Vous êtes sur la page 1sur 16

Academy of Management fleview

1998, Vol. 23, No. 4, 741-755.

TOWARD A TAXONOMY OF
MULTIDIMENSIONAL CONSTRUCTS
KENNETH S. LAW
Hong Kong University of Science & Technology

CHI-SUM WONG
Chinese University of Hong Kong

WILLIAM H. MOBLEY
Hong Kong University of Science & Technoiogy
PDI Global Research Consortia

We propose a taxonomy oi multidimensional constructs based on the relations be-


tween the construct and its dimensions. Multidimensional constructs that exist at
deeper levels than their dimensions we term latent model. We call constructs formed
as algebraic functions of their dimensions aggregate model, whereas constructs
formed as different profiles of dimensional characteristics we term profile model. We
discuss the nature of multidimensional constructs defined under these models and
their operationalizations in empirical research.

We refer to a construct as multidimensional management literature, in which relations with


when it consists of a number of interrelated their dimensions are unspecified. A good exam-
attributes or dimensions and exists in multidi- ple is organizational citizenship behavior (OCB).
mensional domains. In contrast to a set of inter- Organ (1988) proposes that OCB consists of five
related unidimensional constructs, the dimen- dimensions: civic virtue, sportsmanship, altru-
sions of a multidimensional construct can be ism, conscientiousness, and courtesy. However,
conceptualized under an overall abstraction, there is no discussion in the literature on OCB
and it is theoretically meaningful and parsimo- about the relations between OCB as a multidi-
nious to use this overall abstraction as a repre- mensional construct and its five dimensions. We
sentation of the dimensions. For example, an argue that the nature of a multidimensional con-
employee's satisfaction with his or her work, struct differs when different interpretations are
coworkers, supervisor, promotion, and pay are attributed to the relations between the overall
five interrelated constructs that can be grouped construct and its dimensions. For example, the
together theoretically under an overall abstrac- meaning of OCB when it is defined as the com-
tion called ;ojb safis/action (Locke, 1969; Smith, mon factor underlying its five dimensions is dif-
Kendall, & Hulin, 1969). Since the dimensions ferent from the case when OCB is defined as a
enrichbut at the same time constrainour un- simple sum of its dimensions. The relations be-
derstanding of the overall multidimensional tween OCB and other constructs in its nomolog-
construct, a necessary condition for a multidi- ical network also differ when different mean-
mensional construct to be well defined is that ings are attributed to the OCB construct based
the relations between the overall construct and on its relations with its dimensions.
its dimensions must be specified. Without a Therefore, our major purpose in this article is
specification of these relations, one cannot de- to delineate the various possible ways a multi-
rive the overall construct from its dimensions dimensional construct can relate to its dimen-
and can only conduct research at the dimen- sions. The discussion results in a proposed tax-
sional level, even though these dimensions are onomy of multidimensional constructs. The
claimed theoretically to be under an overall con- framework of the article is as follows. We first
struct. discuss the necessary elements of a good clas-
Unfortunately, there are many multidimen- sification system and explain why our proposed
sional constructs developed by scholars in the taxonomy is an acceptable one. This is followed
741
742 Academy of Management Review October

by an introduction of the two decision rules for mensional constructs in the next section. We
our proposed taxonomy, and the definitions and also discuss operationalization of the taxonomy
characteristics of the three models of multidi- (step 2), as well as conceptual similarities and
mensional constructs formed using these two differences (step 3) between the models. Finally,
decision rules. We then discuss the conceptual in order to distinguish clearly between the three
and practical problems of declaring a construct models of multidimensional constructs, we in-
as multidimensional without specifying the re- troduce two decision rules (step 4)relational
lation between the dimensions and the overall level and relational formto guide us in our
construct. Finally, we discuss issues concerning classification of multidimensional constructs.
operationalization of multidimensional con-
structs and the implications of different types of
multidimensional constructs for research de- THE PROPOSED TAXONOMY OF
sign. MULTIDIMENSIONAL CONSTRUCTS
We highlight here that, although many exam- According to Fleishman and Quaintance
ples used in the following discussions are (1984), the most powerful determinant of a clas-
drawn from the literature of microlevel organi- sification system is its intended purpose. Our
zational behavioral research, the conceptualiza- major purpose for proposing a classification
tion of multidimensional constructs can be system for multidimensional constructs is to
equally generalized to other areas of manage- challenge researchers to clearly define the rela-
ment studies. In fact, we have deliberately cho- tions between a multidimensional construct and
sen some examples from macrolevel strategic its dimensions. Without specifying these rela-
research in order to illustrate the broad applica- tions, the various dimensions are simply a col-
tion of our discussion. lection of related variables, and there is no need
to label them as components of a multidimen-
sional construct. In order to define our three
OUR PROPOSED TAXONOMY AS A
models of multidimensional constructs, we in-
CLASSIFICATION SYSTEM
troduce two classification criteria: (1) relational
Fleishman and Mumford (1991) define a clas- level and (2) relational form.^
sification system as a set of specified rules for Relational level indicates whether the multi-
describing the structure of and relations among dimensional construct exists at a deeper and
a set of objects drawn from some domain that more embedded level than its dimensions or
permits similar units to be assigned to a smaller whether it exists at the same level, as a combi-
number of categories. They identify four basic nation of its dimensions. In the former case, the
steps that are required to generate a classifica- multidimensional construct is an unobservable
tion: construct that underlies different dimensions.
Each dimension of the multidimensional con-
1. specification of the domain of objects to be struct is a different manifestation or realization
classified,
2. definition and measurement of the essen- of the construct. Conceptually, one can think of
tial properties of objects lying in this do- the multidimensional construct under this con-
main, dition as a higher-order abstraction underlying
3. appraisal of the relative similarity of these its dimensions. We label multidimensional con-
objects to each other, and structs that belong to this category as latent
4. specification of decision rules for determin- model, li the multidimensional construct exists
ing when objects display sufficient similar-
ity to permit assignment to a common cate- at the same level as its dimensions and is de-
gory. fined as a combination of its dimensions, in-
stead of as a latent construct underlying its di-
We have strictly followed Fleishman and mensions, we need to invoke the relational form
Mumford's guidelines in our development of the criterion.
proposed taxonomy of multidimensional con-
structs. The domain of objects to be classified
(step 1) in our proposed taxonomy is all multidi-
mensional constructs in the management stud- ' We thank an anonymous reviewer, who assisted us in
ies area. We define the three models of multidi- developing these two classification criteria.
1998 Law, Wong, and Mobley 743

