You are on page 1of 23

^ Academy of Management Revi0w. 1989, Vol. 14, No. 3, 423-444.

The Concept of Fit in Strategy Research:


Toward Verbal and
Statistical Correspondence
N. VENKATRAMAN
Massachusetts Institute of Teclinology
This article develops a conceptual framework and identifies six per-
spectives of fit—fit as moderation, fit as mediation, fit as matching, fit
as gestalts, fit as profile deviation, and fit as covariation—each im-
plying distinct theoretical meanings and requiring the use of specific
analytical schemes. These six perspectives highlight the isomorphic
nature of the correspondence hetween a particular concept and its
subsequent testing scheme(s), but it appears that researchers have
used these perspectives interchangeably, often invoking one per-
spective in the theoretical discussion while employing another in the
empirical research. Because such research practices weaken the crit-
ical link between theory development and theory testing, this article
argues for explicit links between theoretical propositions and opera-
tional tests.

The concept of fit has serveci as an important A lack of correspondence between the con-
building block for theory construction in several cept and its mathematical formulation weakens
areas of research (Aldrich, 1979; Fry & Smith, the link between theory building and theory test-
1987; Thompson, 1967; Van de Ven & Drazin, ing (Blalock, 1965), which may be a major rea-
1985), including strategic management (Miles & son for inconsistent research results. Indeed,
Snow, 1978; Snow & Miles, 1983; Venkatraman Galbraith and Nathanson's observation, made
& Camillus, 1984). However, a major problem is a decade ago, is equally valid today: "Although
the lack of corresponding schemes by which fit the concept of fit is a useful one, it lacks the
has been tested. Although it is common for the- precise definition needed to test and recognize
orists to postulate relationships using phrases whether an organization has it or not" (1979, p.
and words such as matched with, contingent 266). In a similar tone. Van de Ven has argued
upon, consistent with, fit, congruence, and that inadequate attention to specifying the form
coalignment, precise guidelines for translating of fit could fundamentally alter the meaning of
these verbal statements to the analytical level the theory itself (Van de Ven, 1979; Van de Ven &
are seldom provided. Consequently, research- Drazin, 1985).
ers choose an available or convenient mathe- This paper has been developed along the fol-
matical form and perform statistical tests without lowing premises: (a) fit is central to both theoret-
examining the validity of their choice (cf. Drazin ical discussions and empirical research in stra-
& Van de Ven, 1985; Schoonhoven, 1981, for ex- tegic management; (b) the role of fit in strategy
ceptions). research has been severely handicapped by the

423
absence of appropriate links between the con- employing them to identify six distinct perspec-
cept (verbalizing fit-based relationships) and tives of fit. Figure 1, a conceptual framework,
theory testing (statistical testing of such relation- uses these two dimensions to map fit as (a) mod-
ships); and (c) although the operationalization of eration, (b) mediation, (c) matching, (d) gestalts,
fit has been discussed generally (e.g., Drazin & (e) profile deviation, and (f) covariation. In the
Van de Ven, 1985; Joyce, Slocum, & Von Glinow, following section, each of these perspectives is
1982; Miller, 1982; Schoonhoven, 1981), this re- discussed independently regarding the central
search has not addressed strategy research in issue of concepts and testing, and subsequently,
particular or the many testing issues that are a set of issues pertaining to theory construction
central to linking concepts with empirical tests. is discussed to develop implications for future
research.
Alternative Perspectives of Fit Fit as Moderation
in Strategic Management Conceptualization. Because investigators
have followed the general axiom that no strat-
A Classificatory Framework
egy is universally superior, irrespective of the
When most researchers decide to use a spe- environmental or organizational context, they
cific concept of fit, they are faced with two fun- have commonly used a contingency perspective
damental decisions. The first is to choose the de- (Harrigan, 1983; Hofer, 1976; Ginsberg & Venka-
gree of specificity of the theoretical relation- traman, 1985) that has been operationalized
ship(s), which indicates the level of precision in within a moderation (interaction) perspective.
the functional form of fit. In some cases, a As noted by Schoonhoven: "When contingency
precise functional form of the relationship be- theorists assert that there is a relationship be-
tween the underlying variables can be speci- tween two variables . . . which predicts a third
fied, but in other cases, certain variables are variable, , . , they are stating that an interacfion
said to fit together, without describing a precise exists between the first two variables" (1981, p.
form. Typically, when two variables are in- 351, emphasis added), thus highlighting the
volved, it is possible to be more precise. For ex- popularity of the moderation perspective in or-
ample, Gupta and Govindarajan (1984) speci- ganizational research.
fied the fit between two variables—strategy and According to the moderation perspective, the
a managerial characteristic—in terms of inter- impact that a predictor variable has on a crite-
active effects, whereas Miller and Friesen (1984) rion variable is dependent on the level of a third
explored the differing patterns of configurations variable, termed here as the moderator. The fit
among a larger set of variables between suc- between the predictor and the moderator is the
cessful and unsuccessful groups of firms. primary determinant of the criterion variable.
The second decision is either to anchor the Researchers invoke this perspective when the
concept (and tests) of fit to a particular criterion underlying theory specifies that the impact of the
(e.g., effectiveness) or to adopt a criterion-free predictor (e.g., strategy) varies across the differ-
specification. In some settings, researchers ent levels of the moderator (e.g., environments).
specify fit that is intrinsically connected to spe- In more general terms, a moderator can be
cific criterion variables, but in other settings they viewed categorically (types of environment,
adopt a criterion-free specification, which has stages of product life cycle, organizational
universal applicability. types) or characteristically (degree of business-
These dimensions of specificity and anchoring relatedness, degree of competitive intensity),
are best applied to strategic management by and it will affect the direction or the strength of

424
LOW MANY
Fit as Profile Deviation Fit as Gestalts
(V) (IV)

CO

3 c
a o D O
"M

all Fit as Mediation Fit as Covariation II


(II) (VI) O 0)
Q) • S
(D - ^ O

PI ii

Fit as Moderation Fit as Matching


(I) (ID)
HIGH FEW
Criterion-specific Criterion-free

Choice of anchoring the specification


of fit-based relationships
Figure 1. A classificatory framework for mapping the six perspectives of fit in strategy research.

the relation between a predictor variable (e.g,, best understood when only two variables are
strategy) and a dependent variable (e.g., per- involved. Figure 2 is a schematic representation
formance). In a formal representation Z is a of this perspective.
moderator if the relationship between two (or Analytical Issues. Analytical issues are dis-
more) variables, say X and Y, is a function of the cussed here and in the following sections to
level of Z. The following is a mathematical rep- highlight the underlying link between conceptu-
resentation: alization and testing, and although they are
y = fix, z, X • z) (1)
where Y — performance, X = strategy, and Z = X {e.g.. Strategy)
the contextual variable that fits with strategy for
performance improvement; here X • Z reflects
the joint effect of X and Z.
Z (e.g., Context) (Performance)
At this stage, positioning this perspective
within the classificatory framework (Figure 1) is
more easily understood because it is clear that
this perspective derives its meaning through its X • Z (Interaction)
anchor with a particular criterion, Y. Further, as
is apparent later through a discussion of the an- Figure 2. A schematic representation of fit as
alytical issues, the theoretical meaning of fit is moderation.

