,”
“i,” “j,” Griffeth et al. (2000); “q,” Hackett (1989); “c,” Judge et al.; “l,” “m,” “v,” “p,” Koslowsky et al. (1997); “e,” “f,” LePine et al. (2002);
“b,” “r,” Mathieu and Zajac (1990), composited with Meyer et al. (2002); “u,” Mitra et al. (1992); and “d,” Riketta (2002).
   Original analyses include “g,” “k,” “n,” “o,” and “t.” A correlation from Angle and Perry (1981) was removed from estimates “i,” “m,”
“p,” and “v.”

over. Our review of relationships between contex-                               expectations, contextual performance was nega-
tual performance and withdrawal behavior (Hy-                                   tively related to all three withdrawal behaviors
potheses 1–3) further implies that citizenship                                  and positively related to focal performance. Cor-
behaviors are withheld prior to the decision to miss                            rected (for unreliability) estimates were moderate
work (Chen et al., 1998).                                                       in size for relationships with turnover (␳ˆ ⫽ ⫺.22),
   Because the published meta-analyses on which                                 lateness (␳ˆ ⫽ ⫺.15), absenteeism (␳ˆ ⫽ ⫺.26), and
our meta-matrix is based did not distinguish pri-                               focal performance (␳ˆ ⫽ .23). All 95% confidence
mary studies with regard to time (except Bycio                                  intervals for these estimates excluded zero, sup-
[1992]), we returned to the original data to differ-                            porting Hypotheses 1– 4. We note again that these
entiate studies with predictive designs (e.g., job                              results are not biased by percept-percept infla-
satisfaction measured before absenteeism) from                                  tion (Doty & Glick, 1998; Harrison & McLaughlin,
those with postdictive designs (e.g., absenteeism                               1996; Podsakoff, MacKenzie, Lee, & Podsakoff,
measured before job satisfaction). A finding of                                 2003) as we included only those studies for
stronger predictive than postdictive effect sizes                               which correlations were taken from noncommon
would be consistent with the temporal order im-                                 sources.
plied by models A to E.                                                            For two of the hypothesized links (contextual
                                                                                performance with turnover and with focal perfor-
                           RESULTS                                              mance), there was considerable variability in effect
                                                                                sizes not accounted for by sampling error. This
Hypotheses 1–4: Links between Contextual
                                                                                variability implied the presence of potential mod-
Performance and Other Job Behaviors
                                                                                erators, which will be discussed later. Our goal for
  Meta-analytic results for tests of Hypotheses 1                               the present study was to provide the best summary
through 4 appear in Table 1. In keeping with                                    estimates of the links between contextual perfor-
2006                                        Harrison, Newman, and Roth                                            315