Relational iorm indicates whether the multi- Type I: The Latent Model
dimensional construct can be formed as an
algebraic function of its dimensions. This We classify a multidimensional construct as
classification rule applies only if the multi- latent model if it is a higher-level construct that
dimensional construct does not exist at a deeper underlies its dimensions. For latent multidimen-
conceptual level than its dimensions. In some sional constructs, the dimensions are simply dif-
multidimensional constructs the dimensions of ferent forms manifested by the construct. It is
the construct can be algebraically amalgam- possible to compare the dimensions of a latent
ated into an overall representation of the con- multidimensional construct with the different
struct. We label constructs in this category as items used to measure a unidimensional con-
aggregate model. In other cases, because of the struct in a measurement scale. In a measure-
theoretical nature of the construct, the multidi- ment scale all items are measuring the same
mensional construct is interpreted as various construct, with a certain degree of error. Simi-
profiles formed by pairing the characteristics of larly, different dimensions of a latent multidi-
different dimensions. Here, levels of the multidi- mensional construct represent the same con-
mensional construct are determined by profiling struct, with different degrees of accuracy. Under
levels of each of the dimensions. We label this the latent model, therefore, we define the multi-
as the profile model of multidimensional con- dimensional construct as the commonality
struct in our proposed taxonomy. among the dimensions. However, there is a lim-
Figure 1 shows a diagrammatic representa- itation to this comparison. Whereas the items
tion of the two criteria in this proposed taxon- used to measure a unidimensional construct are
omy and the three resulting models of multidi- all observable variables, the dimensions of a
mensional constructs. In Table 1 we summarize multidimensional construct usually are unob-
the three models of multidimensional constructs servable latent variables.
and give examples from the literature for each One good example of a latent multidimen-
model. Given our two classification criteria and sional construct is general mental ability (GMA),
the three resulting types of multidimensional in the area of selection testing. If we follow the
constructs, we now turn to a discussion of the two-factor theory of Spearman (1927), GMA is
characteristics of each model, some examples of defined as the general latent construct underly-
each, and their similarities and differences. ing different ability dimensions, such as verbal

FIGURE 1
Proposed Taxonomy of Multidimensional Construct

Relational level
(Does the multidimensional
construct exist at the same
level as its dimensions?)

.r I
No Yes

Latent Relational form


model (Can the dimensions be algebraically
combined to form an overall
representation of the construct?)
\
r
No Yes

Profile Aggregate
model model
744 Academy of Management Review October

TABLE 1
A Taxonomy of Multidimensional Constructs, with Examples
Models Examples

Latent model: constructs as underlying higher-order End-user computing satisfaction (Doll, Xia, & Torkzadeh,
abstractions behind their dimensions 1994)
g-factor (Spearman, 1927)
Personality traits (Rounds & Tracey, 1993)
Physical distribution activities (Granzin, 1980)
Role stress (McGee, Ferguson, & Seers, 1989)
Upward-influence tactics (Farmer, Maslyn, Fedor, &
Goodman, 1997)
Aggregate model: constructs exist at the same level as
their dimensions and are formed as a mathematical
function of their dimensions
Additive Dysfunctional thought process (Judge & Locke, 1983)
Job characteristics (Stone, 1976)
Job-related tension (Kahn, Wolfe, Quinn, Fnoek, &
Rosenthal, 1964)
Job satisfaction (Lawler, 1983; Locke, 1969)
Organic management system (Keller, 1978)
Social readjustment (Holmes & Rahe, 1976)
Multiplicative Job motivation (Hackman & Oldham, 1976, 1980)

Profile model: constructs exist at the same level as their Foci of commitment (Becker & Billings, 1993)
dimensions and are formed as various combinations Influence tactics (Kipnis, Schmidt, Swaffin-Smith, &
of their dimensional characteristics Wilkinson, 1984)
Organizational environment (Duncan, 1972)
Personality traits (Cattell, Eber, & Tatsuoka, 1970; Gough,
1987; Hathaway & McKinley, 1967)
Vocational interests (Holland, 1973; Strong, 1943)
Unclassified: constructs without clear specification of the Culture (Hofstede, 1984)
relations between the overall construct and the Organizational citizenship behaviors (Organ, 1988)
dimensions Self-concepts (Marsh & Gouvernet, 1989)
Support for innovation (Siegel & Kaemmerer, 1978)
Time urgency (Landy, Rastegary, Thayer, & Colvin, 1991)

ability, quantitative ability, reasoning ability, construct underlying its three dimensions of
and associative memory. customer relations, inventory management, and
In addition to GMA, there are many constructs system design.
in the management literature that are defined We derive our third example of a latent mul-
as latent model. For example, Granzin (1980) de- tidimensional construct from Farmer, Maslyn,
fines 5 orthogonal second-order factors behind Fedor, and Goodman's (1997) study of the factor
the 15 first-order factors extracted from 58 items structure of upward influence tactics. Farmer et
measuring physical distribution activities. al. uses Schriesheim and Hinkin's (1990) instru-
These five second-order factors are (1) internal ment to identify 3 broad upward-influence strat-
materials movement, (2) information-system op- egies (hard, soft, and rational) out of an 18-item
erations, (3) facilities planning, (4) traffic man- instrument measuring 6 influence tactics (asser-
agement, and (5) external materials movement. tiveness, rationality, coalition, upward appeal,
Since these 5 second-order factors are defined ingratiation, and exchange). These 3 broad in-
as the latent constructs underlying the 15 first- fluence strategies were defined by Farmer et al.
order factors, Granzin has defined 5 multidimen- as the second-order factors behind the 6 influ-
sional constructs with the 15 first-order factors ence tactics, which were, in turn, defined as the
as their dimensions. One of these five con- first-order factors underlying the 18 measure-
structsinformation-system operations, for ex- ment items. Farmer et al. have, therefore, iden-
ampleis defined as a latent multidimensional tified three multidimensional constructs (hard.
1998 Law, Wong, and Mobley 745

