revisiting the miles and snow strategic framework

Strategic Management JournalStrat. Mgmt. J., 26: 47–74 (2005)

Published online 28 October 2004 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/smj.431

REVISITING THE MILES AND SNOW STRATEGICFRAMEWORK: UNCOVERING INTERRELATIONSHIPSBETWEEN STRATEGIC TYPES, CAPABILITIES,ENVIRONMENTAL UNCERTAINTY, AND FIRMPERFORMANCE

WAYNE S. DESARBO,1* C. ANTHONY DI BENEDETTO,2 MICHAEL SONG3

and INDRAJIT SINHA2

1 Smeal College of Business Administration, Pennsylvania State University, UniversityPark, Pennsylvania, U.S.A.2 Fox School of Business Administration, Temple University, Philadelphia, Pennsyl-vania, U.S.A.3 School of Business, University of Washington, Seattle, Washington, U.S.A.

The Miles and Snow strategic type framework is re-examined with respect to interrelationshipswith several theoretically relevant batteries of variables, including SBU strategic capabilities,environmental uncertainty, and performance. A newly developed constrained, multi-objective,classification methodology is modified to empirically derive an alternative quantitative typologyusing survey data obtained from 709 firms in three countries (China, Japan, United States).We compare the Miles and Snow typology to the classification empirically derived utilizingthis combinatorial optimization clustering procedure. With respect to both variable batteryassociations and objective statistical criteria, we show that the empirically derived solutionclearly dominates the traditional P-A-D-R typology of Miles and Snow. Implications anddirections for future research are provided. Copyright 2004 John Wiley & Sons, Ltd.

INTRODUCTION

For nearly a quarter century, the Miles and Snow(1978) strategic choice typology has been widelyembraced and been a subject of considerableresearch attention in both the management andmarketing strategy literatures (see, for example,Conant, Mokwa, and Varadarajan, 1990; Ham-brick, 1983; McDaniel and Kolari, 1987; McKee,Varadarajan, and Pride, 1989; Ruekert and Walker,

Keywords: strategic types; constrained classification;SBU strategic capabilities; environmental uncertainty;SBU performance*Correspondence to: Wayne S. DeSarbo, Smeal College of Busi-ness Administration, Pennsylvania State University, Room 701D,Business Administration Building, University Park, PA 16802-3007, U.S.A. E-mail: [email protected]

1987; Shortell and Zajac, 1990). Authors attributethe typology’s longevity and excellence to itsinnate parsimony, industry-independent nature, andto its correspondence with the actual strategic pos-tures of firms across multiple industries and coun-tries (Hambrick, 2003). Miles and Snow (1978)originally envisaged strategy as an agglomera-tion of decisions by which a strategic businessunit (SBU) aligns its managerial processes (includ-ing its capabilities) with its environment. Accord-ingly, businesses were conceptually classified onthe basis of their patterns of decisions into the nowfamiliar Prospector–Analyzer–Defender–Reactor(P-A-D-R) framework. Prospectors are technolog-ically innovative and seek out new markets; Ana-lyzers tend to prefer a ‘second-but-better’ strategy;

Copyright 2004 John Wiley & Sons, Ltd. Received 8 July 2002Final revision received 25 May 2004

48 W. S. DeSarbo et al.

Defenders are engineering-oriented and focus onmaintaining a secure niche in relatively stable mar-ket segments; and Reactors lack a stable strategyand are highly responsive to short-term environ-mental exigencies.

The Miles and Snow framework continues tobe the most enduring strategy classification systemavailable (Hambrick, 2003). Despite the entrenchednature of the Miles and Snow (1978) framework,a number of researchers have commented on theneed for further empirical validation and test-ing of its underlying assumptions (Conant et al.,1990; Zajac and Shortell, 1989; Shortell and Zajac,1990). Such authors have noted the fact that theoriginal Miles and Snow research was limitedin the number of industries and the range ofcapabilities studied. They did not systematicallystudy all the possible linkages between capabilitiesand strategic type, nor did they attempt to provethe validity of their typology across other indus-try types. Similarly, for the most part, they didnot determine empirically whether, for example,Prospectors might outperform Defenders undersome circumstances (one exception was Zajac andShortell, 1989, who found that Prospectors outper-formed Defenders in the volatile healthcare indus-try).

Hambrick (1983) noted that the parsimoniousMiles and Snow model offers an incomplete viewof strategy. Its generic character ignores industryand environmental peculiarities. In fact, Miles andSnow (1978) stressed that the various strategictypes would perform equally well in any industry,provided that the strategy was well implemented.This latter stance is inconsistent with the moretypical view that an environment favors certaintypes of strategies. As Hambrick (1983) noted,little consideration of the environment–strategylink has been given in Miles and Snow, and nosystematic evidence has been provided on howstrategic types differ in their functional attributes(Miles and Snow, 1978: 7).