mance and other behavioral criteria across settings.          dex (TLI; also known as the nonnormed fit index
Having done so, we could proceed with tests of the            [Tucker & Lewis, 1973]), Jöreskog and Sörbom’s
compatibility principle.                                      (1989) adjusted goodness-of-fit index (AGFI), and
                                                              Steiger’s (1990) root-mean-square error of approxi-
                                                              mation (RMSEA). The last is really a “badness-of-
Models A–E: Increasingly Compatible Job
                                                              fit” index, with larger values indicating greater
Attitude–Job Behavior Constructs
                                                              misfit.
   We tested the compatibility principle by compar-              Two trends are apparent in the Table 3 results.
ing structural equation models (Edwards, 2001) fit-           First, models with more general or more unified
ted to the meta-analytic, corrected correlations              conceptualizations of behavioral criteria fit better
shown in Table 2. Path coefficients for each model            than models with more diversified criterion con-
are shown in Figure 1. Fit indexes for each model             ceptualizations, as evidenced by all indexes. That
are reported in Table 3. As correlations in each cell         is, model D (unified model: CFI ⫽ .92, TLI ⫽ .86,
reflected different sample sizes, the model fit in-           AGFI ⫽ .93, and RMSEA ⫽ .10) fits better than
dexes were based on the harmonic mean across                  model C (diversified model, but unitary with-
meta-analytic cells (Viswesvaran & Ones, 1995). As            drawal construct: CFI ⫽ .87, TLI ⫽ .79, AGFI ⫽ .89,
shown in Figure 1, organizational commitment and              and RMSEA ⫽ .12), which in turn fits better than
job satisfaction were modeled as a single, higher-            model A (diversified model: CFI ⫽.77, TLI ⫽ .65,
order, overall attitude construct in all cases.               AGFI ⫽ .81, and RMSEA ⫽ .16). This finding, based
   Models of the criterion space took several forms.          on a very large sample of employees in natural
These included diversified models A and B, with               settings, supports the compatibility principle. It
five separate criteria; diversified model C, with             also empirically attests to the viability of individual
three separate criteria (lateness, absenteeism, and           effectiveness as a higher-order criterion for overall
turnover reflecting a higher-order withdrawal con-            job attitude. It is reasonable to think of job satisfac-
struct); and unified models D and E, represented              tion and organizational commitment not as unique
with a single, higher-order effectiveness construct           predictors of specific performance criteria or with-
for all the behavioral criteria. Model B was nested           drawal tendencies, but as predictors of a general
within A, and model E, within D, so that chi-square           response that involves the overall engagement
differences could be used to test changes in fit              with, or contribution of favorable efforts to, one’s
within these pairs (Hu & Bentler, 1999). The se-              work role.
quence of all five models was not nested; therefore,             A second trend evident in Table 3 is the superi-
comparisons were made on other indexes. Because               ority of models with progression of withdrawal.
the various indexes differ in specific assumptions,           Model B fits much better than model A (⌬␹2 ⫽
the use of multiple indexes is recommended (Jöres-           648.7, df ⫽ 2, p ⬍ .01), and model E fits consider-
kog & Sörbom, 1989). We included Bentler’s (1990)            ably better than model D (⌬␹2 ⫽ 185.6, df ⫽ 2, p ⬍
comparative fit index (CFI), the Tucker-Lewis in-             .01). Even when an effectiveness construct is spec-

                                                  TABLE 3
                                    Attitude-Behavior Model Comparisonsa

                               Number of
                              Performance
         Models               Dimensions                ␹2         df       RMSEA          TLI        CFI       AGFI

Model A: Diversified               5             1,144.42**        14          .16         .65        .77         .81
 criteria
Model B: Diversified               5               495.68**        12          .11         .81        .89         .90
 criteria, progression of
 withdrawal
Model C: Diversified               3               596.42**        13          .12         .79        .87         .89
 criteria, unitary
 withdrawal construct
Model D: Unified criterion         1               380.76**        13          .10         .86        .92         .93
Model E: Unified criterion,        1               195.15**        11          .07         .92        .96         .96
 progression of
 withdrawal

  a
   Harmonic mean N ⫽ 3,120.
  ** p ⬍ .01
316                                         Academy of Management Journal                                     April