soft, and rational upward-influence strategies), other. If the construct is formed from its dimen-
each with two dimensions, and each dimension sions, our second classification criterion
measured by three items in the scale. relational formspecifies how.
In fact. Farmer et al.'s findings provide a per- There are two possible answers to the ques-
fect justification for our argument that a multi- tion of relational form: (1) the dimensions may
dimensional construct is not well defined unless be combined algebraically or (2) they may not.
its relations with its dimensions are theoreti- Constructs defined under the first case are ag-
cally specified. The Farmer et al. results seem to gregate multidimensional constructs, whereas
confirm the theorization of the three multidimen- those defined under the second case are profile
sional constructs labeled as hard, soft, or ration- multidimensional constructs. Under the aggre-
al upward-influence strategies. However, one gate model, the multidimensional construct is
would need further evidence that the three sec- formed as an algebraic composite of its dimen-
ond-order factors collapse into one single under- sions. The multidimensional construct can be a
lying third-order factor if one is interested in an linear or nonlinear function of its dimensions,
overall multidimensional construct defined as and the dimensions may also have unit weights
"the general use of upward-influence strate- or differential weights while forming the overall
gies." In other words, one can end up with three construct.
multidimensional constructs or one single mul- A good example of an aggregate multidimen-
tidimensional construct of upward-influence sional construct is the motivating potential of a
tactics, depending on how one specifies the re- job. Hackman and Oldham (1976, 1980) define
lations between the construct and its dimen- motivating potential as a multidimensional con-
sions. That is, the multidimensional construct of struct, with the degree of skill variety, task au-
upward-influence tactics may indicate an over- tonomy, task significance, task identity, and the
all latent construct showing the degree to which amount of feedback its five dimensions. They
an employee would try to influence his or her also explicitly specify a nonlinear functionthe
supervisor. But upward-influence tactics may Motivating Potential Score (MPS)of the five
also indicate a collection of three multidimen- core job characteristic dimensions in their orig-
sional constructs representing whether an em- inal conceptualization of the job motivation con-
ployee would try to influence his or her supervi- struct.
sor through soft, hard, or rational means. This construct measured by MPSnamely, the
This example illustrates that simply showing motivating potential of a jobis a very good
the existence of some distinct but interrelated example to illustrate the aggregate model. Al-
dimensions does not clearly define what we though it is perfectly legitimate to argue that
mean by the multidimensional construct. Re- these five dimensions form an overall represen-
searchers have to specify the ultimate level tation of the how motivating a job is, it makes no
where the multidimensional construct is defined sense to argue that there exists a latent con-
before they can determine the meaning of the struct called motivating potential that can be
construct. To make the issue more complicated, manifested solely as, for example, task auton-
being the latent higher-level construct underly- omy or task significance. The construct of moti-
ing its dimensions is only one possible way of vating potential, therefore, should, logically be
specifying the relations between a construct defined under the aggregate model, instead of
and its dimensions. We discuss the other possi- the latent model, of multidimensional con-
bilities in the following sections. structs.
Another good example of an aggregate multi-
Type II: The Aggregate Model
dimensional construct is job satisfaction. Locke
(1969) and Lawler (1983) define overall job satis-
The latent model requires that the dimensions faction as the sum of such facets as satisfaction
are manifestations of the multidimensional con- with pay, coworkers, work, supervisor, and pro-
struct, which implies that the construct leads to motion. They also explicitly argue that it was
the dimensions. If this were not true, the only not necessary in their study to use a weighted
possibility would be that the construct is formed sum, because the satisfaction scores themselves
from its dimensions, because the construct and took importance into account (see, for example,
the dimensions cannot be independent of each Mobley & Locke, 1970). No dimension alone can
746 Academy of Management Review October

represent a latent construct called overall job sent 16 different possible personality profiles
satisfaction; this construct exists only as the of individuals.
summed aggregate of its dimensions. Other personality instruments, such as the
Our third example of an aggregate multidi- Minnesota Multiphasic Personality Inventory
mensional construct is social readjustment. (Hathaway & McKinley, 1967), the California
Holmes and Rahe (1978) introduced the construct Psychological Inventory (Gough, 1987), the Six-
of social readjustment to measure how much teen Personality Factor Questionnaire (Cattell,
readjustment a person needs to respond satis- Eber, & Tatsuoka, 1970), and the Big Five person-
factorily to stressful life-changing events. They ality scale (McCrae & Costa, 1989), all attempt to
designed the Social Readjustment Rating Scale represent the personality construct as different
(SRRS) to measure the number of occurrences of profiles of personality dimensions. Scholars nat-
43 life-changing events, such as death of a urally describe personality as various profiles
spouse, marriage, pregnancy, and vacation, in or types, because it is theoretically meaningless
the lives of the respondents during the preced-
to algebraically aggregate introversion and
ing year. For each respondent the number of
occurrences was multiplied by scaling weights conscientiousness to represent how strong the
and summed to create a life-changing score, personality of a person is. In addition, one
representing the multidimensional construct of would not say that there is a latent construct
social readjustment. Each of these 43 life- called "personality" that is manifested as vari-
changing events, therefore, can be viewed as a ous personality dimensions. Instead, each and
dimension. The dimensions combine together every researcher on personality would represent
into a linear composite to form the multidimen- the construct as a profile combination of differ-
sional construct of social readjustment. ent personality dimensions.
Similar to the case of the motivating potential From the area of organizational commitment
of a job, it is illogical to say that social readjust- comes the concept of foci of commitment,
ment is a higher-order latent construct underly- which forms another example of the profile
ing the 43 life-changing events. Instead, each of model. In their conceptualization of employee
the possible events is a component of a con- commitment in the workplace, Becker and Bill-
struct representing the degree of social readjust- ings (1993) developed a scale to measure the
ment a person has to face in a given year. commitment of employees to four foci (the or-
ganization, top management, supervisor, and
work group) and identified four profiles of
Type III: The Profile Model commitment: (1) the "globally committed," who
have strong commitment to the organization
There are some multidimensional constructs and top management; (2) the "committed," who
scholars theorize at the same level as their commit an equal extent to each of the four foci;
dimensions but do not define as algebraic func- (3) the "uncommitted," who have low commit-
tions of their dimensions. Because of their theo- ment to each of the four foci; and (4) the
retical nature, the dimensions of these multidi- "locally committed," who have strong commit-
mensional constructs cannot be combined
algebraically. As a result, researchers usually ment to the work group only.
specify various levels of their dimensions and Our third example of a profile multidimen-
interpret the construct by profiling the levels. sional construct is the widely cited concept of
One obvious example of a profile multidi- organizational environment (Duncan, 1972) in
mensional construct is personality. The Myers- the area of strategic management. Duncan con-
Briggs Type Indicator (MBTI), for example, was ceptualizes organizational environment by two
developed to represent the personality of a dimensionsnamely, the simple-complex and
person in four dimensions: (1) introversion- the static-dynamic dimensions. These two di-
extroversion, (2) sensing-intuition, (3) thinking- mensions are dichotomized and fully crossed to
feeling, and (4) judging-perception (Myers & form four profiles of organizational environ-
McCaulley, 1985). Myers and McCaulley then ment. According to Duncan, these profiles would
fully crossed these 4 dimensions of the multi- have different impacts on various organization-
dimensional construct of personality to repre- al structures and processes.
1998 haw, Wong, and Mobley 747