Hambrick (1983) attempted to relate differencesin strategy to differences in performance condi-tioning on environmental and functional attributes.Many of his findings conflict with those pre-dicted from Miles and Snow (1978), and suggestthat deeper empirical research into the relation-ships between capabilities, strategic type, and per-formance across a wider range of industries iswarranted. Yet, despite the inferred role of envi-ronmental factors (Hambrick, 1983; Zajac and

Shortell, 1989), environmental effects remainedempirically uninvestigated. Additionally, capabil-ities other than marketing-related ones were notexplored in the Conant et al. (1990) study. Anoptimal approach should be one that explicitlyaccounts for both capabilities and environmentalattributes. In fact, Conant et al. (1990) called forfuture research to examine the synthesis and inte-gration across multiple sets of items. Finally, Ham-brick (1984) also criticized such general classifi-cation schemes in not being quantitatively based.‘Typologies represent a theorist’s attempt to makesense out of non-quantified observations. Theymay have the advantage of being ‘poetic’ . . ., thatis ring true, often sounding very plausible. How-ever, since they are the product of rather personalinsight, they may not accurately reflect reality. Or,more likely, they may serve well for descriptivepurposes but have limited explanatory or predic-tive power’ (Hambrick, 1984: 28).

The research objective of this manuscript isto introduce a new quantitative methodology toderive strategic typologies empirically in an at-tempt to resolve some of these criticisms lev-eled against the Miles and Snow typology overthe years. In particular, we address the criticismsof Hambrick (1983, 1984) discussed above byexplicitly including environmental attributes andSBU strategic capabilities in empirically derivingstrategic type via an objective quantitative method-ology, as well as exploring performance differ-ences across strategic types. Our procedure per-mits a more comprehensive modeling and group-ing of SBUs, and provides the flexibility of test-ing alternative taxonomies in a comparative man-ner. Our goal is not to uncover generic strate-gic types that could be necessarily generalizedacross all time periods, industries, data samples,etc., as we believe this would be impossible to do.Rather, we propose a quantitative methodology tobe utilized across any scenario in order to derivestrategic types for a given empirical application(e.g., for a given time period, industry). We reportan extensive empirical research study conductedwith 709 SBUs in three different countries (China,Japan, United States). We find that the particularfour mixed-type solution derived by our proposedmethodology for this particular application empir-ically dominates the Miles and Snow classifica-tion in terms of objective statistical criteria andprovides much better explanatory power in termsof relationships with related batteries of variables

Copyright 2004 John Wiley & Sons, Ltd. Strat. Mgmt. J., 26: 47–74 (2005)

Revisiting Miles and Snow 49

such as strategic capabilities, environmental uncer-tainties, and performance. While some similarity tothe Miles and Snow typology is witnessed, we do,however, uncover significant differences in termsof strategic capabilities, environmental uncertain-ties, and performance compared to the Miles andSnow typology as applied to this sample. Since wedo not impose any a priori single or fixed classifi-cation scheme to the data, our quantitative frame-work that we tailor to this particular classificationproblem allows for the optimally derived typol-ogy to be objectively compared in terms of overallfit with any alternative typology (e.g., Miles andSnow) that may be imposed from our dataset.

THEORETICAL BACKGROUND

The Miles and Snow typology

Based on their field studies conducted in fourindustries (textbook publishing, electronics, foodprocessing, and health care), Miles and Snow(1978) proposed a strategic typology classifyingbusiness units into four distinct groups: Prospec-tors, Analyzers, Defenders, and Reactors. Prospec-tors lead change in their industries, principally bylaunching new products and identifying new mar-ketplace opportunities. Defenders find and seekto maintain a secure niche in a stable productarea. Rather than concentrating on new productor market development, Defenders stay withina limited range of products, focusing more onresource efficiency and process improvements thatcut manufacturing costs. Analyzers share traits ofboth Prospectors and Defenders. While defendingpositions in some industries, they may selectivelymove quickly to follow promising new productor market developments. Although they may initi-ate product or market development, Analyzers aremore likely to follow a second-but-better strategy.A business pursuing an Analyzer strategy com-petes sometimes as a Defender, and other times asa Prospector, since it requires substantial resourcesto be able to do both simultaneously. These threestrategic types (Prospectors, Defenders, and Ana-lyzers) are consistent in their strategic selection,and will perform well so long as their implemen-tation is effective. They tend to outperform Reac-tor businesses who lack a consistent strategy, andrespond, usually inappropriately, to environmen-tal pressures as they arise. For a more extensive

discussion of the strategic types, see Hambrick(1983), Conant et al. (1990), and Walker et al.(2003).