ified as the behavioral criterion for overall job atti-       the predictive effect sizes were larger than postdic-
tude, fit improves as models include an ordered               tive effect sizes (binomial p ⫽ .003). These trends,
sequence of relationships from the unique variance            however, should not be taken as a definitive test of
of lateness to absenteeism to turnover. In fact,              temporal sequence, as confidence intervals for pre-
model E, which has the unified effectiveness crite-           dictive and postdictive estimates overlapped (with
rion and progression of withdrawal, demonstrates              the exception of relationships in which absentee-
fit that might be considered a close approximation            ism was a consequence of job attitudes). In sum,
to true structural relationships in the population            our synthesis of 221 primary studies that employed
(Hu & Bentler, 1999: CFI ⫽.96, TLI ⫽ .92, AGFI ⫽              time-lagged designs gave initial evidence about at-
.96, and RMSEA ⫽ .07). The significant chi-square             titude-behavior sequencing. It revealed a statisti-
suggests there is still room for improvement, but             cally significant overall trend favoring temporal
this statistic is sensitive to trivial model departures       precedence for attitudes and a progression of with-
in our very large samples, and it is also inflated by         drawal behaviors.
the nonnormal distributions of turnover, lateness,
and absenteeism (Harrison & Hulin, 1989).
                                                                                 DISCUSSION
   Despite the inclusion of progression of with-
drawal, the unified criterion model E remains con-               This study addresses several theoretical ques-
sistent with the compatibility principle of attitude          tions that previous empirical research has left un-
theory. In this model, the structural path that joins         answered. We provide the first large-scale empiri-
overall job attitude with effectiveness is rather             cal test of the compatibility principle (Fishbein &
strong (␥standardized ⫽ .59, p ⬍ .01). As this model is       Ajzen, 1975) for job attitudes and work behavior,
based on only a single predictor and a single crite-          following conjectures from theorists such as Fisher
rion, the standardized structural path can be taken           (1980) and Hulin (1991). By noting that job satis-
as a correlation between latent constructs. This la-          faction and organizational commitment are atti-
tent correlation is markedly stronger than even the           tudes that specify a target but do not specify any
recently updated job satisfaction–performance esti-           particular action, we hypothesized and demon-
mate from Judge et al. (2001; ␳ˆ ⫽ .30). Therefore, to        strated that a general set of actions at work—not
answer the question in our paper’s title, overall job         specific behaviors—serves as the best criterion con-
attitude has considerable importance for under-               struct for overall job attitudes. According to com-
standing behavioral outcomes.                                 peting theoretical positions, job attitudes should
                                                              preferentially predict withdrawal behaviors
                                                              (Vroom, 1964), or job attitudes serve simply as a
Time Sequencing of Job Attitudes and Job
                                                              common cause for a variety of otherwise unrelated
Behaviors
                                                              behaviors (Johns, 1998). Instead, results of our
   To test the time ordering between variables im-            meta-analytic study support a unified criterion
plied by the directed arrows in models A through E,           model on the basis of its relative and absolute fit,
we meta-analyzed primary studies using time-                  and show that when attempting to understand pat-
lagged designs. All correlations involving turnover           terns of work behavior from attitudes such as job
were predictive (turnover was always a lagged cri-            satisfaction and organizational commitment, re-
terion). Our literature search found 564 of the orig-         searchers should conceptualize the criterion at a
inal 667 primary effects from the published and               high level of abstraction. A general job attitude is
new meta-analyses. Because 50 of these failed to              strongly linked to a general behavioral criterion. In
report enough information on the time ordering                our conceptualization, the higher-order effective-
between measures, 514 were eventually coded for               ness construct might be defined as a general ten-
time sequencing. Overall, 82 effects were predic-             dency of employees to contribute desirable inputs
tive, 139 effects were postdictive, and 292 effects           toward their work roles rather than withhold those
were concurrent. Table 4 presents results of this             inputs. A simple label for such a conceptualization
analysis.                                                     might be the “attitude-engagement” model of job
   Of the 14 bivariate relationships shown in the             attitudes and behaviors.
table, 2 (focal performance with lateness and job                An alternative, formative view of these behav-
satisfaction with contextual performance) provided            ioral criteria might characterize our models as mis-
no data to permit comparisons of predictive versus            specified (e.g., Jarvis, MacKenzie, & Podsakoff,
postdictive designs. Of the remaining 12 relation-            2003). In such a view, each behavior does not “re-
ships, 10 were in the hypothesized direction (i.e.,           flect” an underlying construct (as in the conven-
consistent with the directions of arrows in Figure            tional approach to constructs [Edwards & Bagozzi,
1). That is, for 10 out of 12 bivariate relationships,        2000]), but instead, all behaviors add together to
2006                                                 Harrison, Newman, and Roth                                               317

                                                  TABLE 4
             Attitude-Behavior (A-B) and Behavior-Behavior (B1-B2) Meta-Analytic Correlations by
                                       Temporal Order of Constructsa

                                                                                                            Percentage of Variance
                                                  Uncorrected          Total   95% Confidence   Corrected      Accounted for by
                 Relationship                         r          k      N         Interval          r           Sampling Error

Job satisfaction (A) with focal performance (B)
  Predictive: A measured, then B                      .17        23 3,251         .14 to .20       .28               30%
  Concurrent: A and B measured at same                .18       130 23,045        .17 to .19       .30               28
     time; separate sources
  Postdictive: B measured, then A                     .19        47    8,713      .17 to .21       .31               17