425
termed analytical issues, they are best viewed statistic [Arnold, 1982, p. 152].) Investigators also
as issues that link the conceptual and statistical use complementary moderated regression anal-
domains. For the moderation perspective, four ysis (MRA) to test the form of moderation, and it
issues deserve particular attention: (a) the dis- is more commonly used in strategic manage-
tinction between form and strength of modera- ment research (e.g., Gupta & Govindarajan,
tion, (b) the role and impact of multicoUinearity, 1984, 1986; Hitt, Ireland, & Palia, 1982; Hitt, Ire-
(c) the comparison of main versus interaction ef- land, & Stadter, 1982; Prescott, 1986). MRA can
fects, and (d) the requirement of partialling out be represented by the following set of equations:
quadratic effects for testing the moderating ef- Y = QQ + a^X + ajZ + e (2)
fects. Y = ao + a^X + a^ -h agX • Z + e (3)
The distinction between the form and the
strength of moderation (Arnold, 1982) illustrates The moderation hypothesis is supported if the
both theoretical and analytical issues. If a re- unstandardized coefficient, a-^,, differs signifi-
searcher, using the moderation perspective, hy- cantly from zero, attesting to the effects of fit be-
pothesizes that the predictive ability of certain tween X and Z on performance, Y.
strategies differs across different environments, Because a particular data set may support
this hypothesis reflects the strength of modera- one effect and not the other (Arnold, 1982), it is
tion and can be tested using subgroup analysis. critical that researchers articulate their concep-
Conversely, if a researcher specifies that the tualization of moderation and justify their choice
performance outcome is jointly determined by of analytical technique (subgroup analysis vs
the interaction of the predictor and the modera- MRA) to ensure correspondence between their
tor, then this hypothesis reflects the form of mod- theory and their tests. Recognizing this distinc-
eration and can be tested using moderated re- tion, Prescott (1986) evaluated the nature of the
gression analysis (Arnold, 1982; Sharma, Du- moderating role of the environment and con-
rand, & Gur-Ane, 1981). Thus, even when using cluded that environments moderate the
the moderation perspective of fit, the investiga- strength, but not the form of the strategy-
tor must be more precise in developing the theo- performance relationship, which implies that it
retical position. is no longer adequate for researchers to theorize
The researcher addresses the strength of mod- a general form of fit-based relationship and to
eration by using subgroup analysis which typi- treat MFiA and subgroup analysis as substitut-
cally splits the sample into groups based on the able for testing the proposition.
contextual variable (Z). (In some cases, it may be The second issue pertains to the role and im-
more appropriate to split the sample on the basis pact of multicoUinearity, a statistical problem
of X to evaluate the moderating role of X on the that arises when correlations between indepen-
relationship between Z and y. In either case, the dent variables are extremely high, producing
analytical issues are the same, although the large standard errors of regression coefficients
theoretical meanings could differ.) The hypoth- and unstable coefficients. This is relevant for es-
esis of fit as reflected in the strength of modera- timating equation (3) because the cross-product
tion is supported when statistically significant term, X • Z, is likely to be strongly correlated
differences exist in the value of correlation coef- with X and Z. This issue is important because
ficients between strategy and performance some researchers explicitly invoke multicoUin-
across the environmental groups. (For two earity as a reason for not testing fit within this
groups, the difference in the correlation coeffi- perspective (cf. Dewar & Werbel, 1979; Drazin h
cients can be tested as a f-statistic [Bruning & Van de Ven, 1985).
Kintz, 1987, pp. 226-228], but for multiple Although statistical estimation problems are
groups, the difference is tested as a chi-squared posed by multicoUinearity, it is not problematic

426
for establishing the existence of moderation ef- the researchers can test for the existence of in-
fects. For interval-level measurements, a simple teraction effects but cannot compare the relative
transformation of the scale of origin reduces the influence of main and interaction effects be-
level of correlation between the cross-product cause the standardized coefficients are mean-
term, X • Z, and the original variables (X, Z). ingless. However, if it is possible to develop ratio
(Southwood, 1978; p. 1198, demonstrated that scales for X and Z, then their relative effects
any equation [X^, X2, X^ — X^ • X^] can be trans- can be compared (Allison, 1977; Friedrich, 1982;
formed into one [X[. X'^, X'3 = X[ • X^] where X\ Southwood, 1978). Thus, if the theory relies on a
= [Xi + c] and X'2 = [X2 + k], that is, the two comparison of main and interaction effects, the
equations express the same surface in the three moderation perspective may be limited to cases
dimensions [Y, Xj, X^], except for a shift in the that accommodate ratio levels of measure-
axes of X. However, if the variables are mea- ments.
sured using ratio scales, such a transformation The fourth issue focuses on partialling out the
alters the meanings of the measures and cannot quadratic effects of the original variables (X and
be so applied.) Such a transformation does not Z) in order to establish the presence or absence
affect the test for the unstandardized coefficient of multiplicative effects. Southwood (1978) ar-
03, although it affects the tests for both unstan- gued that in order to be sure that the relation-
dardized coefficients, a^ and a2, and all stan- ship is one of interaction rather than parabolic
dardized coefficients (cf. Southwood, 1978, pp. curvilinearity, it is necessary to control by using
1166-1167; 1199-1201). Thus, contrary to some the terms X, Z, X^, Z^ and X Z simultaneously.
investigators' beliefs in rejecting MRA for testing A test of the partial correlation between the de-
fit due to multicoUinearity problems (Dewar & pendent variable Y and X • Z, after partialling
Werbel, 1979; Drazin & Van de Ven, 1985), it is a out the effects of X, Z, X^ and Z^, provides sup-
valid analytical method for testing fit as moder- port for multiplicative interaction, whereas tests
ation (Allison, 1977; Arnold & Evans, 1979), of the partial correlation coefficients (a) between
which implies that if researchers conceptualize Y and X^, after partialling out the effects of X, Z,
their theories within the moderation perspec- X • Z and Z^ and (b) between Y and Z^, after
tive, they should not dismiss MRA for multicol- partialling out the effects of X, Z, X • Z and X^,
linearity reasons without evaluating this impor- provide evidence of curvilinearity. A review of
tant property for interval-level measurement. strategy and organizational research adopting
The third issue pertains to comparing main the moderation perspective indicates that extant
versus interaction effects within the moderation research has not incorporated such controls,
perspective. For example, a researcher may be which weakens the interpretations. It is hoped
interested in both the effects of a chosen busi- that future attempts at modeling fit as modera-
ness-unit strategy on performance and the ef- tion will rule out the rival explanation of plausi-
fects of the fit between strategy and administra- ble curvilinear effects of the original variables
tive characteristics, such as structure (Chandler, when establishing the existence of joint, multi-
1962), systems (Lorange & Vancil, 1978), or man- plicative effects.
agerial traits (Gupta, 1984). Such concepts in- Limitations. The moderation perspective of fit
clude the joint assessment of main and interac- is limited in at least two ways. One concerns the
tion effects, which cannot be accomplished at inability to separate the existence of fit from the
an interval-level of measurement. As discussed effects of fit. It is important for investigators to
previously, testing for a3 = 0 cannot be accom- remember that statistical tests within this frame-
plished along with tests for a^ and 02 because of work involve either the assessment of the signif-
the arbitrary scale of origin for interval-level icance of a^ in Equation 3, which is an indica-
measurement. Thus, by using a typical MRA, tion of the effect of fit (depicted by the interaction

427
term) on the criterion variable Y, or the differ- derlying theory should be questioned because
ence in the correlation coefficients, involving the the moderation view of fit (viewed in terms of
dependent variable Y across various groups multiplying individual strategy variables and
defined by either X or Z. Because specifying fit individual environmental variables) has little
as moderation refers to a criterion variable, its theoretical basis; hence, it is not surprising that
meaning is intricately tied to the specific perfor- Jauch and his colleagues (1980) did not find sig-
mance variable and may not be generalizable nificant multiplicative effects. Indeed, a careful
to other performance measures. Thus, if a re- review of the research literature arguing for en-
searcher is not interested in the performance ef- vironment-strategy coalignment indicates that
fects of fit, but is interested in the existence of fit the proponents of this view (e.g., Andrews, 1980;
in different subsamples, this perspective is inap- Bourgeois, 1980; Miles & Snow, 1978) invoked
propriate. the notion of fit rather metaphorically, and it is
The second limitation pertains to the difficulty up to the empirical researcher to translate such
involved in attaching theoretical meanings to a theoretical position into a moderation perspec-
the interaction terms, especially multiple sets of tive for analysis.
interactions and higher order interactions. The The problem inherent in higher order interac-
problem inherent in the multiple sets of interac- tions relates to the ambiguity of attaching any
tions can be illustrated as follows: If strategy is clear theoretical meaning to these terms. Al-
measured using a set of n variables and envi- though Allison (1977) has shown that three-way
ronment is captured through a set of m vari- interaction terms and higher order terms are not
ables, each variable reflects a specific theoreti- susceptible to multicoUinearity problems (for in-
cal content. Therefore, the moderating effects of terval-level measurements), there is an inherent
strategy and environment would be specified in theoretical limitation. Given that higher order
terms of a set of interaction variables (nXm). interactions often lack theoretical relevance,
However, this set collectively may not reflect the their use should be carefully justified by sub-
moderating effects because relationships that stantive rather than empirical reasons. Thus, re-
are implied by the set of individual interaction searchers should be particularly persuasive
components may not adequately represent the when attaching meaning to the higher order
nature of systemic interaction, implying an error terms.
of logical typing (Bateson, 1979; Van de Ven &
Drazin, 1985). Given these critical limitations, it is important
to recognize that as investigators should strive to
An example of this problem is considered in a avoid inappropriately linking theoretical and
study by Jauch, Osborn, and Glueck (1980), the empirical domains, they should view mod-
which examined the financial performance im- eration as one of many schemes to conceptual-
plications of the environment-strategy fit using a ize fit, and they should treat MRA as one of
multiplicative or moderation perspective. They many ways in which to test fit-based relation-
modeled fit as a set of 72 interaction terms, but ships. Toward this end, five other perspectives
none of the 72 possible interactions were statis- are discussed.
tically significant at thep < .05 level. Should this
be interpreted as a rejection of the performance Fit as Mediation
effects of environment-strategy fit (as they noted) Conceptualization. The mediation perspective
or should it more appropriately be understood specifies the existence of a significant interven-
as empirical results that did not support the hy- ing mechanism (e.g., organizational structure)
pothesis when tested using a moderation per- between an antecedent variable (e.g., strategy)
spective? If the latter is concluded, the appropri- and the consequent variable (e.g., perfor-
ateness of using this perspective to test the un- mance). Thus, while moderation specifies vary-