The Miles and Snow (1978) typology has beenextensively applied in the strategy literature, andhas generally been supported (Snow and Ham-brick, 1980; Hambrick, 1983, 1984; McDanieland Kolari, 1987; McKee et al., 1989; Shortelland Zajac, 1990; Webster, 1992). Conant et al.(1990) developed an 11-item scale to classify firmsinto strategic types; this scale has been success-fully applied elsewhere (e.g., Dyer and Song, 1997;DeSarbo et al., 2004).

The role of firm strategic capabilities

Strategic capabilities have been defined as ‘com-plex bundles of skills and accumulated knowledgethat enable firms [or SBUs] to coordinate activ-ities and make use of their assets’ (Day, 1990:38) to create economic value and sustain competi-tive advantage. Many kinds of strategic capabilitiesthat are common to businesses can be identified.Technological, product development, productionprocess, manufacturing, and logistics capabilitiesallow a firm to keep costs down and/or differen-tiate its offerings. Increased production efficiencyreduces costs, improves consistency in delivery,and ultimately increases competitiveness (Day,1994). Market sensing, channel and customer link-ing, and technology-monitoring capabilities allowa business to respond swiftly to changing customerneeds and to exploit its technological strengthsmost effectively (Day, 1994). Marketing capabili-ties, such as skills in segmentation, targeting, pric-ing, and advertising, permit the business to takeadvantage of its market sensing and technologicalcapabilities and to implement effective marketingprograms. Capabilities in information technology(IT) help the firm diffuse market information effec-tively across all relevant functional areas that it canexploit to direct the new product development pro-cess. Finally, management-related capabilities sup-port all of the above and include human resourcemanagement, financial management, profit andrevenue forecasting, among others.

Miles and Snow (1978) had reported theexistence of relationships across strategic typesand such firm strategic capabilities. Prospectors,for example, tend to compete by anticipatingnew product or marketplace opportunities andthrough technological innovation. These firms



thrive in unstable, volatile environments—thosemarked by rapid technological change such as inthe biotechnology, medical care, and aerospaceindustries (Walker et al., 2003). Prospectors usea first-to-market strategy and typically succeed bybeing able to develop new technologies, products,and markets rapidly (McDaniel and Kolari, 1987;Conant et al., 1990). Walker et al. (2003) notethat Prospectors require strength in product R&Dand product engineering, and perform best whenthe amount spent on product R&D is high. Theyalso rely on solid market research and build closeties with distribution channels to ensure that theR&D produces products that meet customer needs(Hambrick, 1983; McDaniel and Kolari, 1987;Shortell and Zajac, 1990). Also, IT capabilitiesfacilitate internal communication and functionalintegration that are critical to new productsuccess (Swanson, 1994; Moenaert and Souder,1996; Griffin and Hauser, 1996; Bharadwaj,Bharadwaj, and Konsynski, 1999). Miles andSnow (1978) have noted that Prospectors needto have the most complex coordination andcommunication mechanisms, as they are mostreliant on new product development to sustaincompetitiveness (Robinson, Fornell, and Sullivan,1992).

In contrast, Defenders attempt to locate andmaintain a secure niche in a relatively stable prod-uct or service area. They do not look outside theirestablished product-market domain to identify newopportunities (McDaniel and Kolari, 1987; Shor-tell and Zajac, 1990). They tend to offer a morelimited range of products or services than theircompetitors, and try to protect their domains byoffering higher quality, superior service, and lowerprices (Hambrick, 1983). Clearly, to be effectivein achieving these objectives, Defenders need topossess a high level of marketing and market link-ing capabilities (Conant et al., 1990; Walker et al.,2003), and have to concentrate on resource effi-ciency, cost-cutting, and process improvements.

According to Miles and Snow (1978), successfulprospecting will have the effect of strengtheningtechnology and R&D capabilities. In other words,‘Prospectors tend to want to continue prospect-ing’ (Hambrick, 1983), since this is what theydo best. Similarly, Defenders will likely keep ondefending, while Analyzers will build upon bothprospecting and defending capabilities. Reactorsdo not capitalize on the set of capabilities they

already have built up, but rather they shift strate-gic orientation in reaction to competitive pressures,thus they will usually be at a disadvantage to thosefirms that are competing from an established posi-tion of strength.

The role of environmental uncertainty

The strategy literature generally posits that strategyselection is conditional on how closely a businessis aligned with its environment (e.g., Hofer andSchendel, 1978; Porter, 1980). For example, inconditions of high uncertainty in technology, cus-tomer, or competitive environments, the firm mustbe able to accommodate to environmental change(Miller and Friesen, 1978, 1983; Utterback, 1979).Environmental uncertainty may require a firm to beable to respond more rapidly to unforeseen changein order to survive (Lawrence and Lorsch, 1967;Covin and Slevin, 1989).