A Comparison and Contrast of the Three among the dimensions, the dimensions of con-
Models of Multidimensional Constructs structs under the latent model have to be corre-
In the discussion above we have used our two lated. This restriction, however, is not required
evaluation criteriarelational level and rela- for both the aggregate and profile models. We
tional formto distinguish among the three discuss this characteristic of the latent model
models of multidimensional constructs. In addi- more in-depth in the section on operation-
tion to these two evaluation criteria, we can also alization of the three models.
compare and contrast the three proposed mod- In addition to the meaning of the construct, the
els of multidimensional constructs from at least nature of the multidimensional construct is dif-
two different perspectives: (1) interpretation of ferent under the three models. If the dimensions
the nature of the construct and (2) definition of of the construct are continuous in nature, the
true variances of the construct. theorized higher-order construct underlying the
dimensions in the latent model also will be con-
Nature of the construct under the three mod- tinuous. Similarly, an aggregate multidimen-
els. Under both the latent and aggregate mod- sional construct usually will be continuous if the
els, a multidimensional construct can be sum- dimensions are continuous, because the mathe-
marized as a single overall representation of all matical functions used to combine the dimen-
the dimensions; in the latent model the overall sions usually are continuous functions.
construct is the latent commonality underlying
the dimensions, whereas in the aggregate However, this is not true for profile multidi-
model the overall construct is the mathematical mensional constructs. In order to form various
composite formed from the dimensions. In con- profiles, one needs to partition each dimension
trast, a profile multidimensional construct can artificially into different discrete levels, even if
be interpreted only as a set of profiled charac- the dimensions are originally continuous in na-
teristics of the dimensions; there is not a single ture. As a result, researchers theorizing about a
theoretical overall construct that summarizes profile multidimensional construct usually
and represents all the dimensions. For example, would end up classifying subjects as belonging
although a person can be identified as high or to or excluded from a particular profile. For ex-
low in general mental ability and job satisfac- ample, a researcher using the MBTI as a mea-
tion, one cannot say that a person is high or low sure of personality usually would conclude
in personality. Personality as a profile multidi- whether a person belongs to the ISTJ type in-
mensional construct can be interpreted only un- stead of how strong an ISTJ type the person is.
der various profiles. A person can score high or Although dichotomization of profiles (i.e., speci-
low in one personality profile but not high or low fying whether or not a person belongs to that
in personality as an overall construct. profile) is the common practice of researchers
defining multidimensional constructs under the
The meaning of a multidimensional construct profile model, we would argue that this is not a
classified as latent model is also totally differ- mandatory requirement of the profile model. We
ent from one classified as aggregate model. Un- discuss the possibility of interpreting a profile
der the latent model the overall latent construct as a continuous variable in the section on opera-
leads to various dimensions of the construct, tionalization of the three models.
because the dimensions are simply different Definition of true variances of the construct.
ways the construct is realized. In contrast, the We can also compare and contrast the three
overall construct is formed from its dimensions models of multidimensional constructs using
under the aggregate model. Since the relational the variance partitioning perspective. Nunnally
direction between the overall construct and its (1978) proposes that the observed variances of a
dimensions is exactly opposite under the latent variable can be partitioned into four elements:
and aggregate models, the meaning of the con- (1) common variances, (2) group variances, (3)
struct also will be theoretically different. This specific variances, and (4) random variances.
difference in meaning between the latent and We would apply this argument to each dimen-
aggregate models also imposes some restric- sion of a multidimensional construct. We show
tions on the relations among the dimensions of three of these four components of the variances
the construct. Since the latent construct under of a dimension in Figure 2 under the latent
the latent model is defined as the commonality model.
748 Academy ol Management Review October

Common variances are those shared by all struct. We treat variances specific to one dimen-
dimensions of the multidimensional construct. sion (specific variances), covariances shared by
They are represented by the blackened area in some dimensions only (group variances), and
the latent model (top diagram) in Figure 2. random variances as error variances in the la-
Group variances are those shared by only tent model. As a result, only the black area in
some dimensions and are represented by the Figure 2 is considered as the true variances of
striped area in the latent model in Figure 2. the latent multidimensional construct.
Specific variances are the variances unique to Under the aggregate model, the multidimen-
a single dimension. They appear as shaded sional construct is defined as an algebraic com-
areas in the latent model in Figure 2. Random posite formed by the dimensions. Specific vari-
variances are observed variances caused by ances and group variances, therefore, are also
random factors. In order to keep the diagrams part of the true variance of the construct, and
simple and clear, we omit random (error) vari- only random variances are considered as error
ances from Figure 2. variances. As a result, we consider the whole
Under the latent model, the multidimensional area marked in gray in the second diagram in
construct is the latent higher-order construct un- Figure 2 as true variances of the aggregate mul-
derlying the dimensions. Therefore, we consider tidimensional construct.
only common variances or covariances shared For the profile model we use the simplified
by all dimensions as true variances of the con- case of dichotomizing each dimension in order

FIGURE 2
Three Proposed Models of Multidimensional Constructs from the Variance Partitioning
Perspective
Dimension 1