428
ing effects of an independent variable on a de- diation perspective, but through the latter they
pendent variable as a function of the moderat- are able to recognize the differences in the var-
ing variable, this perspective specifies the ious stages of a system of relationships. Al-
existence of intervening (indirect) effects be- though neither perspective is superior, the
tween an antecedent variable and its conse- choice should be predicated by theoretical con-
quent variable. siderations. For example, Hambrick, Mac-
The position of this perspective in Figure 1 Millan, and Day (1982) treated relative market
should be explained. Like moderation, this per- share as a moderator when analyzing effective-
spective is anchored with respect to a particular ness of strategies across the four cells of a busi-
criterion variable. However, the functional form ness portfolio matrix, an analysis that is consis-
of fit is, viewed simply as indirect effects, less tent with portfolio matrix theory. Alternatively,
precise than the moderation perspective in examining the nature of the relationship (di-
(strength, form, quadratic effects, etc.). More- rect vs spurious) between market share and
over, more than two variables can be incorpo- profit, Prescott, Kohli, and Venkatraman (1986)
rated within this perspective, thus reducing the treated market share as a mediator in a system
level of precision that can be reflected in speci- of relationships between strategy and profit
fying the functional form of fit. across different environmental contexts. Both
The mediator variable accounts for a signifi- studies addressed the fit that strategy-market-
cant proportion of the relation between the pre- share has on performance, using different theo-
dictor and criterion. Stated formally, Z is a me- retical and analytical perspectives.
diator of the probabilistic relation Y = f{X) if Z is Similarly, if the classical industrial organiza-
a probabilistic function of X (i.e., Z = f[X]) and Y tion economics paradigm (Scherer, 1980; Porter,
is a probabilistic function of Z (i.e., Y = f[Z]), 1981) of structure -^ conduct -^ performance is
where X, Z, and Y include different theoretical used to test the role that firm-level strategic ac-
content (Roseboom, 1956). Figure 3 is a sche- tions have in influencing the relationship be-
matic representation of fit as mediation involv- tween market-structure characteristics and firm
ing three variables, where Z (e.g., strategy) acts performance, both perspectives could be used
as a mediating mechanism (fit) between X {e.g., with differing theoretical meanings. The media-
market-structure characteristics) and Y (e.g., tion perspective decomposes the effects that
firm performance). This is supported by the fol- market-structure characteristics have on firm
lowing set of equations: performance into direct effects versus indirect ef-
fects. In contrast, the moderation perspective is
= QQ + ajZ -I- a2X + e (4)
useful to evaluate the differences in the relation-
X = bo + b^Z + e (5)
ships between strategy and performance across
Strategy researchers have embraced the various environments (Prescott, 1986).
moderation perspective more often than the me- Analytical Issues. Tests of fit as mediation usu-
ally are carried out within a path-analytic
framework (cf. Alwin & Hauser, 1972; Blalock,
X {e.g., Strategy) 1969; Duncan, 1972). Consistent with the pri-
mary objectives of this article, only issues that
link theory and statistics, namely the distinction
between complete versus partial mediation and
Z (e.g.. Context) Y (Performance) the test for the performance effects of fit, are dis-
cussed.
^ Figure 3. A schematic representation of fit as The first issue, the distinction between com-
' mediation. plete versus partial mediation, has important

429
theoretical implications for strategy research 1986, for details of estimation of these coefficients
that are best highlighted using a set of coeffi- within strategy research):
cients of the mediational model. For example, r = aj -I- a2 • bl + spurious effects (6)
this issue can address the dilemma of the rela-
tive importance of market-structure factors ver- where r^ is the observed zero-order correlation
sus strategic-choice decisions (Figure 3). If the between X and Y. (It is worthwhile to restrict the
coefficient a^ in Equation 4 is not statistically dif- discussion to cases in which this zero-order cor-
ferent from zero, the strongest support for the relation is large enough to make the decompo-
mediating effects of Z is obtained. This implies sition insightful.) The relative proportion of the
that the presence of Z is necessary for the trans- two effects—indirect (a2 ' bi) versus direct (aj)—
mission of effects of X on Y, and this is termed the provides an index of the relative effect of fit
complete mediational model. Thus, for testing (namely, indirect effects) versus the direct ef-
theoretical propositions rooted in the structure —> fects. More formal, statistical corroboration is
conduct—* performance paradigm, firm conduct provided through a test of the statistical signifi-
plays a critical role in translating market- cance of the indirect effects through the media-
structure opportunities into firm performance. tor. Because this involves testing a product of
Alternatively, if the coefficient a^ is not zero, a two regression coefficients, a standard f-test
direct effect between X and Y exists, and an in- cannot be adopted. Hence, an approximation
direct effect between X and Y through Z exists, based on Sobel (1988) follows:
implying a partial mediational model. This im-
plication means that partial effect is obtained
= (a, • b,)/V(bl +
due to market structure, and partial effect is ob- where a2 and bj are as defined earlier and se
tained due to strategic decisions because of the refers to the standard error of estimates.
market-structure characteristics. Such tests di- It is important to note that although the previ-
rectly address the theoretical position that firm ous discussion focused on a simple system of
performance is a function of both structural fac- relationships involving three variables {X, Z,
tors and strategic choice (e.g., Andrews, 1980; and Y), it can be conceptually and analytically
Bourgeois, 1980; Schendel & Hofer, 1979). Fi- extended to multiple variables (Duncan, 1972),
nally, if the direct effect, aj, is much stronger even though estimation and interpretation prob-
than the indirect effects, this implies that firm lems could be more pronounced. Thus, the fit as
conduct plays an insignificant role. mediation perspective is positioned differently
The second issue, the test for the performance from the moderation perspective in the classifi-
effects of fit, is important because most theorists catory scheme.
seek to specify the effects of performance that
are attributable to fit. Thus, the usefulness of this Fit as Matching
perspective depends on the availability of the
test statistic for the effects of fit. In other words, Concepfuaiizafion. This perspective is in-
although the test for a^ in Equation 3 provided voked for strategy concepts in which fit is a theo-
systematic statistical evidence for moderating retically defined match between two related
effects, a corresponding test for fit as mediation variables. This is a major point of departure
is needed. However, this issue is complicated from the previous two perspectives because fit is
because the impact of fit is given by the product specified without reference to a criterion vari-
of two path coefficients, b^ • ag. Based on the able, although, subsequently, its effect on a set
Simon-Blalock decompositional technique (Dun- of criterion variables could be examined. Stated
can, 1972), the following equation can be spec- differently, a measure of fit between two vari-
ified (see Prescott, Kohli, and Venkatraman, ables is developed independent of any perfor-

430
mance anchor, which is unlike the previous two clude (a) the potential unreliability of the fit mea-
perspectives. sure because the reliability of a difference score
An example of this perspective is found in IX - Zl is less than the average reliability of its
Chandler's (1962) classical thesis that a diversi- component parts {X and Z); (b) the possibility of
fication strategy requires a multidivisional struc- spurious association with an external vari-
ture, whereas a geographical expansion strat- able—even if the difference score has accept-
egy requires field units, and the absence of such able reliability, it may be spuriously related to
a match between structure and strategy leads to the criterion variable through the effects of the
administrative inefficiency or weaker perfor- original components {X and Z); and (c) the
mance. Thus, the measure of strategy-structure knowledge of generally weak discriminant va-
fit (0 or 1) can be derived based on the underly- lidity if the transformed variable is not differen-
ing theory without reference to performance. tiable from its component variables. Because re-
Such a theoretical proposition is most appropri- searchers who employ difference scores to mea-
ately operationalized within a matching per- sure fit within this perspective do not discuss
spective (Grinyer, Yasai-Ardekani, & Al-Bazzaz, their response to such issues, it is not possible to
1980; Rumelt, 1974) and tested with the external systematically assess the impact these scores
criterion of performance. Similarly, Chakravar- have on research results, but it is hoped that
thy (1987) stated that fit, within a planning- investigators will be more sensitive to the poten-
system context, existed when the planning sys- tial problems in future studies.
tem matches the required ideal system, and he Residual analysis reflects Dewar and Wer-
used this definition to test whether such a match bel's (1979) operationalization of fit. In this
enhances system performance. method, the residuals from the regression of one
Analytical Issues. The matching perspective variable (e.g., X) on the other (e.g., Z) are used
has been supported by three somewhat related to reflect fit, which subsequently can be related
analytical schemes: the deviation score analy- to the criterion variable (Y). According to this
sis, the residual analysis, and the analysis of perspective's proponents, the main advantages
variance (ANOVA). are that "one can simultaneously test both uni-
The deviation score analysis is based on a versalistic and contingency predictions . . .
premise that the absolute difference between [and] one can obtain some idea of the magni-
the standardized scores of two variables indi- tude of the different effects by comparing the
cates a lack of fit (e.g., Alexander & Randolph, coefficients" (p. 436). But, in arguing for this ap-
1985; Bourgeois, 1985). For instance, in the proach, these authors reject the multiplicative
three-variable system discussed earlier, IX — approach for multicoUinearity reasons, which is
Zl indicates the lack of fit between X and Z, and ill-founded. This analysis is appropriate to test
the performance implications of fit are tested by fit-based relationships if the underlying theory is
examining the impact that this variable has on conceptualized as deviation and if the limita-
performance. The formal specification of the tions of deviation scores are not serious. How-
equation follows: ever, it is inappropriate for researchers to con-
ceptualize fit as moderation and to reject MRA
Y = + + + - Zl) + e (7) and favor the residual analysis because they
If the coefficient a^ is statistically significant in are not interchangeable.
Equation 7, then a hypothesis of performance The residual approach does include limita-
effects of fit is supported. tions (Dewar & Werbel, 1979; Miller, 1982): (a) the
This analysis has intuitive appeal, but the problems of choosing an appropriate baseline
problems pertaining to the difference scores model to calculate the residuals; (b) the con-
(Johns, 1981) should be recognized. These in- founding of error variance in residuals, in addi-