Researchers commenting on the Miles and Snowtypology have noted that different environmentalcircumstances may be conducive to certain strate-gic types (e.g., Hambrick, 1983). Factors of envi-ronmental uncertainty that are likely to be per-ceived important by managers include such issuesas the degree of predictability of financial andcapital markets, government regulation and inter-vention, actions of competitors, actions of sup-pliers, and general conditions facing the organi-zation (Hrebiniak and Snow, 1980). According toWalker et al. (2003), the following environmentalcharacteristics tend to favor Prospector strategies:(1) the industry is in the early stage of the prod-uct life cycle (PLC); (2) market segments are stillunidentified or undeveloped; (3) industry technol-ogy is newly emerging; (4) there are few estab-lished competitors; (5) industry structure is still inthe process of evolving; and (6) industry concen-tration is high, e.g., one firm holds most of themarket share. The reverse conditions tend to favorDefender strategies, whereas Analyzer strategiesare favored in the ‘middle ground.’ As an exam-ple, if a large number of competitors exist, butindustry structure is still evolving and a shakeoutis inevitable, an Analyzer strategy may be moreappropriate.

Comparatively few research studies haveattempted empirically to support the proposed rela-tionships between environment, strategic capabil-ity, and Miles and Snow strategic types as outlinedby Walker et al. (2003). Hambrick (1983) has



examined the effects on strategic choice of twoenvironmental variables, product life cycle stage,and industry innovation, using the PIMS database.Zajac and Shortell (1989) find that Prospectorsand Analyzer hospitals outperform Defender hos-pitals in the rapidly changing health care environ-ment. There is a need for a greater considerationof the effects of the environment and capabilitieson strategic choice.

Strategic capabilities and environmentaluncertainty as antecedents to strategic choice

It is clear from the above discussion that relation-ships exist across business capabilities, the envi-ronment, and strategic type. Yet, the precise natureof the relationships among the antecedent vari-ables and strategic type remains unclear and inneed of further empirical investigation. The fourMiles and Snow types, originally developed for asmall number of industries, may not sufficientlywell describe the strategic types that exist in otherindustry settings. For example, a disparate set ofantecedent circumstances may lead a business toadopt a Prospector strategy. It may be in a rapidlychanging market, and rely on leadership in newproduct development to stay ahead of the compe-tition. Or, it may be in a relatively stable market,but seek to exploit a new emerging technologyto serve customer needs in novel ways. Similarly,several different paths may lead to the adoption ofa Defender strategy. A market share leader in apredictable market may choose to defend its posi-tion, as may a business seeking to reduce riskexposure in an extremely unpredictable market.In sum, a business will select a particular strate-gic type based on its particular internal strengths(capabilities) and external (environment) circum-stances, and the strategic types that are actuallyemployed may not, in fact, be cleanly interpretableas the four Miles and Snow (1978) categories. Thissuggests the implementation of a contingency-based approach that can empirically derive suchsituation-specific strategic types.

Performance and strategic types

As stated earlier, Miles and Snow (1978) suggestedthat the three ‘archetypal’ strategic types (Prospec-tors, Analyzers, and Defenders) should all per-form well, and should also all outperform Reactors

due to the latter’s lack of a stable strategy. Ham-brick (1983) points out that the original Miles andSnow model does not seek to predict which of thearchetypal strategic types would be highest in per-formance, or under what circumstances; in fact,‘performance’ had not been clearly defined. Henotes that his comment is not to be taken as a crit-icism of the Miles and Snow model, as their intentwas to develop a typology of corporate strategy,not to explore the performance consequences. Nev-ertheless, it is clear that more research was neededon the topic of strategic type and performance.Subsequent empirical tests of the Miles and Snowframework (e.g., Conant et al., 1990; Dyer andSong, 1997) have generally supported the expecta-tion that the three archetypal strategic types wouldoutperform Reactors. More research is still war-ranted, particularly with regard to whether industryclassification, environmental factors, or other vari-ables might affect the performance achieved bydifferent strategic types.

Next, we proceed with describing the methodol-ogy to be utilized for an empirical contingency-based approach for deriving strategic types. Asindicated before, we do not begin the analysiswith any a priori expectations about the numberof strategic types, nor concerning their nature andcomposition, nor in how they differ. We allow theselection of the optimal typology to be objectivelyand empirically determined by the structure in thedata and the statistical fit of the models. Analyz-ing the characteristics of the identified strategictypes in the empirically derived ‘best’ solution per-mits us to explore the interrelationships discussedabove. Hence, the result of our analysis is a strate-gic typology that yields significantly richer insightsabout the actual strategic postures adopted by firmsacross industries and that speaks to the associa-tions between strategic capabilities, environmentalattributes, strategic choice, and resulting perfor-mance.