Common variances |

Specific variances

Dimension 2 Dimension 3

Latent model

Aggregate model Profile model


1998 Law, Wong, and Mobley 749

to come up with a profile representing the mul- relations between the construct and its dimen-
tidimensional construct for illustrative pur- sions.
poses. The straight lines that cut each ellipse We argue that multidimensional constructs
into two halves are the dichotomization criteria. for which relations with their dimensions are
For one specific profile, which is defined as high not specified are not well developed. The rela-
in all three dimensions, we consider only the tion between the construct and its dimensions is
areas marked in dots as part of the true vari- a necessary element in the definition of a mul-
ances of this particular profile. It is clear in this tidimensional construct for at least three rea-
illustration that, similar to the aggregate model, sons: (1) definition of the research question,
common variances, specific variances, and (2) theoretical parsimony, and (3) relations with
group variances are all considered in defining other constructs.
this profile multidimensional construct. As a re-
sult, depending on the dichotomization criterion,
the portion that is considered as true variances Definition of the Research Question
for the Figure 2 profile may or may not cover the Without correct specifications of the relations
common variances, group variances, or specific between multidimensional constructs and their
variances of a specific dimension. dimensions, one would set up research hypoth-
Viewing multidimensional constructs from eses at the construct level, conduct analyses at
this variance partitioning perspective shows the dimension level, but make conclusions at
that we are not suggesting a system that cate- the construct level. Using national culture as an
gorizes multidimensional constructs based sim- example, if individualism and power distance
ply on their operational definition. Instead, we (Hofstede, 1984) are only two correlated con-
argue that multidimensional constructs defined structs, they can be treated as distinct con-
under the three models are theoietically differ- structs, and it may not be necessary to label
ent constructsdifferent because the definition them as different dimensions of national cul-
of the true variances of the construct differs, and ture. We would not use individualism and power
we cannot claim that they are only different distance as two separate independent con-
operationalizations of the same construct. structs and draw the conclusion that "culture"
has an effect on a dependent variable when
either individualism or power distance shows a
significant relation with that dependent vari-
MULTIDIMENSIONAL CONSTRUCT: THE
able.
UNCLASSIFIED CASES
In contrast, we suggest that the profile model
There are some multidimensional constructs may be a good choice for a multidimensional
in the literature for which the author does not construct such as national culture. Researchers
specify the relations between the overall may develop different profiles of culture and
construct and its dimensions. For example, test the relations between some predictor and
Hofstede (1984) develops four dimensions of the criterion variables under different cultural pro-
culture of a society: individualism, power dis- files. In a more recent work, Hofstede (1994) ap-
tance, masculinity, and uncertainty avoidance. pears to head in that direction.
However, he presents no theoretical discussion
of the relations of these cultural dimensions
with the multidimensional construct of culture. Theoretical Parsimony
Other examples of multidimensional constructs Whereas analyses of the relations between
under this unclassified category include organ- specific dimensions of a multidimensional con-
izational citizenship behavior (Organ, 1988), pro- struct with other constructs may enrich our un-
cedural justice (Folger & Konovsky, 1989), per- derstanding of the construct, treating dimen-
ceived social support (Pierce, Sarason, & sions as a set of individual variables precludes
Sarason, 1991), multidimensional self-concepts any general conclusion about the relations be-
(Marsh & Gouvernet, 1989), and time urgency tween a multidimensional construct and other
(Landy, Rastegary, Thayer, & Colvin, 1991). The constructs. Only when the interrelations be-
authors label all these constructs as multidi- tween a multidimensional construct and its di-
mensional, but leave unclear or unspecified the mensions are specified can we derive overall
750 Academy of Management Review October

and parsimonious conclusions about the role of poses only. Most of the basic ideas in the
the multidimensional construct in its nomologi- following discussion could still be applied when
cal network. Again, it is difficult to draw conclu- using other analytical techniques.
sions about the overall construct unless empiri-
cal analyses are conducted at the construct The Latent Model
level instead of the dimension level.
Since latent multidimensional constructs are
realized through their dimensions, one possible
Relations with Other Constructs way to operationalize the latent model is to rep-
We have discussed that the definition of true resent the multidimensional constructs as the
variance of a multidimensional construct differs common factor underlying their dimensions. In
among the three models. A researcher may, covariance structure analysis the dimensions of
therefore, arrive at different conclusions when a multidimensional construct usually are esti-
using different definitions of a multidimen- mated by a number of observed variables or
sional construct. In fact, MacCallum and Browne indicators. We may view the dimensions as
(1993) and Law and Wong (in press) have dem- common factors underlying the indicators and
onstrated the differences in parameter esti- operationalize the overall multidimensional
mates with different specifications of the rela- construct as a higher-order factor underlying the
tions between the dimensions and the overall dimensions (Hunter & Gerbing, 1982). Depending
multidimensional construct in structural equa- on the nature of the construct, the overall latent
tion modeling. construct(s) can be defined as a group of second-
order factors or one single higher-order factor
underlying the first-order dimensions.
OPERATIONALIZATION OF THE PROPOSED
For example, in the case of general mental
TAXONOMY
ability, we define the g-factor as a single higher-
After having discussed the theoretical impli- order factor underlying the specific ability di-
cations of the proposed taxonomy, we now dis- mensions. In the case of upward-influence tac-
cuss how to operationalize the three possible tics, we define three interrelated constructs
models of multidimensional constructs in actual (hard, soft, and rational strategies) as the factors
research practices. We illustrate this issue in underlying the six influence strategy dimen-
the context of covariance structure analysis or sions.
structural equation modeling. Since we define the latent overall construct as
We have chosen covariance structure as the the commonality among its dimensions, we can
illustrating tool for three reasons. First, factor estimate it from the variance-covariance matrix
analysis and regression analysis are the two among the dimensions of the construct using the
most common techniques used in studies of mul- technique of confirmatory factor analysis. This
tidimensional constructs. Scholars have shown characteristic of the latent model is in sharp
that both of these techniques are simply special contrast to the nature of aggregate multidimen-
cases of the technique of covariance structure sional constructs.
analysis. Therefore, we discuss the empirical
application of the models in a more general The Aggregate Model
context by framing them under the technique of
covariance structure analysis. Second, covari- For the aggregate model we define the multi-
ance structure analysis is gaining widespread dimensional construct as an algebraic function
acceptance and is growing in importance in of its dimensions. As we have discussed, a la-
management research (Medsker, Williams, & tent multidimensional construct can be inter-
Holahan, 1994). Thus, our discussion of the opera- preted as a common factor of its dimensions; a
tionalization of the three models with this tech- common factor is also a linear function of its
nique would be beneficial to the increasing component variables. There are, however, fun-
number of researchers using this technique. damental differences between the conceptual
We emphasize that our discussion of the opera- meaning of the functional relations between the
tionalization of the three models using covari- construct and its dimensions under the latent
ance structure analysis is for illustrative pur- and aggregate models. We can estimate a mul-
1998 Law, Wong, and Mobley 751