Organizational commitment (A) with focal
    performance (B)
  Predictive                                          .12         9    2,439      .08 to .16       .16a              34%
  Concurrent, same source                             .17        17    4,634      .14 to .20       .23               17
  Concurrent, separate sources                        .12        28    7,682      .10 to .15       .17               86
  Postdictive                                         .11        16    3,602      .08 to .14       .15a              90

Job satisfaction (A) with absenteeism (B)
  Predictive                                         ⫺.15         7      997    ⫺.21 to ⫺.09     ⫺.24a               91%
  Concurrent, same source                            ⫺.31         2      435    ⫺.39 to ⫺.22     ⫺.49                n.a.
  Concurrent, separate sources                       ⫺.14         1      139    ⫺.30 to .03      ⫺.22                n.a.
  Postdictive                                        ⫺.06        17    2,312    ⫺.10 to ⫺.02     ⫺.10a               39%

Organizational commitment (A) with
    absenteeism (B)
  Predictive                                         ⫺.12        10    3,484    ⫺.15 to ⫺.09     ⫺.18a               38%
  Concurrent, same source                             .03         1      252    ⫺.09 to .15       .04                n.a.
  Concurrent, separate sources                       ⫺.21         1      114    ⫺.38 to ⫺.03     ⫺.30                n.a.
  Postdictive                                        ⫺.05        14    1,501    ⫺.10 to .00      ⫺.07a               37%

Focal performance (B1) with absenteeism (B2)b
  Predictive: B1 measured, then B2                   ⫺.33         8    1,134    ⫺.45 to ⫺.21     ⫺.57a               54%
  Concurrent: B1, B2 measured at the same            ⫺.20        15    7,749    ⫺.26 to ⫺.15     ⫺.35                53
    time; same source
  Concurrent: B1, B2 measured at the same            ⫺.26         5     722     ⫺.33 to ⫺.19     ⫺.45                38
    time; separate sources
  Postdictive: B2 measured, then B1                  ⫺.11        19    4,463    ⫺.18 to ⫺.05     ⫺.20a               25

Job Satisfaction (A) with lateness (B)
  Predictive                                         ⫺.16         2      677    ⫺.23 to ⫺.08     ⫺.19a               n.a.
  Concurrent (same source)                           ⫺.09         5    1,455    ⫺.14 to ⫺.04     ⫺.11                21%
  Postdictive                                        ⫺.14         7    1,235    ⫺.19 to ⫺.08     ⫺.17a               52%

Organizational commitment (A) with lateness
    (B)
  Predictive                                         ⫺.20         1      402    ⫺.29 to ⫺.10     ⫺.21a               n.a.
  Concurrent                                         n.a.       n.a.    n.a.        n.a.          n.a.               n.a.
  Postdictive                                        ⫺.07         4    1,059    ⫺.13 to ⫺.01     ⫺.08a               27%

Focal performance (B1) with lateness (B2)
  Predictive                                         n.a.       n.a.    n.a.        n.a.          n.a.               n.a.
  Concurrent (same source)                           ⫺.02         2      545    ⫺.11 to .06      ⫺.03                n.a.
  Postdictive                                        ⫺.25         2    1,325    ⫺.29 to ⫺.19     ⫺.30                n.a.

Lateness (B1) with absenteeism (B2)
  Predictive                                          .31         1      324      .21 to .41       .44a              n.a.
  Concurrent (same source)                            .48         4      880      .42 to .53       .67               14%
  Concurrent (separate sources)                       .32         7    2,295      .28 to .35       .45               18%
  Postdictive                                         .24         1      324      .14 to .34       .34a              n.a.