METHODOLOGY

The constrained multi-objective classificationapproach

The procedure we employ here is a gen-eral, but flexible, modeling framework forconstrained, multi-objective classification (calledNORMCLUS) that is well suited for groupingentities such as firms or customers. DeSarbo



and Grisaffe (1998) have formulated the originalNORMCLUS procedure which we modify hereoffering greater flexibility in accommodating mul-tiple data batteries collected on the same firms,and in permitting a variety of constraints such asallowing for different types of clusters, controllingthe minimum size (e.g., number of firms) of theresulting clusters, etc. The overall benefits of thisconstrained, multi-objective, clustering methodol-ogy over traditional classification schemes lie inits ability to satisfy a number of key empiricallygrounded and managerially relevant criteria (seeDeSarbo and Grisaffe, 1998, for a more completediscussion). What makes NORMCLUS particu-larly useful to the problem of strategic choice hereis its ability to appropriately accommodate mul-tiple batteries of variables (here, strategic types,firm capabilities, environmental uncertainty, andfirm performance) in a multi-objective functionsetting. In addition, NORMCLUS provides a sta-tistical framework in which competing solutionscan be formally compared in terms of comparativegoodness-of-fit statistics (pseudo F -tests) explic-itly taking into consideration the number of esti-mated parameters and model degrees of freedom.This aspect is of particular interest in the strate-gic choice application to follow in deriving anoptimal solution in terms of the ‘best’ number ofstrategic types. Also, the four strategic type solu-tion of Miles and Snow (1978) provides a natural‘straw man’ or ‘benchmark’ against which to com-pare alternative empirically derived solutions onstatistical fit grounds (we will also examine tradi-tional cluster analysis as an alternative comparisonstandard). Finally, NORMCLUS has the ability toaccommodate user-specified constraints to controlvarious aspects of the resulting cluster solution(e.g., minimum cluster size).

The NORMCLUS methodology

NORMCLUS utilizes recent developments in com-binatorial optimization in finding partitions offirms to optimize a user-specified objective func-tion subject to user-specified constraints. Supposethere are m = 1, . . ., M objective functions thatare comparably scaled as to range and distribution,and that a particular clustering/classification prob-lem implies their joint optimization (minimizationor maximization). In the utility function methodof multi-criteria optimization, a utility function

Um(fm) is defined for each objective, fm, depend-ing on the importance of fm compared to the otherobjective functions. Then, one can define a totalutility function U as:

U =M∑

m=1

Um(fm) (1)

A solution vector θ∗ is then found by optimizingU subject to user-specified constraints:

hj (θ) = 0 j = 1, . . . , J (2)

gs(θ) ≤ 0 s = 1, . . . , S (3)

A specific form for Equation 1 above can be givenby:

U =M∑

m−1

Um =M∑

m−1

αmfm(θ), (4)

where αm is a scalar weighting or importance fac-tor associated with the mth objective function,fm(θ), with

∑m αm = 1, and hj (θ) and gs(θ) are

linear or non-linear equality and inequality con-straints respectively. In the absence of theory, αm

are typically set equal to 1/M . Rao (1996) calls thisthe ‘weighting function method’ for solving multi-criteria optimization problems that typically gener-ate Pareto optimal solutions. Rao (1996) describesa number of alternative multi-criteria optimizationframeworks such as the inverted utility method,the global criterion method, the bounded objec-tive function method, and lexicographic method,which can all be accommodated in NORMCLUS.Appendix 1 summarizes the various types of con-straints that can be implemented or tailored to anyparticular taxonomy problem.

Estimation algorithms for strategic typesanalysis

A variety of optimization procedures are avail-able in NORMCLUS for parameter estimationincluding ordinary least-squares, constrained least-squares, and a host of combinatorial optimiza-tion procedures employing genetic algorithms (cf.Rao, 1996, for a survey), simulated annealing(cf. DeSarbo, Oliver, and Rangaswamy, 1989),lambda-opt procedures (cf. Lin and Kernighan,1973), as well as a variety of heuristics such asgreedy algorithms and taboo search. The particular



selection of which combinatorial optimization pro-cedure to use depends very much on the structureof the classification problem at hand.

The primary objective for this strategic typeclassification problem is to derive strategic typeclusters that satisfy many of the criteria dis-cussed above. More specifically, we want clus-ters of firms whose strategic choices are different,whose environmental characteristics are different,whose strategic capabilities are distinct, with dif-ferent performance levels, whose sizes are substan-tial, and whose results can be easily projected tothe entire customer base. Given these objectives,and the multidimensional nature of the problem,a multi-criteria objective function is defined asearlier described in the weighted utility functionmethod. Let X denote Conant’s strategic type bat-tery of variables, Y the environmental uncertaintybattery of variables, P for the performance vari-able battery, and Z the firm strategic capabilitiesbattery of variables. For this particular application,we define four separate parts of the combined util-ity function that is to be maximized:

f1 = |M1/T1| (5)

f2 = |M2/T2| (6)

f3 = |M3/T3| (7)

f4 = |M4/T4| (8)

where Mj = the between cluster sum-of-squaresand cross-products for battery j ; Tj = the totalsum-of-squares and cross-products for battery j ;and | | = the determinant operator.