431
tion to measurement error; (c) the arithmetic sign should be trying to find frequently recurring
of the residuals; and (d) the need to distinguish clusters of attributes or gestalts" (1981, p. 5).
between the residual of Z on Z and the residuals Along similar lines. Miller and Friesen noted:
of Z on X, although both reflect the fit between X "Archetypes appear to represent a set of rela-
and Z. At a minimum, researchers employing tionships which are in a temporary state of bal-
this analysis should discuss the relevance and ance. The . . . situations which are described
implications of these issues in their specific re- seem to form a number of gestalts. There is
search context. something holistic and ordered about the pat-
The third approach, the analysis of variance tern of . . . attributes" (1977, p. 264, emphasis
(ANOVA), typically is used to test for interaction added), and such a pattern could provide useful
effects, but Joyce, Slocum, and Von Glinow insights into a powerful concept of equifinality or
(1982) used it to develop useful tests of fit reflect- the feasible sets of internally consistent and
ing the matching perspective. They distin- equally effective configurations.
guished among three forms of fit—effect, gen- Instead of gestalts, if the system were decom-
eral, and functional—in a general congruency posed into a set of bivariate contingencies, it is
model that hypothesized interaction effects, but plausible that internal inconsistencies (mutually
they emphasized the similarity and matching conflicting directions) among multiple pairwise
levels of independent variables. A specific ad- contingencies would exist. As Child remarked:
vantage of their analytical scheme is that they "What happens when a configuration of differ-
could test competing perspectives of fit within a ent contingencies are found, each having dis-
common analytical framework (Joyce et al., tinctive implications for organizational design?"
1982). (1975, p. 175). A partial solution to this problem
Thus far, these three perspectives share one may be the development of gestalts, which log-
common feature, namely, they are appropriate ically extends the bivariate fit perspective
for specifying bivariate fit, fit specified in various through a multitiered taxonomical approach
functional forms between two variables, al- (Hambrick, 1984; Miller & Friesen, 1984). The po-
though the second perspective can accommo- sition of the fit-as-gestalts perspective within the
date a larger system of relationships. In contrast, classificatory framework (Figure 1) is deter-
the following perspectives are appropriate for mined by its being criterion-free and minimally
simultaneously specifying and testing fit among precise.
a larger set of variables. Analytical Issues. There are two major ana-
Fit as Gestalts lytical issues regarding fit as gestalts, the de-
scriptive validity of the gestalts and the predic-
Conceptualization. When fit is conceptualized tive validity of the gestalts.
and specified using two variables, it is possible Descriptive validity requires that the gestalts
for investigators to invoke alternate perspectives be interpretable in terms of the theoretical posi-
that have precise functional forms, but when tions implied by fit. This is especially important
many variables are used, the degree of preci- because most analytical schemes that are avail-
sion must be relaxed. One such multivariate able for developing gestalts (e.g., cluster anal-
perspective is the identification of gestalts, ysis, g-factor analysis) are inductive. (For a com-
which is defined m terms of the degree of inter- prehensive discussion on alternate analytical
nal coherence among a set of theoretical at- approaches to gestalt development, see Miller
tributes. and Friesen, 1984.) Although Miller (1981) ar-
The role of gestalts has been best described by gued that gestalts are "relatively few and very
Miller: "Instead of looking at a few variables or different from one another, both in terms of the
at linear associations among such variables we scores of, and relationships among, variables,"

432
(p. 1) it is important to develop a set of formal types or multiple configurations of equally suc-
criteria to judge the descriptive validity of ge- cessful strategies. For instance, Etzioni argued
stalts. These criteria should include (a) testing that "congruent [organizational] types are more
the number of gestalts using formal statistical effective than incongruent types" (1961, p. 14,
procedures like the C-H index (Milligan & Coo- emphasis added), and Child reported that high-
per, 1985), (b) demonstrating cluster stability us- performing organizations had internally consis-
ing cross-validation of results with holdout sam- tent structural configurations, which were ab-
ples, and (c) describing the gestalts based on the sent in poor-performing organizations. If such
theory that guided the selection of input vari- propositions can be directly tested within a ge-
ables for analysis. stalt perspective, the challenge is to incorporate
Although the first two criteria address statisti- performance variable(s) into the analysis. One
cal concerns, the last one is important for theo- approach for assessing predictive validity is to
rizing because there is a subtle but important follow Hambrick's suggestion of identifying sub-
distinction between fit as gestalts and strat- samples of high- and low-performing busi-
egy taxonomies. The latter represents an empir- nesses to identify profiles of fit within each sub-
ical identification of naturally occurring strategy sample. This approach would help a researcher
types (Hambrick, 1983b; Miller & Friesen, 1984) to develop distinct profiles of fit across the per-
that does not use internal congruence, except formance categories, and to assess the possibil-
indirectly to name the gestalts. The selection of ity of discovering patterns of equifinality within
the underlying variables for taxonomic inquiry low- and high-performing businesses. Such an
is guided by the need to balance parsimony and approach enabled Hambrick (1983a, 1983b) and
exhaustiveness of coverage. In contrast, speci- Miller and Friesen (1978) to isolate generic suc-
fying the testing of fit as gestalts requires a care- cessful and unsuccessful gestalts in an explor-
ful enumeration of the underlying dimensions so atory fashion. However, because concepts of
that the gestalts can be ordered along their lev- equifinality and generic strategies rest on the
els of fit or congruence. For example, in Miles interplay between sound theoretical and statis-
and Snow's (1978) theory of strategic adaptation tical criteria, more work is needed to develop the
based on the notion of internal consistencies link between theory and data analysis.
among three problem domains (entrepreneur-
Fit as Profile Deviation
ial, engineering, and administrative), the pat-
tern of coalignment results in four strategy types Conceptualization. In the profile-deviation
(prospectors, defenders, analyzers, and reac- perspective, fit is the degree of adherence to an
tors). A rigorous test of this theory requires using externally specified profile, and it is akin to Van
the three problem domains as a basis to develop de Ven and Drazin's (1985) usage of pattern
four gestalts that can be interpreted in the light analysis. The role and use of this perspective
of their theory. This ensures a tighter link be- can be best introduced through the following
tween the underlying theory, the choice of di- case. If an ideal strategy profile (e.g., the level of
mensions, and the interpretations of the clusters resource deployments along a set of strategy di-
in light of the theory. Thus, although Miles mensions) is specified for a particular environ-
and Snow's (1978) typology is popular, unfortu- ment, a business unit's degree of adherence to
nately its use in empirical research studies has such a multidimensional profile will be posi-
not yet been explicitly predicated on the pattern tively related to performance if it has a high
of fit among the three dimensions. level of environment-strategy coalignment.
Predictive validity is important for establish- Conversely, deviation from this profile implies a
ing the performance implications of fit and for weakness in environment-strategy coalignment,
demonstrating the existence of generic strategy resulting in a negative effect on performance.