Job satisfaction (A) with contextual
     performance (B)
  Predictive                                          .22         4      807      .20 to .23       .27              100%
  Concurrent, separate sources                        .26        14    4,492      .22 to .30       .32               53%
  Postdictive                                        n.a.       n.a.    n.a.         n.a.         n.a.               n.a.
318                                                  Academy of Management Journal                                                April

                                                                 TABLE 4
                                                                (Continued)

                                                                                                                 Percentage of Variance
                                                    Uncorrected             Total   95% Confidence   Corrected      Accounted for by
                   Relationship                         r               k    N         Interval          r           Sampling Error

Organizational commitment (A)
    with contextual performance (B)
  Predictive                                              .21        9      2,231      .13 to .29       .28a              27%
  Concurrent, same source                                 .19       24      6,288      .15 to .22       .25               66
  Concurrent, separate sources                            .19       16      4,060      .15 to .23       .25               67
  Postdictive                                             .16        3        363      .10 to .22       .21a             100

Contextual performance (B1) with focal
    performance (B2)
  Predictive                                              .20        3      1,102     .09 to .31        .23a              26%
  Concurrent, separate sources                            .21       16      8,182     .17 to .25        .25               18
  Postdictive                                             .07        5        698    ⫺.02 to .16        .09a              67

Contextual performance (B1) with lateness
    (B2)
  Predictive                                            ⫺.14            3    291       .02 to ⫺.31    ⫺.19a               47%
  Concurrent, separate sources                          ⫺.10            1     94          n.a.        ⫺.13                n.a.
  Postdictive                                           ⫺.07            1    193          n.a.        ⫺.09a               n.a.

Contextual performance (B1) with absenteeism
    (B2)
  Predictive                                            ⫺.21            2    339     ⫺.20 to ⫺.22     ⫺.24               100%
  Concurrent, separate sources                          ⫺.18            3    331     ⫺.17 to ⫺.19     ⫺.21               100
  Postdictive                                           ⫺.26            3    287     ⫺.12 to ⫺.40     ⫺.30                59

  a
      Predictive-postdictive difference is in hypothesized direction.
  b
      Estimates taken from Bycio (1992).

“form” a new construct. In the present context,                             (see Locke, 1970), but the combined evidence from
such an argument would also maintain that overall                           dozens of time-lagged studies tends to favor the
job attitude first contributes to individual tenden-                        attitude-behavior mechanism.
cies to engage in specific behaviors and that these                            Work behaviors also appear to display a theoret-
behaviors then coalesce or combine (no positive                             ically meaningful time sequence. Although we con-
covariation is necessary) to cause the general be-                          ceptualized effectiveness as a unified tendency, ac-
havioral criterion. It is crucial for us to note that                       tions reflecting this tendency come about in a way
this formulation is theoretically at odds with the                          that suggests progression of withdrawal (Rosse &
compatibility principle, which posits that the gen-                         Miller, 1984). Time-lagged data show that lateness
eral tendency to behave favorably toward an object                          tends to precede absence, and absence predicts
(in this case, one’s work role) is caused by an over-                       turnover. Additionally, as Chen and coauthors
all, positive evaluation of that object. Our conten-                        (1998) explained, the withholding of contextual
tion is that we tested substantive models specifying                        performance behaviors may be a part of this pro-
that (1) there is a general construct of effectiveness                      gression sequence (as the first signal of reduction in
and (2) job behaviors reflect it, but not in identical                      work role inclusion), invoked prior to tardiness or
ways (specific work criteria are congeneric, not par-                       quitting. The time-lagged studies concur, again
allel). Both of these statements are consistent with                        showing a (slight) tendency for contextual perfor-
attitude theory (Ajzen, 1988).                                              mance to precede lateness. To further integrate this
   In addition to validating a unified criterion for                        concept with our theoretical models, we tried a
job attitudes, we found that attitudes and behaviors                        modification to model E, specifying contextual per-
may occur according to a particular time-ordered                            formance as the first element in the progression
sequence. For six out of seven attitude-behavior                            sequence among disturbance terms (i.e., contextual
pairings, predictive correlations (attitude to behav-                       performance to lateness, lateness to absence, and
ior) were stronger than postdictive correlations (be-                       absence to turnover) and found a slight improve-
havior to attitude). We cannot rule out the possibil-                       ment in overall model fit (⌬␹2 ⫽ 5.1, df ⫽ 1, p ⬍
ity that reciprocal causal processes are operating                          .05). Thus, both our static model and our lagged
You can also read