Thus, fi are eta square measures which mea-sure separation in the component battery clustervariables. Note that all fi range between 0 and 1,as does the combined function U in Equation 1.Here, we set α = 0.25 to weigh each compo-nent of U (strategic choice, environment, strate-gic type, performance, and strategic capabilities)equally given the nature of this specific applicationand the lack of any strong a priori theory. Thus,we are looking to summarize the associations andinterrelationships (not causal) between these fourdistinct batteries of variables and simultaneouslyderive a strategic type classification that optimizesthese interrelationships. That is, what taxonomycan be estimated that maximizes the separationor differences between each of the variable itemswithin each battery (i.e., quantitatively optimizeEquation 4 above)?

In addition, we imposed a number of other con-straints on the final solutions. First, a mutuallyexclusive partitioning of the sample into sepa-rate, non-overlapping clusters was desired giventhe Miles and Snow (1978) tradition. Second, nosingle cluster was specified to contain less than5 percent of the sample to avoid outlier small clus-ters. Appendix 2 describes the modified lambda-opt constrained combinatorial optimization algo-rithm especially formulated for this particularapplication.

RESEARCH DESIGN

Instrument development

Our constructs are defined based on competitivecapability theory (Day, 1994; Conant et al., 1990).Our review of the marketing and management lit-erature, however, found no existing scales for mostof the SBU capabilities being studied. We thereforecarried out a multi-step instrument developmentprocedure to be certain of the validity and relia-bility of the operationalized constructs (Churchill,1979).

Step 1: Measurement items for each capabilitytype

We identified relevant measurement scales fromthe marketing and management literatures, andgrouped them into the five capability types toform the initial pool of items. In cases where wefelt all the dimensions of the construct had notbeen adequately covered, we created new addi-tional items. We refined the scales through focusinterviews with managers in two SBUs, askingthem to assess the completeness of the scales.The results of the interviews suggested that thescales were relevant and complete. The respon-dents were also asked to rate their own SBU rela-tive to major competitors on each scale item using11-point Likert-type scales (0 = much worse thancompetitors, 10 = much better than competitors);they had no difficulty completing this task.

Step 2: Scale development

Seven judges (two professors and five doctoralstudents with background in measurement devel-opment) were asked to sort a list of scale items intothe strategic capability scales, following Davis’s



(1986) procedure. Working independently, eachjudge sorted the scale items into the strategic capa-bility types. Following Davis (1986, 1989), con-struct convergence and divergence were examinedby assessing inter-rater reliability. The percent-age of correct placement of items was calculatedas the proportion of items placed by the sevenjudges within the intended theoretical construct.The minimum percentage obtained was 84 percent.The items that were frequently incorrectly placedwere eliminated from the pool. As a second testof inter-rater reliability, Cohen’s kappa (Cohen,1960), measuring level of agreement in categoriza-tion, was calculated for each pair of judges. Kapparanged from 0.97 to 0.82, exceeding the minimumacceptable level of 0.65 (Jarvenpaa, 1989). Weconcluded that the items demonstrated convergentvalidity within the related capability and discrimi-nant validity across the capabilities (Davis, 1986,1989).

Step 3: Instrument pre-testing

To further assess scale validity and reliability, theremaining items were combined into a single ques-tionnaire, which was pre-tested with 32 managersin the two SBUs. The pre-test resulted in the dele-tion of two more scale items. A second pretest, thistime with 41 EMBA students taking a new prod-uct development class, was then conducted. Theresults were factor analyzed and scale reliabilitieswere assessed. Two additional items were deleted,resulting in the final set of capability scale items.

Step 4: Cross-cultural validation of the researchinstrument

Here, we explicitly sought to avoid an Americanbias to the problem. Following Song and Parry(1996), the above three steps served only as a start-ing point for our scale development, recognizingthat conducting international research requires an‘inside-out’ approach in order to develop valid andreliable measures of appropriate constructs. Theresearch was designed with the intent of estab-lishing equivalent measures for the study of theJapanese and Chinese firms as appropriate anddeveloping new measures and/or constructs as nec-essary.

To make sure the translation was accurate andthat the question meanings were not altered, weused a double-translation method to translate the

questionnaire into Japanese and Chinese (Songand Parry, 1996). A comparison of the resultingquestionnaires revealed considerable consistencyacross translators. The questionnaires were firsttranslated into the foreign language by a translatorand then translated back into English by a differenttranslator to ensure translation equivalence.

After the translation procedure, we conductedfield research in six Japanese firms and two Chi-nese firms in which we examined SBU capabili-ties and innovation strategies. The purposes of theadditional field research were: (1) to establish thecontent validity of the concepts and the hypoth-esized relationships among the constructs; (2) toestablish equivalence of the constructs, concepts,measures, and samples; and (3) to assess the pos-sibility of cultural bias and response format bias(Song and Parry, 1996). The field research stud-ies were conducted over a 9-month period withmultiple visits to the companies.