433
Figure 4, a schematic representation of this per- a sample of PIMS-based businesses across eight
spective, uses six illustrative dimensions of strat- environments over two different time periods.
egy. (X3 refers to the values in the study sample, They found strong, consistent support for their
and X^ refers to the calibration sample.) hypotheses because deviations in resource allo-
This perspective is different from the gestalt cation patterns from corresponding environ-
perspective because the referent profile is an- ment-specific profiles were strongly (negatively)
chored to a specific criterion (typically, perfor- related to performance. This perspective also
mance), and thus it is positioned differently in can be logically extended to other research is-
the classificatory framework. Therefore, in con- sues in which ideal profiles are specified along a
trast to an exploratory development of multivari- set of theoretically related variables.
ate profiles of coherence, this perspective allows Analytical Issues. The interplay of critical is-
a researcher to specify an ideal profile and to sues IS important in this conceptualization of fit
demonstrate that adherence to such a profile for (a) developing the ideal profile, (b) adding
has systematic implications for effectiveness. Al- differential weights for the multiple dimensions,
though this perspective can be adopted in a va- and (c) using a baseline model to assess the
riety of research situations, it is particularly use- power of the test.
ful for testing the effects of environment-strategy The first issue focuses on developing a profile
coalignment (Andrews, 1980; Bourgeois, 1980; that can serve as a benchmark for calibrating
Miles & Snow, 1978) because deviations in strat- the strategies of the businesses in a study. Re-
egy from an environment-specific ideal profile garding this issue, there a r e two obvious
should be negatively related to performance. choices. The first choice is theoretical, that is,
Venkatraman and Prescott (in press) argued that specifying the profile along a set of dimensions
this perspective best reflects the theoretical that are appropriate for a particular environ-
proposition of performance effects of environ- ment. Although this choice is intuitively appeal-
ment-strategy coalignment and employed it on ing, the operational task of developing such a

Strategy Standardized Scale for


Dimensions Importance Measuring the Dimensions
-1 0 +1

X2

X,

X,

Figure 4. A schematic representation of fit as profile deviation.

434
profile with numerical scores along a set of di- sional scaling (e.g., Carroll & Chang, 1967) for
mensions is difficult. (For example. Porter, 1980, assessing the effects of weighted versus un-
provided a conceptual base for identifying the weighted distances in the measurement of fit
ideal strategies for different generic environ- can be adapted.
ments, but it is not possible to translate his ver- The third issue pertains to the power of the
bal statements into numerical scores of ideal re- test, which requires the researcher to specify a
source deployments for each environment.) baseline model to demonstrate that the predic-
The second option is empirical, that is, devel- tive power of the measure of coalignment (cal-
oping this profile (cf. Drazin & Van de Ven, 1985; culated as a weighted euclidean distance using
Venkatraman & Prescott, in press) using a cali- the profile of the calibration sample) is signifi-
bration sample. In a strategy research context, cantly better than a measure calculated as de-
Venkatraman and Prescott (in press) developed viation from a random profile. This nontrivial is-
their profile of ideal strategic resource deploy- sue is best explained through an analogy to the
ments using the mean scores of a calibration discriminant analysis, where the power of the
sample—defined as the businesses that lie in discriminant function is demonstrated by its
the top 10 percent on the performance scale ability to discriminate among certain groups
within an environment—that subsequently was that use a set of discriminating variables. For
not used for the study sample. Because it is im- this purpose, the classificatory accuracy of the
portant to arrive at a valid ideal profile, it is sig- model is compared against a chance or naive
nificant for investigators to use a hold-out sam- model (Morrison, 1969). In Van de Ven and
ple, jackknifing for different proportions of the Drazin's (1985) pattern analysis, a negative and
sample and subsample replications, in order to significant correlation coefficient between the
increase the robustness of the results. coalignment measure and the criterion demon-
The second issue concerns developing a mul- strates the performance effects of fit. Because the
tidimensional profile by using equal weights for power of their test is unknown, it weakens the
the dimensions or by differentially weighting the interpretive ability of the empirical result. How-
dimensions, based on their relative importance ever, more powerful tests that discount rival hy-
to the context. Drazin and Van de Ven (1985) potheses (e.g., deviations from any random pro-
adopted the unweighted scheme but noted that file would exhibit an equally strong and signifi-
equal weights "can be relaxed by introducing cant negative correlation coefficient) are needed
the possibility of differentially weighting the im- to empirically test this perspective in strategy re-
portance of deviation . . . in determining search.
performance" (p. 351). For strategy research, as-
Fit as Covariation
suming equal weights for the underlying strat-
egy dimensions is untenable because an effec- Conceptualization. According to this perspec-
tive package of strategic resource deployments tive, fit is a pattern of covariation or internal con-
should reflect differential emphasis, which de- sistency among a set of underlying theoretically
pends on the importance of a particular dimen- related variables, and it can be best described
sion to the particular environment. Adopting through an illustration. If a business-unit strat-
such a view, Venkatraman and Prescott (in egy in a particular context is best represented as
press) derived the weighting scheme from the the pattern of consistent, concurrent resource al-
beta weights of the regression equations using locations to the areas of research and develop-
the array of strategy variables on performance ment, design, manufacturing, and marketing,
for each environment in the calibration sample. any one area is insufficient for an effective strat-
Alternatively, where feasible, preference- egy, which requires consistent attention to all
mapping procedures that include multidimen- areas. For such a theoretical position, fit must be

435
specified among the four dimensions, and it is The first issue, distinguishing between the ex-
most appropriately captured as the pattern of ploratory and confirmatory approach, is an age-
covariation. old social science research dilemma. Opera-
Fit as covariation is similar to megastrategy tionalizing fit as covariation is based on the ba-
(Mintzberg, 1978) or strategy as a pattern or sic principles of factor a n a l y s i s , that is,
stream of major and minor decisions (Grant & explaining covariation among a set of indicators
King, 1982; Miles & Snow, 1978). Both megastrat- in terms of a smaller set of factors (first-order
egy and a pattern of decisions are best repre- factors) and explaining the covariation among
sented as covariation among the constituent di- the first-order factors in terms of second-order
mensions, rather than as any of the earlier per- factor(s). More specifically, fit as covariation is
spectives (with the possible exception of gestalts specified as a second-order factor, and the first-
because this perspective also seeks to identify order factors represent the dimensions to be
similar units based on the scores along a set of coaligned (Venkatraman, 1986).
chosen variables and makes no claim regarding Covariation can be modeled in two ways: as
the pattern of internal consistency among the exploratory factor analysis (EFA) or as confirma-
variables). Even though these perspectives may tory factor analysis (CFA). Specifying a second-
appear similiar, gestalts are viewed as products order factor within EFA involves using a com-
of cluster analysis (grouping of observations mon factor analysis with oblique rotations and
based on a set of attributes), whereas covaria- correlations among the first-order factors to ex-
tion is modeled as factor analysis (grouping of amine the possibility of second-order factors (cf.
attributes based on data from a set of observa- Cattell, 1978, Chapter 9, for details). In contrast,
tions). Therefore, the position of fit as covariation CFA (Joreskog & Sorbom, 1979) is used to test
within the classificatory framework (Figure 1) one or more theoretically specified second-order
differs from position of the perspective of fit as factor(s) against the observed data. Thus, unlike
gestalts only in relation to the degree of specifi- EFA in which predefined mathematical criteria
cation of the functional form. This perspective are used to determine the optimal rotation that
requires a much greater precision in the pattern may result in uninterpretable theoretical mean-
of logical consistency among the factors and the ing, CFA provides a basis to explicitly evaluate
explication of the underlying logical link among a second-order factor model against an alterna-
the attributes. For this perspective, for example, tive model (one that may, for example, specify
general linear models like regression analysis only intercorrelated first-order factors).
have limited use because they miss "the concept
of a central thread or internal logic underlying a When the relative benefits of both types of fac-
strategy, . . . [and the] regression coefficient tor analysis are weighed (cf. Bagozzi, 1980;
may have statistical significance, but may indi- Joreskog & Sorbom, 1979), the confirmatory ap-
cate no apparent logical linkage among the var- proach is preferred for modeling fit as covaria-
ious independent variables" (Hambrick, 1980, p. tion. Figure 5, a schematic representation of this
571, emphasis added). Thus, it can be a major perspective, pits two models against each other.
challenge for the researcher to develop models The left part of the model specifies the direct ef-
that best reflect the degree of covariation among fects that four dimensions have on a criterion,
a set of attributes. which is ineffectual for providing any insights
into the pattern of covariation among the four
Analytical Issues. There are two critical issues dimensions. In contrast, the right side of the fig-
regarding specifying and testing fit as covaria- ure explicitly specifies the covariation among
tion: using the exploratory versus confirmatory the four dimensions as reflecting an internally
approach to specifying fit and testing the impact consistent business strategy, which in tum has
that performance has on fit. an effect on the criterion. The coalignment