tidimensional construct under the latent model common assumption of a linear relationship be-
from the covariance structure of its dimensions tween the construct and its dimensions and es-
because it is conceptualized as the commonality timation of the weights of different dimensions
among the dimensions. In contrast, we concep- based on the relations between the construct
tualize an aggregate multidimensional con- and other constructs in its nomological network.
struct as a composite formed from its dimen- For example, job satisfaction may be defined as
sions. The mathematical relations between the a mathematical aggregate of its five dimensions
composite and the dimensions can be totally with the weights in forming the aggregate un-
independent of the covariance structure of the known. If turnover and absenteeism are two con-
dimensions. For example, multidimensional structs theorized as outcomes of job satisfaction,
constructs under the aggregate model can have the weights of the dimensions in forming the
dimensions that are totally uncorrelated (such aggregate job satisfaction construct can be es-
as the 43 life-changing events on social read- timated through the covariance structure of the
justment)a phenomenon that is unacceptable five job satisfaction dimensions and turnover/
under the latent model. absenteeism. A major characteristic of this
In the context of covariance structure analy- structural model is that there is a core construct
sis, this conceptual difference between the la- called job satisfaction, which has structural re-
tent and aggregate model can be visualized in a lations with its dimensions and the two depen-
path diagram. A latent multidimensional con- dent variables of turnover and absenteeism, but
struct implies that it would have structural there is not any observable indicator of this core
paths or arrows pointing to its dimensions. In construct of job satisfaction.
contrast, there would be structural arrows point- MacCallum and Browne (1993) have specified
ing from the dimensions to form an aggregate two necessary conditions for identifying a struc-
multidimensional construct.^ With exactly oppo- tural model involving the kind of unobserved
site structural directions between the dimen- variables defined as linear composites of other
sions and the construct, the resulting estimates variables. First, a scale must be established for
of structural parameters would be diametrically the composite variable either (1) by fixing the
different for each of the two models. weight for one of the paths from an indicator to
Since we cannot derive, under the aggregate the composite variable at some nonzero value,
model, the exact mathematical form between usually 1.0, or (2) by fixing the variance of the
the construct and its dimensions from the co- composite variable at some nonzero value, usu-
variance structure of the dimensions, we have ally 1.0. Second, the composite variable must
two possibilities for determining the relative emit paths to at least two different latent vari-
weights of each dimension in forming the final ables. These requirements have to be fulfilled
construct. In some cases the exact algebraic for structural models involving aggregate mul-
function can be determined theoretically. For tidimensional constructs when the weights of
example, Hackman and Oldham (1976, 1980) ex- the dimensions in forming the aggregate are
plicitly define overall job characteristics as a unknown.
multiplicative nonlinear function of the five core
job characteristics. Similarly, Locke (1969) ex-
plicitly defines overall job satisfaction as the The Profile Model
Simple sum of its dimensions. A profile multidimensional construct will
There are, however, cases when theories con- cause complicated issues in structural equation
cerning the construct are not detailed enough to analysis. Usually, researchers using the profile
prescribe the exact algebraic relation between model will qrtificially dichotomize each dimen-
the multidimensional construct and its dimen- sion and use different combinations of the di-
sions. For such cases we suggest the use of the chotomized dimensions to form various profiles
of the multidimensional construct. Using Becker
and Billings' (1993) commitment foci as an exam-
^ Note that the .structural paths or arrows do not imply that ple, respondents either belong to the profile la-
these dimensions are external "causes" of the construct. The
dimensions are not causes of the multidimensional con- beled "globally committed" or not, although we
struct; they are part of the definition of the multidimensional have interval measures of their commitments to
construct. different foci.
752 Academy of Management Review October

This artificial dichotomization of multidimen- ing potential that is manifested through differ-
sional constructs after they are defined under ent job characteristics. In other words, it does
the profile model may reduce their observed cor- not make any conceptual sense to argue that a
relations with other variables. To minimize the job is motivating and so it has more job charac-
effects of this problem, we suggest that each teristics (e.g., skill variety). On the contrary, it
profile of the construct can be operationalized makes perfect sense to argue that a job is rich in
empirically as the centroid of all the data points job characteristics (e.g., skill variety) and so it is
classified under the profile using the dichotomi- motivating. Unfortunately, job design studies in
zation criterion. The profile centroid is the point the literature utilizing structural equation mod-
in the multidimensional space where the score eling totally ignore this conceptual and theoret-
on each dimension is equal to the average of all ical characteristic of the construct and use the
the data points classified under this profile. As a dimensions of job characteristics as indicators
result, it possesses the "typical" characteristics of the overall construct (see, for example, James
of all the points defined in the profile. The sim- & Tetrick, 1986, and Mathieu, Hofmann, & Farr,
ilarity of each data point with this profile then 1993) in confirmatory factor analyses.
can be defined as the reversed Euclidean dis- Similarly, although primary job satisfaction
tance between the data point and the profile theorists such as Locke (1969) and Lawler (1983)
centroid in the multidimensional space. Once define job satisfaction as the composite formed
the degree of profile characteristics is defined, from its dimensions, Schmitt and Bedeian (1982),
the multidimensional construct then can be Netemeyer, Johnston, and Burton (1990), and
modeled with other constructs in its nomologi- Vandenberg and Scarpello (1990) operationalize
cal network. job satisfaction as the common factor underly-
Although this operationalization of the profile ing its dimensions in their confirmatory factor
model of a multidimensional construct can solve analyses. It is interesting to note that Parson
the problem of artificial categorization of contin- and Hulin (1982) and Judge and Locke (1993) both
uous dimensions, one limitation of this ap- use the argument that job satisfaction can be
proach is that the profile centroid is empirically defined as the common factor underlying its di-
defined. One would, therefore, need a large rep- mensions because the dimensions are substan-
resentative sample in order to obtain stable and tially correlated. We do not think that a high
valid estimates of the profile centroid. correlation among the dimensions is good evi-
dence that the variables can be defined as la-
tent model. Although we acknowledge that, as
An Integrative Discussion science progresses, some earlier conceptions
Once again, we emphasize the classification about constructs will need reinterpretation, we
of multidimensional constructs into the three feel such reinterpretation should be made with
proposed models as a theoretical, rather than sound theoretical justification and strong empir-
empirical, issue. There is ample evidence that ical support. A high correlation among the di-
many constructs have a strong theoretical basis mensions of a multidimensional constructby
to be classified under one of the three models of itself, and without conceptual justificationis
multidimensional constructs. Personality and not strong enough evidence to reinterpret the
career interests seem to be concrete examples of construct as latent model instead of aggregate
profile constructs, because researchers' primary model.
purpose in introducing these constructs is to Finally, there are special operational con-
group individuals into different profiles. Spear- cerns for multidimensional constructs for which
man's (1927) definition of the g-factor clearly relations between the constructs and their di-
makes it a latent construct. In contrast, job mo- mensions are not specified. Without a strong
tivating potential as defined by Hackman and theoretical basis to define these constructs un-
Oldham (1976) is clearly aggregate model. It is der one of the three models, we suggest that
illogical to represent job motivating potential as logical considerations may be used with empir-
latent model because Hackman and Oldham ical means to test the choice of a specific theo-
discuss job motivating potential as an outcome retical model of multidimensional constructs.
of job characteristics. It is difficult to argue that We use the construct of organizational citizen-
there exists an abstract construct of job motivat- ship behaviors (OCB) as an illustration.
1998 Law, Wong, and Mobiey 753