The field research studies were important forseveral reasons. First, they facilitated an assess-ment of construct equivalence (i.e., conceptual,functional, and category equivalence). Second,they indicated that (with minor modifications) themeasurement scales were appropriate for study-ing strategic capability and strategic types inJapanese and Chinese context. Third, the resultsfrom these field research studies suggested thatit is more appropriate to ask the respondents torate their SBU on each of the strategic capabilityscale items relative to their major competitors (seeAppendix 3 for exact wording). Fourth, 11-pointLikert-type scales (from 0 to 10) were chosen forthe four strategic capability scales. Song and Parry(1997a, 1997b) also suggest that this format is bet-ter understood across multiple nations than are the1–7 or 1–6 scales more commonly seen in NorthAmerican research, because of their structural sim-ilarities to the metric system. Thus, we used an11-point scale to elicit levels of agreement, withvalues ranging from 0 (much worse than our com-petitors) to 10 (much better than our competitors).

Measures

Appendix 3 presents all of the measures utilized inthese data collection phases. Five major strategiccapability areas were explicitly measured:

1. Market-Linking Capabilities. These capabilitiesrelate to focused market sensing and linking



outside the organization and were rated accord-ing to several scale items developed from Day’s(1994) descriptions of such capabilities. Thecomponent items measure the relative capa-bilities in creating and managing durable cus-tomer relationships, creating durable relation-ships with suppliers, retaining customers, andbonding with channel members such as whole-salers and retailers.

2. Technological Capabilities. Technological capa-bilities, pertaining to production process effi-ciency, cost reduction, greater consistency indelivery, and greater competitiveness, were rat-ed according to scale items drawn from Day’s(1994) set of such capabilities. The items mea-sure relative capabilities in the prediction oftechnological change, technology and new prod-uct development, and product facilities.

3. Marketing Capabilities. Marketing capabilitieswere measured using a set of scale itemsdrawn from the Conant et al. (1990) study ofmarketing capability and strategic type. Theseinclude knowledge of customers, knowledge ofcompetitors, integration of marketing activities,skills in segmentation and targeting, and effec-tiveness of pricing and advertising programs.

4. Information Technology Capabilities. Informa-tion technology capabilities refer to the rela-tive capabilities that help an organization createtechnical and market knowledge and facilitateintra-organizational communication flow. Wedeveloped items that measure the possession ofinformation technology systems for new prod-uct development, cross-functional integration,technology and market knowledge creation, andinternal communication. These items were sub-ject to the measurement development proceduredescribed above.

5. Management Capabilities. These capabilitiesinclude the ability to integrate logistics systems,control costs, manage financial and humanresources, forecast revenues, and manage mar-keting planning. We developed a set of six itemsthat measure the possession of these manage-ment capabilities.

The second battery of survey items pertains toenvironmental uncertainty. Here, we employedthree separate scales of six items each to assess dif-ferent aspects of the business environment uncer-tainty. The technological environment uncertaintyscale included the assessment of technological

change, the extent of technical opportunity, thedifficulty of technological forecasting, and otheraspects of technology. The assessment of the mar-ket environment uncertainty was based on changesin customer preferences, customer price sensitivity,customer product needs, changing customer base,and ease of forecasting marketplace changes. Thecompetitive environment uncertainty scale assessedthe extent of promotion and price wars, ability offirms to match competitive offers, and other com-petitive aspects. On all of these scales, a higherscore means that the environment is more uncer-tain.

The third battery of items concerned the classifi-cation of Miles and Snow strategic types. Here, weused the data collected in the second phase of thedata collection process to classify the SBU/divisioninto the four strategic types. The 11-item scale isthe same one previously developed and validatedby Conant et al. (1990).

Finally, performance data were collected. Of709 SBUs, we collected complete performancedata for 549 of them. The measures collectedwere: PROFIT (i.e., total revenue—total variablecosts)/total revenue); ROIPEC (i.e., an averagepercentage of the return on investment in thisbusiness unit over the past 3 years); ROI (returnon investment); ROA (return on assets); RMS(relative market shares); CUSRET (overall cus-tomer retention); CUSRET2 (retention of majorcustomers); SALESGR (sales growth rate); PERF1(overall profit margin relative to the objective forthis business unit); PERF2 (overall sales relativeto the objective for this business unit); and PERF3(overall return on investment relative to the objec-tive for this business unit). Note, the NORMCLUSprocedure was modified accordingly to accommo-date missing data concerning the performance bat-tery.