436
among them is formally specified as an unob- As described by Joreskog and Sorbom (1979),
servable theoretical construct at a higher plane if these two models are nested (one is a subset of
than the individual functional dimensions. the other), it is possible to use the difference in a
There are no directly observable indicators for chi-squared statistic (which follows a chi-
this construct; its meaning is derived through di- squared distribution) for assessing model supe-
rectly operationalized first-order factors that in- riority. If the two parts of Figure 5 are not nested,
clude observable indicators reflecting the re- a chi-square difference test criterion (Joreskog,
source allocation patterns to the functions (e.g., 1971) cannot be employed. Thus, a preliminary
indicators of functional strategy). In this specific indication of the model superiority can be pro-
example, internal consistency is formally repre- vided by comparing the coefficients of determi-
sented in the model, and its effect on perfor- nation of the two models (analogous to R^). A
mance can be directly assessed. more formal assessment requires the calculation
The second issue, pertaining to testing the im- of a target coefficient {T). In keeping with Marsh
pact that performance has on fit modeled as co- and Hocevar (1985), a second-order factor model
variation can be described in terms of three is merely a parsimonious explanation of the co-
complementary test statistics: a comparison of variation among the first-order factors. Conse-
the coefficients of determination, the calculation quently, even if the second-order factor model
of the target coefficient T (Marsh & Hocevar, effectively explains the covariation among the
1985), and the statistical significance of the sec- first-order factors, the goodness of fit can never
ond-order factor loadings (Venkatraman, 1986). exceed that of the first-order factor model. Marsh

Main effects moder Covariation model

° Correlations among Xs are not drawn for


schematic clarity
Figure 5. A schematic representation of fit as covariation.

437
Table 1
Comparing Alternate Perspectives oi tbe Concept of Fit in Strategy Research

Perspectives
Key Characteristics Fit as Moderation Fit as Mediation Fit as Matching

Underlying conceptualization Interaction Intervention Matching


of fit
Verbalization of a strategy The interactive effects of Market share is a key The match between strategy
proposition strategy and managerial interverung variable and structure enhances
characteristics have between strategy administrative efficiency
implications for performance and performance
Number of vanables m the Two Two to multiple Two
specification of fit
Analytical schemeCs) for Analysis of variance Path-analysis ANOVA
testing fit Moderated regression analysis Deviation scores
Subgroup analysis Residual analysis
Measure of fit Statistical derivation Statistical derivation Interval-level measure
Illustrative references Gupta & Govindarajan (1984) Prescott, Kohh, & Chandler (1962)
Prescott (1986) Venkatraman (1986) Bourgeois (1985)
Joyce, Slocum, & Von
Glinow (1982)
Dewar & Werbel (1979)
Required for model identification purposes.

and Hocevar (1985) proposed using a Target co- Beyond a General Metaphor of Verbal
efficient (T)—the ratio of the chi-square of the and Statistical Correspondence
first-order model to the chi-square of the second-
order model, which has an upper bound of 1.00. In Table 1, the six perspectives of fit described
A value close to 1.00 will support using the co- in this article are compared along several key
variation model versus the main-effects model, characteristics to highlight this fundamental dif-
and this model can be interpreted in a manner ference. From this table, it is clear that the con-
similar to Bentler and Bonett's (1980) delta index, cept of fit is more complex than a general met-
until the distributional properties of T are estab- aphor requiring aligning distinct theoretical
lished. The third criterion is to assess the statisti- meanings with analytical requirements.
cal significance of the four loadings {a^ through Theorists in strategic management must move
04) on the second-order factor, termed as from treating fit as a general metaphor that has
coalignment. If the four first-order factors consis- universal applicability. Researchers in general
tently contribute to the second-order factor, all do not provide guidance for operationalizing
four coefficients are expected to be significant. theories of fit (cf. Bourgeois, 1980; Chakravarthy,
If these three criteria are satisfied, fit as covari- 1987; Jauch & Osborn, 1981; Snow & Miles, 1983).
ation is the preferred specification, and the per- Words such as congruence, fit, and alignment
formance effect of fit is given by the magnitude should be accompanied by descriptive guide-
and significance of coefficient a^. lines that, at minimum, specify specific func-

438
Table 1—Continued
Comparing Alternate Perspectives of the Concept of Fit in Strategy Research

Perspectives
Fit as Gestalts Fit as Profile Deviation Fit as Covariation
Intemal congruence Adherence to a specified profile Internal consistency

The nature of intemal congruence The degree of adherence to a The degree of intemal consistency in
among a set of strategic variables specified profile has significant resource allocations has significant
differs across "high" and "low" effect on performance effect on performance
performance businesses
Multiple Multiple Four'^ to multiple

Numerical taxonomical The calculation of deviation as a Second-order factor analysis


methods—cluster analysis, factor eucledian distance in an (confirmatory)
analysis n-dimensional space MDS
Ordinal/interval measure Interval measure Interval measure
Miller and Friesen (1984) Van de Ven & Drazin (1985) Venkatraman (1986)
Hambrick (1984) Venkatraman & Prescott (in press) Venkatraman & Walker (1989)

tional form(s). This classificatory scheme, based fit may now be more appropriate, but as the
on six perspectives of fit described in this article research stream matures, using confirmatory
should serve as a common reference point for perspectives would be more appropriate.
discussions, and it is hoped that future work will Explicit Justification of the Specification of Fit.
exhibit increased sensitivity to the widely differ- Researchers must explicitly justify their specifi-
ing theoretical and empirical meanings of fit. cation of fit within a particular research context.
They should abandon random and convenient
Choosing Among the Six Perspectives statistical methods, recognizing that research
Is any perspective better than the others? Al- results are sensitive to this choice. Just as strat-
though it is tempting to rank the appropriateness egy researchers routinely demonstrate the link
of these perspectives for different theoretical between defining their concept and operation-
questions, this may be a futile exercise because alizing their measures (Venkatraman & Grant,
of the intrinsic nature of the links among theory, 1986), they also should justify the specification of
method, and data. The same theoretical ques- fit between or among the constructs.
tion can be used for more than one perspective, For example, Gupta and Govindarajan (1984)
which would be useful for cumulative theory developed a hypothesis that reflected the per-
building. Moreover, for a particular research spective of fit as moderation; however, fit could
stream, using exploratory perspectives that are have been operationalized through a matching
less precise in specifying the functional form of perspective or as a mediational perspective.

439
Their empirical results would have been richer if cases in which the extant research may not be
they had compared the moderation perspective powerful enough to suggest rejecting the theo-
against the alternative perspectives. Because ries outright. For example, in situations involv-
this process was not followed, the empirical ing the fit between only two concepts, only mod-
demonstration of the impact that the strategy- eration, mediation, and matching could be
manager fit has on performance may not be considered. Although not a generalizable state-
generalizable beyond the moderation perspec- ment, these three could be considered as com-
tive. Similarly, are the results of Bourgeois' (1985) peting perspectives for most research questions.
study of the difference-score approach of the Unfortunately, this issue has been left open for
mediation perspective generalizable? Also, interpretation according to strategic manage-
would Chakravarthy's (1987) results on the ab- ment literature. Researchers have invoked theo-
sence of performance effects of tailoring strate- retical perspectives that are relevant to moder-
gic planning systems to the context hold up if ation, mediation, and matching in very general
they were tested within the perspective of profile terms, but have specified a functional form of fit
deviation? that is appropriate for only one perspective.
Such questions stem from the triangulation Since this article has shown that moderation,
concept, which advocates testing theoretical re- mediation, and matching are theoretically and
lationships by using multiple measures and empirically different and it has advocated that fit
multiple methods (Denzin, 1978; Jick, 1979), and as gestalts and fit as profile deviation can be
they are relevant because the hypothesis of the used as complementary perspectives when mul-
effects of performance on structure-context fit tiple variables are involved, identifying relevant
was supported using some perspectives and not competing models in a particular study could
others (cf. Drazin and Van de Ven, 1985; Joyce et lead to more useful and powerful operational-
al., 1982). Also, the theory of the impact that per- ization, data collection, data analysis, and inter-
formance has on the environment-strategy pretation.
coalignment was supported using subgroup Multiple Statistical Tests Within a Data Set: A
analysis, reflecting the strength of moderation "Triangulation-Trap"? Do researchers have the
(Prescott, 1986), for fit as profile deviation (Ven- option of using multiple tests of fit within a single
katraman & Prescott, in press), but not within a empirical study? When Tosi and Slocum (1984)
moderated regression perspective (Jauch et al suggested guidelines for contingency theory re-
1980). search, they noted that "statistical techniques
have frustrated researchers' attempts to test for
Role of Multiple Perspectives the interaction effects being modeled because
each technique has implied biases. Researchers
need to compare the utility of each of these sta-
For the investigator, a critical issue in study- tistical techniques using the same data set" (p.
ing strategic management is the benefit of using 16, emphasis added). Van de Ven and Drazin
multiple perspectives. This general theme is dis- (1985) made a similar recommendation: "Studies
cussed in two ways, the use of multiple concepts should be designed to permit comparative eval-
and the use of multiple statistical tests within a uation of as many forms of fit as possible. . . .
data set. Examining multiple approaches to fit in contin-
Multiple Concepts of Fit. Which perspective is gency studies and relating the findings to
most appropriate for a given research question? unique sample characteristics can greatly aid
One way to answer this question is for the re- the development of mid-range theories of what
searcher to consider multiple specifications as approach to fit applies where" (pp. 358-360).
competing theories or models, especially in Specifically, by using multiple specifications