Organ (1988) proposes that OCB consists of DISCUSSION AND CONCLUSIONS


five dimensions: (1) civic virtue, (2) sportsman-
ship, (3) altruism, (4) conscientiousness, and Although there has been limited discussion in
(5) courtesy. Since Organ does not clearly define the existing literature about how constructs can
the interrelations between the construct and its be formed from their dimensions (Foa, 1965),
dimensions, we argue that OCB may be defined scholars have never proposed a systematic tax-
under all three models of multidimensional con- onomy of multidimensional constructs. We pro-
struct. pose, in this article, a taxonomy based primarily
on the relations between a multidimensional
One can, for example, define OCB as an alge- construct and its dimensions. We argue that a
braic functionor simple sumof the five di- multidimensional construct in which relations
mensions. Under this aggregate model, an em- with its dimensions are not specified is not well
ployee is expressing OCB if she or he shows one defined. Laskey, Day, and Crask (1989) have
or more of the five named behavioral dimen- identified three requirements for a good classi-
sions. However, one can also define OCB as the fication scheme. First, a good classification
conimon or latent factor underlying these five scheme should lead to mutually exclusive and
dimensions. An employee should then express a exhaustive categories. Second, the classifica-
great extent of each and every citizenship be- tion scheme should capture meaningful differ-
havior dimension in order to score high on the ences of the objects being classified, while re-
OCB construct. maining parsimonious. Third, the classification
Finally, one can identify different profiles of scheme should be operational. There should ex-
employees' expression of OCB and use these ist some strategies so that objects are consis-
profiles to represent the multidimensional con- tently categorized according to the rules for
struct of OCB. For example, one may use the classification.
technique of cluster analysis to group the sub- Our proposed classification scheme has met
jects in a sample into different clusters with all three requirements. First, we have argued
different characteristics on civic virtue, sports- that the three models of multidimensional con-
manship, altruism, conscientiousness, and cour- structs should be theoretically determined by
tesy. After identifying one meaningful cluster as our two evaluation criteria: relational level and
the OCB profile of interest (e.g., high on altruism relational form. The two evaluation criteria will
and low on all other four dimensions), one can lead to three mutually exclusive (latent, aggre-
then define the level of OCB of a specific subject gate, and profile) models of multidimensional
as the negative Euclidean distance of the OCB constructs. Since the answers to our two evalu-
ratings of this subject from the centroid of the ation criteria are either yes or no, our proposed
identified cluster in the multidimensional classification must be exhaustive. Second, we
space. have discussed the conceptual differences in
After showing the above three operation- the meaning of multidimensional constructs de-
alizations of the multidimensional construct of fined under these three models. We have, there-
OCB, a researcher then can test the comparative fore, captured meaningful differences among
accuracy of the three definitions of OCB. We different multidimensional constructs, while
suggest that model accuracy can be determined keeping the total number of models to a parsi-
by at least two evaluation criteria. First, under a monious number of three. Finally, our two eval-
correct definition of OCB, the OCB construct uation criteria of relational level and relational
should have expected relations with other con- form will lead to clear operational guidelines for
structs in its nomological network in a criterion researchers in defining their multidimensional
validity sense. Second, with the correct specifi- constructs.
cation of the relations between the construct We believe this proposed classification
and its dimensions, one should attain accept- scheme will clarify the fundamental character-
able goodness-of-fit indices in the covariance istics of various multidimensional constructs.
structure analysis. The model that best fits the This clarification is important for both concep-
above two criteria then could be concluded to be tual and empirical reasons. Conceptudlly, the
an acceptable definition of the multidimen- meaning of the multidimensional constructs un-
sional construct of OCB. der the three models will be totally different.
754 Academy of Management Review October

Empirically, a multidimensional construct cate- Granzin, K. L. 1980. Physical distribution: A description of its
structure by means of second-order factor analysis. Jour-
gorized as latent model may be operationalized nal of Business Research, 8: 213-223.
as the common factor underlying its dimensions.
Hackman, J. R., & Oldham, G. R. 1976. Motivation through the
Under the aggregate model, the multidimen- design of work: Test of a theory. Organizational Behav-
sional construct may be operationalized as the ior and Human Performance, 16: 250-279.
composite construct formed from its dimensions. Hackman, J. R., & Oldham, G. R. 1980. Work redesign. Read-
Finally, we suggest that the reversed distance of ing, MA: Addison-Wesley.
a data point from the centroid of the profile can Hathaway, S. R., & McKinley, J. C. 1967. Manuai for the
be used as an operational definition of the mag- Minnesota Multiphasic Personality Inventory. New York:
nitude of profile characteristics of the data Psychological Corporation.
point. Hofstede, G. 1984. Culture's consequences: International dif-
Given the serious conceptual and empirical ferences in wort-related values. Newbury Park, CA:
implications for the classification of multidi- Sage.
mensional constructs and the existence of un- Hofstede, G. 1994. Management scientists are human. Man-
classified constructs in the literature, we agement Science, 40: 4-13.
strongly recommend more research on the re- Holland, J. L. 1973. Maiing vocational choices: A theory of
examination of these unclassified constructs. careers. Englewood Cliffs, NJ: Prentice-Hall.
Future empirical research should clearly dis- Holmes, T. H., & Rahe, R. H. 1976. The social readjustment
cuss and classify the multidimensional con- rating scale. Journal of Psychosomatic Research, 1: 213-
structs so that the appropriate means of opera- 218.
tionalization can be adopted. Hunter, J. E., & Gerbing, D. W. 1982. Unidimensional mea-
surement, second order factor analysis, and causal mod-
els, flesearch in Organizational Behavior. 4: 267-320.
REFERENCES James, L. R., & Tetrick, L. E. 1986. Confirmatory analytic tests
of three causal models relating job perceptions to job
Becker, T. E., & Billings, R. S. 1993. Profiles of commitment: An
satisfaction. Journal of Applied Psychology, 71: 77-82.
empirical test. Journal of Organizational Behavior, 14:
177-190. Judge, T. A., & Locke, E. A. 1993. Effect of dysfunctional
thought processes on subjective well-being and job sat-
Cattell, R. B., E3Der, H. W., & Tatsuoka, M. M. 1970. Handbook
isfaction. Journal of Applied Psychology, 78: 475-490.
for Sixteen Personality Factor Questionnaire (16PF).
Champaign, IL: Institute for Personality and Ability Kahn, R. L., Wolfe, D. M., Quinn, R. P., Fnoek, J. D., &
Testing. Rosenthal, R. 1964. OrganizafionaJ stress: Studies in role
conflict and ambiguity. New York: Wiley.
Doll, W. J., Xia, W., & Torkzadeh, G. 1994. A coniirmatory
iactor analysis of the end-user computing satisfaction Keller, R. T. 1978. Dimensions of management system and
instrument. MIS Quarterly, 18: 453-461. performance in continuous-process organizations. Hu-
man i?elations, 31: 5975.
Duncan, R. B. 1972. Characteristics of organizational envi-
ronments and perceived environmental uncertainty. Ad- Kipnis, D., Schmidt, S. M., Swaffin-Smith, C, & Wilkinson, I.
ministrative Science Quarterly, 17: 313-327. 1984. Patterns of managerial influence: Shotgun manag-
ers, tacticians, and bystanders. Organizational Dynam-
Farmer, S. M., Maslyn, J. M., Fedor, D. B., & Goodman, J. S.
ics, 12(Winter): 58-67.
1997. Putting upward influence strategies in context.
Journal of Organizational Behavior, 18: 17-42. Landy, F. J., Rastegary, H., Thayer, J., & Colvin, C. 1991. Time
urgency: The construct and its measurement. Journal of
Fleishman, E. A., & Mumford, M. D. 1991. Evaluating classi-
Applied Psychology, 76: 644-657.
fications of job behavior: A construct validation of the
ability requirement scales. Personnel Psychology, 44: Laskey, H. A., Day, E., & Crask, M. R. 1989. Typology of main
523-575. message: Strategies for television commercials. Journal
of Advertising, 18: 36-41.
Fleishman, E. A., & Quaintance, M. K. 1984. Taxonomies of
human performance: The description of human tasks. Law, K. S., & Wong, C. S. In press. Multidimensional con-
Orlando, FL: Academic Press. structs in structural equation analysis: An illustration
using the job perception and job satisfaction constructs.
Foa, U. G. 1965. New developments in facet design and
Journal of Management.
analysis. Psychological Review, 72: 262-274.
Lawler, E. E., III. 1983. Satisfaction and behavior. In R. M.
Folger. R., & Konovsky, M. A. 1989. Effects of procedural and
Steers, & L. W. Porter (Eds.), Motivation and work behav-
distributive justice on reactions to pay raise decisions.
ior {3rd ed.): 332-345. New York: McGraw-Hill.
Academy o/. Management Journal, 32: 115-130.
Locke, E. A. 1969. What is job satisfaction? Organizational
Gough, H. G. 1987. California psychological inventory ad-
Behavior and Human Performance, 4: 309-336.
ministrator's guide. Palo Alto, CA: Consulting Psychol-
ogists Press. MacCallum, R. C, & Browne, M. W. 1993. The use of causal
1998 law, Wong, and Mobley 755