440
they uncovered insightful nuances of the struc- researchers should be cautious about triangula-
tural contingency theory that otherwise may not tion within a data set, after the fact.
have been possible. A Monte Carlo simulation can answer this
Philosophically, this recommendation is fine, question by accommodating varying patterns of
but, before researchers use multiple perspec- correlations among X, Y, and Z and by perform-
tives within a single study, they should evaluate ing tests of convergence. Because the probabil-
both the appropriateness of each perspective ity of convergence across the perspectives of
and the design for collecting data because not moderation, matching, and mediation is close to
all theoretical statements are amenable to the zero, recommendations employing after-the-fact
six perspectives, and measurement schemes attempts (Tosi & Slocum, 1984) to assess triangu-
may limit the use of some perspectives. More lation are of limited use. Indeed, if a random
important, the nonconvergence of results across data set is used, the probability of finding sup-
multiple statistical tests can be ambiguous. If re- port for any one perspective is quite high.
sults converge across multiple perspectives, ev- Thus, multiple perspectives can be useful
idence of robustness is provided; however, the when researchers present theoretically argued
converse is not true. For example, for one data competing models, but not when the data are
set, Drazin and Van de Ven (1985) reported that subjected after the fact to multiple statistical tests
their theory was supported using a systems ap- that explore the extent of support for different
proach, but it was not using an interaction ap- perspectives.
proach. Do their results imply differential sup-
port for the competing theoretical perspectives, Longitudinal Operationalizations of Fit
or should the similarities among these perspec-
tives be evaluated using multiple tests within the The perspectives discussed in this article have
same data set? focused on static, cross-sectional approaches for
Additionally, for a random data set, what is specifying and testing fit within strategy re-
the probability of researchers' accepting a theo- search. However, coalignment is a dynamic
retical relationship of the impact that perfor- and never-ending task, whereby the organiza-
mance has on fit across a subset of comparable tion is continually "shooting at a moving target
perspectives? For instance, suppose that three of coalignment" (Thompson, 1967, p. 234).
different researchers are seeking to test the Therefore, no organizational system is in a state
same underlying theory of the impact that per- of perfect dynamic coalignment, but every orga-
formance has on fit between two independent nization is moving toward this state. Several in-
variables X and Z by using the perspective of teresting theoretical issues are based on this dy-
moderation, matching, and mediation. Further, namic perspective of fit, such as the forces that
suppose they independently obtain a similar explain organizational movement toward ver-
pattern of correlations among the three vari- sus away from equilibrium (fit), the short-term
ables, X, Y, and Z. What is the probability that benefits of fit versus the long-term benefits of fit,
their results would converge, irrespective of the the possible trade-offs between short-run fit ver-
specific values in the correlation matrices? An- sus long-term fit. Although these are interesting
swers to this question explicate the fundamental theoretical issues, it is unclear whether the six
nature of the relationships between and among perspectives identified here are appropriate for
the three perspectives that are independent of testing them. Even though this article does not
the specific theory being tested, and they are cover the issue of dynamism, it is recognized as
useful for pointing out the appropriateness of a promising area of research for the develop-
pursuing multiple tests of fit. Until it is shown that ment of appropriate mechanisms to specify and
there is no a priori preference (bias) of one test. test fit within a longitudinal perspective.

441
Summary schemes wherever feasible. Toward this end,
six concepts of fit within strategic management
Fit, as a general concept, is central to the have been discussed. For each perspective,
study of strategic management. Yet, theory in available statistical schemes highlighted the iso-
this area must be extended to clarify the specific morphic nature of the link between the theory
meaning of fit as used in various settings, to de- and the testing scheme, and it was demon-
velop strong links between verbalizations and strated that the statistical methods are not freely
statistical testing schemes, and to test relation- interchangeable without confounding the un-
ships by using multiple comparable statistical derlying theoretical meaning of fit.