indicators in covariance structure models: Some practi- item response theory and hierarchical factor analysis in
cal issues. Psychological Bulletin, 114: 533-541. applications to the measurement of job satisfaction.
Marsh, H. W., & Gouvernet, P. J. 1989. Multidimensional self- Journal of Applied Psychology, 67: 826-834.
concepts and perceptions of control: Construct valida- Pierce, G. R.. Sarason, I. G., & Sarason, B. R. 1991. General
tion of responses by children. Journal of Educational and relationship-based perceptions of social support:
Psychology, 81: 57-69. Are two constructs better than one? Journal of Personal-
Mathieu. J. E., Hofmann, D. A.. & Farr, J. L. 1993. Job percep- ity and Social Psychology, 61: 1028-1039.
tion-job satisfaction relations: An empirical comparison Rounds, J., & Tracey. T. J. 1993. Prediger's dimensional rep-
of three competing theories. Organizational Behavior resentation of Holland's RIASEC circumplex. Journal of
and Human Decision Processes. 56: 370-387.
Applied Psychology, 78: 875-890.
McCrae. R. R.. & Costa. P. T., Jr. 1989. The structure of inter-
Schmitt. N.. & Bedeian. A. G. 1982. A comparison of LISREL
personal traits: Wiggin's circumplex and the five-factor
and two-stage least squares analysis of a hypothetical
model. Journal of Personality and Social Psychology, 56:
life-job satisfaction reciprocal relationship. Journal oi
586-595.
Applied Psychology, 67: 806-817.
McGee, G. W., Ferguson, C. E., Jr., & Seers, A. 1989. Role
conflict and role ambiguity: Do the scales measure these Schriesheim, C. A., & Hinkin, T. R. 1990. Influence tactics
two constructs? Journal of Applied Psychology, 74: 815- used by subordinates: A theoretical and empirical anal-
818. ysis and refinement of the Kipnis. Schmidt, and Wilkin-
son subscales. Journal of Applied Psychology, 75: 246-
Medsker, G. J., Williams, L. J., & Holahan. P. J. 1994. A review 257.
of current practices for evaluating causal models in
organizational behavior and human resources manage- Siegel, S. M., & Kaemmerer, W. F. 1978. Measuring the per-
ment research. Journal of Management, 20: 439-464. ceived support for innovation in organizations. Journal
of Applied Psychology, 63: 553-562.
Mobley, W. H., & Locke, E. A. 1970. The relationship of value
importance to satisfaction. Organizafionai Behavior and Smith, P. C . Kendall, L. M., & Hulin, C. L. 1969. The measure-
Human Per/ormance- 5: 463-483. ment of satisfaction in work and retirement: A strategy
Myers, I. B., & McCaulley, M. 1985. Manuai; A guide to the for the study of attitudes. Chicago: Rand McNally.
development and use of the Myers-Briggs Type Indicator. Spearman, C. 1927. The abilities of man. New York: Macmil-
Palo Alto. CA: Consulting Psychologists Press. lan.
Netemeyer, R. G., Johnston, M. W., & Burton, S. 1990. Analysis Stone, E. F. 1976. The moderating effect of work-related val-
of role conflict and role ambiguity in a structural equa- ues on the job scope-job satisfaction relationship. Or-
tions framework./oumaJ of Applied Psychology, 75: 148- ganizationai Behavior and Human Performance, 15: 147-
157. 167.
Nunnally, J. C. 1978. Psychomefric theory (2nd ed.). New York: Strong, E. K. 1943. Vocafionai interests of men and women.
McGraw-Hill. Stanford. CA: Stanford University Press.
Organ. D. W. 1988. Organizafionai citizenship behavior: The Vandenberg, R. J., & Scarpello. W. 1990. The matching model:
good soldier syndrome. Lexington. MA: Lexington Books.
An examination of the processes underlying realistic job
Parson, C. K., & Hulin, C. L. 1982. An empirical comparison of preview. Journal of Applied Psychology, 75: 60-67.

Kenneth S. Law received his Ph.D. in human resource management from the Univer-
sity of Iowa. He is an associate professor in the Department of Management of the
Hong Kong University of Science & Technology. His research focuses on human
resource management, management in the People's Republic of China, and method-
ological issues in management research.
Chi-Sum Wong is an associate professor of management at the Chinese University of
Hong Kong. He received his Ph.D. in organizational behavior and human resource
management from Purdue University. His research interests include job analysis and
design, application of structural equation modeling, and organizational behavior in
Chinese societies.
William H. Mobley is Vice President of Personnel Decisions International Corporation
(PDI) and Managing Director of PDI's Global Research Consortia. He provides man-
agement and organizational research and consulting services for multinational firms
doing business throughout Asia and elsewhere in the world. He earned his Ph.D. in
industrial-organizational psychology from the University of Maryland. He is best
known for his research, writing, and consulting on employee motivation, turnover, and
selection and international and strategic human resources management.

Vous aimerez peut-être aussi