References
Aldrich, H. E. (1979) Oiganization and environments. Engle- Bruning, J. L., & Kintz, B. L. (1987) Computational handbook
wood, NJ: Prentice-Hall. of statistics. Glenview, IL: Scott, Foresman.
Alexander, ]. W., & Randolph, W. A. (1985) The fit between Carroll, J. D., & Chang, J. J. (1967) fleiatog preference data
technology and structure as a predictor of performance in to multidimensional scaling solutions via a generalization
nursing subunits. Academy of Management Journal, 28, of Coombs' unfolding model. (Available from Bell Labs,
844-859. New Jersey)
Allison, P. D. (1977) Testing for interaction in multiple regres- Cattell R. B. (1978) The logic and use of factor analysis. San
sion. Ameiican Journal of Sociology, 83, 144-153. Francisco: Jossey-Bass.
Alwin, D. F., & Hauser, R. M. (1972) The decomposition of Chakravarthy, B. S. (1987) On tailoring a strategic planning
effects in path analysis. American Sociological Review, system to its context: Some empirical e-vidence. Strategic
40, 37-47. Management Journal, 8, 517-534.
Andrews, K. R. (1980) The concept oi corporate strategy. Chandler, A. D. (1962) Strategy and structure. Cambridge,
New York: Dow Jones-Irwin. MA: MIT Press.
Arnold, H. (1982) Moderator variables: A clarification on Child, J. (1975) Managenal and organization factors associ-
conceptual, analytic and psychometric issues. Organiza- ated with company performance—Part H, A contingency
tional Behavior and Human Performance. 29, 143-174. analysis./ournai o/Managemenf Sfudies, 12, 12-27.
Amold, H. J., & Evans, M. G. (1979) Testing multiplicative Denzin, N. K. (1978) The research act. New York: McGraw-
models does not require ratio scales. Organizational Be- Hill.
havior and Human Performance, 24, 41-59.
Dewar, R., & Werbel, J. (1979) Universalistic and contin-
Bagozzi, R. P. (1980) Causal models in marketing. New York- gency predictions of employee satisfaction and conflict.
Wiley. Administrative Science Quarterly, 24, 426-448.
Bateson, G. (1979) Mind and nature. New York: Dutton.
Drazin, R., & Van de Ven, A. H. (1985) An examination of
Bentler, P. M., & Bonett, D. G. (1980) Significance tests and alternative forms of fit in contingency theory. Administra-
goodness of fit in the analysis of covariance structures. tive Science Quarterly, 30, 514-539.
Psychological Bulletin, 88, 588-606.
Duncan, O. D. (1972) Path analysis: Sociological examples
Blalock, H. M., Jr. (1965) Theory building and the statistical (plus addenda). In H. M. Blalock (Ed.), Causal models in
concept of interaction. American Sociological Review 30 sociai science (pp. 115-138). Chicago: Aldine-Atherson.
374-380.
Egelhoff, W. G. (1982) Strategy and structure in multina-
Blalock, H. M., Jr. (1969) Theory construction: From verbal to
tional corporations: An information processing approach.
mathematical formulations. Englewood Cliffs NJ- Pren-
tice-Hall. Administrative Science Quarterly, 27, 435-458.
Bourgeois, L. J., m (1980) Strategy and environment: A con- Etzioni, A. (1961) A comparative analysis of complex orga-
ceptual integration. Academy of Management Review 5 nizations. New York: Free Press.
25-39. ' ' Friedrich, R. J. (1982) In defense of multiplicative terms in
Bourgeois, L. J., HI (1985) Strategic goals, perceived uncer- multiple regression equations. American Journal of Poht-
tainty, and economic performance in volatile environ- ical Science, 26, 797-833.
ments. Academy of Management Journal, 28, 548-573. Fry, L. W., & Smith, D. A. (1987) Congruence, contingency.
442
and theory building. Academy of Management Review, Hofer, C. W. (1975) Toward a contingency theory of business
12, 117-132. strategy. Academy of iVfanagemenf/ournai, 18, 784-810.
Galbraith, J. R., & Nathanson, D. (1979) The role of organi- Hofer, C. W., & Schendel, D. E. (1978) Strategy formulation:
zational structure and process in strategy implementation. Analytical concepts. St. Paul, MN: West.
In D. Schendel & C. W. Hofer (Eds.), Strategic manage- Jauch, L. R., & Osborn, R. N. (1981) Toward an integrated
ment: A new view of business policy and planning (pp. theory of strategy. Academy of Management Review, 6,
249-283). Boston: Little, Brown. 491-498.
Ginsberg, A., & Venkatraman, N. (1985) Contingency per- Jauch, L. R., Osborn, R. N., & Glueck, W. F. (1980) Short-
spectives on organizational strategy: A critical review of term financial success in large business organizations:
the empirical research. Academy of Management Re- The environment-strategy connection. Strategic Manage-
view, 5, 25-39. ment Journal, 1, 49-63.
Grant, J. H., & King, W. R. (1982) The logic of strategic plan- Jick, T. D. (1979) Mixing qualitative and quantitative meth-
ning. Boston: Little, Brown. ods: Triangulation in action. Administrative Science Quar-
Grinyer, P. H., Yasai-Ardekani, M., & Al-Bazzaz, S. (1980) terly, 24, 602-611.
Strategy, structure, the en-vironment, and financial perfor- Johns, G. (1981) Difference score measures of organizational
mance in 48 U.K. companies. Academy of Management behavior variables: A critique. Organizational Behavior
Journal, 23, 193-220. and Human Performance, 27, 443-463.
Gupta, A. K. (1984) Contingency linkages between strategy Joreskog, K. G. (1971) Statistical analysis of congenenc tests.
and general manager characteristics: A conceptual ex- Psychometrika, 36, 109-133.
amination. Academy of Management Review, 9, 399-412.
Joreskog, K. G., & Sorbom, D. (Eds.) (1979) Advances m fac-
Gupta, A. K., & Go-vindarajan, V. (1984) Business unit strat- tor analysis and structural equation models. Cambridge,
egy, managerial characteristics, and business unit effec- MA: Abt Books.
tiveness at strategy implementation. Academy of Man-
agement Journal, 27, 25-41. Joyce, W., Slocum, J., & Von Glinow, M. (1982) Person-
situation interaction: Competing models of fit. Journal of
Hambrick, D. C. (1980) Operationalizing the concept of busi- Occupational Behavior, 3, 265-280.
ness-level strategy in research. Academy of Management
Review, 5, 567-575. Lorange, P., & Vancil, R. F. (1977) Strategic planning sys-
tems. Englewood Cliffs, NJ: Prentice-Hall.
Hambrick, D. C. (1983a) An empirical typology of mature
industrial product environments. Academy of Manage- Marsh, H. W., & Hocevar, D. (1985) The application of con-
ment Journal, 26, 213-230. firmatory factor analysis to the study of self-concept: First-
and higher-order factor models and their invariance
Hambrick, D. C. (1983b) High profit strategies in mature cap- across groups. Psychological Bulletin, 97, 562-582.
ital goods industries: A contingency approach. Academy
of Management Journal, 26, 687-707. Miles, R. E., & Snow, C. C. (1978) Organiza/ionai strategy,
structure, and process. New York: McGraw-Hill.
Hambrick, D. C. (1984) Taxonomic approach to studying
strategy: Some conceptual and methodological issues. Miller, D. (1981) Toward a new contingency perspective: The
Journal of Management, 10, 27-42. search for organizational gestalts. Journal of Management
Studies, 18, 1-26.
Hambrick, D. C , MacMillan, I. C , & Day, D. L. (1982) Stra-
tegic attributes and performance in the BCG matrix—a Miller, D., & Friesen, P. H. (1977) Strategy making in context:
PIMS-based analysis of industrial product businesses. Ten empirical archetypes. Journal of Management Stud-
Academy of Management Journal, 25, 510-531. ies, 14, 259-280.
Harrigan, K. R. (1983) Research methodologies for contin- Miller, D., & Friesen, P. H. (1978) Archetypes of strategy for-
gency approaches to business strategy. Academy of Man- mulation. Management Science, 24, 921-933.
agement Review, 8, 398-405. Miller, D., & Friesen, P. H. (1984) Organizations: A guanfum
Hitt, M. A., Ireland, R. D., & Palia, K. A. (1982) Industrial view. Englewood Cliffs, NJ: Prentice-Hall.
firm's grand strategy and functional importance: Moder-
ating effects of technology and uncertainty. Academy of Miller, J. P. (1982, August) Toward clarity in contingency hy-
Management Journal, 25, 265-298. pothesis analysis. Paper presented at the meeting of the
Academy of Management, New York.
Hitt, M.A., Ireland, R. D., & Stadter, G. (1982) Functional
importance and company performance: Moderating ef- Milligan, G. W., & Cooper, M. C. (1985) An examination of
fects of grand strategy and industry type. Strategic Man- procedures for determining the number of clusters in a
agement Journal, 3, 315-330. data set. PsychomefriJca, 50, 159-179.
443
Mintzberg, H. (1978) Patterns in strategy formation. Manage- Southwood, K. E. (1978) Substantive theory and statistical
ment Science, 24, 934-948. interaction: Five models. American Journal of Sociology,
Morrison, D. G. (1969) On the interpretation of discriminant 83, 1154-1203.
analysis./ournai of A^arJreting Research, 6, 156-163. Thompson, J. D. (1967) Organizations in action. New York:
Porter, M. E. (1980) Compefi/ive strategy. New York: Free McGraw-Hill.
Press. Tosi, H. L., Jr., & Slocum, J. W., Jr. (1984) Contingency the-
Prescott, J. E. (1986) Environment as a moderator of strategy- ory: Some suggested directions. Journal of Management,
performance relationship. Academy of Management Jour- 10, 9-26.
nal, 29, 329-346. Van de Ven, A. H. (1979) Review of Aldrich's (1979) book—
Prescott, J. E., Kohli, A. K., & Venkatraman, N. (1986) The Organization and environments. Administrative Science
market share-profitability relationship: An empirical as- Quarterly, 24, 320-326.
sessment of assertions and contradictions. Strategic Man- Van de Ven, A. H., & Drazin, R. (1985) The concept of fit in
agement Journal, 7, 377-394. contingency theory. In L. L. Cummings & B. M. Staw
Roseboom, W. W. (1956) Mediation variables in scientific (Eds.), Research in organizational behavior (Vol. 7, pp.
theory. Psychological Review, 63, 249-264. 333-365). Greenwich, CT: JAI Press.
Rumelt, R. P. (1974) Strategy, structure and economic per- Venkatraman, N. (1986, August) Performance implications
formance. Boston: Harvard University, Graduate School of strategic coalignment. Paper presented at the meeting
of Business Administration. of the Academy of Management, Chicago.
Schendel, D. E., & Hofer, C. W. (1979) Strategic manage- Venkatraman, N., & Camillus, J. C. (1984) Exploring the con-
ment: A new view of business policy and planning. Bos- cepts of "fit" in strategic management. Academy of Man-
ton: Little, Brown. agement Review, 9, 513-525.
Schoonhoven, C. B. (1981) Problems with contingency the-
ory: Testing assumptions hidden withm the language of Venkatraman, N., & Grant, J. H. (1986) Construct measure-
contingency "theory." Administrative Science Quarterly, ment in strategy research: A critique and proposal. Acad-
26, 349-377. emy of Management Review, 11, 71-87.
Sharma, S., Durand, R. M., & Gur-Ane, O. (1981) Identifi- Venkatraman, N., & Prescott, J. E. (in press) Environment-
cation and analysis of moderator variables. Journal of strategy coalignment: An empincal examination of its per-
Marketing Research, 18, 291-300. formance implications. Strategic Management Journal.
Snow, C. C, & Miles, R. E. (1983) The role of strategy in the Venkatraman, N., & Walker, G. (1989) Strategic consistency
development of a general theory of organizations. In R. and business performance: Theory and analysis. Working
Lamb (Ed.), Advances in strategic management (Vol. 2, paper, Massachusetts Institute of Technology, Sloan
pp. 231-259). Greenwich, CT: JAI Press. School of Management, Cambridge, MA.

N. Venkatraman (Ph.D., University of Pittsburgh) is


Assistant Professor of Management, MIT Sloan School
of Management, Massachusetts Institute of Technol-
ogy, Cambridge, MA. Correspondence regarding this
article should be addressed to iiim at £52-553 Sloan
Building, MIT, Cambridge, MA 02139.
This is a significantly revised and expanded version
of a paper that appeared in the Academy of Manage-
ment Best Paper Proceedings, 1987. Discussions dur-
ing (he seminars at the Kellogg Graduate School of
Business Administration. Northwestern University,
Graduate School of Business Administration, New
York University, and Sloan School of Management,
Massachusetts Institute of Technology, were useful in
preparing this version. Gregory Dess, Qscar Haupt-
man, John Henderson, John Prescott, Vasu Ramanu-
jam, Harbir Singh, and Meera Venkatraman read
previous drafts and made useful comments and sug-
gestions.

444

You might also like