Data collection procedures

Our data were obtained from a large-scale mailsurvey of the companies listed in Ward’s Busi-ness Directory, Directory of Corporate Affiliations,and World Marketing Directory. A proportionate-stratified random sample of 800 firms from China,Japan, and the United States, with each industryas a stratum, was drawn. The questionnaires forJapanese and Chinese firms were translated intoJapanese and Chinese using a double-translationmethod as noted above. These three countries were



chosen for two reasons. First, they are, in thatorder, the three largest economies in the world,as measured by purchasing power parity (WorldBank, 2000), and are by any measure amongthe top five economies worldwide. Second, theJapanese and Chinese business environments arevery different from the U.S. business environment,which allows us to find evidence of the empiricalgeneralizability of our model. We believe this rep-resents the largest empirical test of the Miles andSnow framework done to date.

The data collection consisted of three stages:pre-survey, data collection on SBU strategies, anddata collection on relative capabilities, environ-mental uncertainty, and performance. In the firststage, we sent a one-page survey and an intro-ductory letter requesting participation to all theselected firms, and offered a list of availableresearch reports to participating firms. Each firmwas asked to select an SBU/division for par-ticipation and provide a contact person in thatSBU/division. Of the 2400 firms contacted, 392 inthe United States, 429 in Japan, and 414 in Chinaagreed to participate and provided the necessarycontacts at the SBU/division level.

In the second stage, concerning strategic types,the designated SBU managers were contacteddirectly by the researchers. A questionnaire, a per-sonalized letter, and the agreement to participationform from the first stage were mailed to eachSBU manager. We employed a three-wave mailingbased on the recommendations of Dillman (1978).We received data on the multi-item measures ofthe strategic types from 308 firms in the UnitedStates, 354 firms in Japan, and 352 firms in China.Two items at the end of the instrument assessedrespondents’ confidence in their ability to answerthe questions. The individuals with a low level ofconfidence (less than 6) were excluded from thefinal sample.

In the third stage, concerning the strategic capa-bilities, performance, and environmental variables,another questionnaire was sent to the SBU man-agers, followed again by a three-wave mailing.We received data on the relative capabilities from216 U.S. firms, 248 Japanese firms, and 245 Chi-nese firms. This constituted our final sample fromeach country, and these sample sizes correspond toresponse rates of 27.0 percent in the United States,31.0 percent in Japan, and 30.6 percent in China.

The final sample includes the following indus-tries: chemicals and related products; electronics

and electrical equipment; pharmaceuticals, drugs,and medicines; industrial machinery and equip-ment; telecommunications equipment; semicon-ductors and computer-related products; instrumentsand related products; and others (air conditioning;transportation equipment, etc.). The majority ofparticipating SBUs/divisions had annual sales of$11–750 million and 100–12,500 employees.

To examine possible non-response bias and therepresentativeness of the participating firms, weperformed a MANOVA to compare early respon-dents with late respondents on all four capabilityvariables. The results were not significant at the95 percent confidence level, suggesting no signifi-cant difference between the early respondents andthe late respondents. Therefore, we concluded thatnon-response bias is not a concern.

We classified the SBUs into the four Milesand Snow strategic types (Prospector, Analyzer,Defender, or Reactor) to permit comparison withour empirically derived strategic types. To do thisclassification, we used the ‘majority-rule decisionstructure’ (see Conant et al., 1990, for details),1

with the following modification: for an SBU tobe classified as a Prospector or a Defender, itmust have at least seven ‘correct’ answers outof the 11 items. Using this procedure, we clas-sified the 709 SBUs/divisions as follows: 234Prospectors, 220 Analyzers, 168 Defenders, and87 Reactors.

EMPIRICAL RESULTS

Fit of new strategic types vs. Miles and Snowstrategic types

We conducted a series of analyses employing theconstrained, multi-objective classification method-ology (NORMCLUS) described in the Methodol-ogy section. Initially, we formed four batteries ofitems: the 11 strategic type items, the 27 strategiccapability items, the 18 environment items, andthe 11 performance variables. Each battery wasequally weighted (0.25) in the analysis. Table 1presents the various goodness-of-fit values for eachbattery as well as the total fit value overall for

1 In this procedure, an SBU is classified as a Prospector if themajority of responses to the 11-item scale correspond to theProspector answers. A similar rule is used to classify SBUs intothe other three strategic types (see the Strategic Typology scalein Appendix 3).



Table 1. Comparative goodness-of-fit values

Number of groups Strategic type Capabilities Environment Performance Total fit

2 0.255 0.465 0.637 0.646 0.501∗∗

3 0.511 0.753 0.622 0.787 0.668∗∗

4 0.689 0.819 0.771 0.881 0.790∗∗

5 0.861 0.740 0.863 0.902 0.841n.s.

M&S(4)∗ 0.999 0.376 0.396 0.570 0.586n.s.KMEANS(4) 0.554 0.567 0.584 0.655 0.592n.s.

∗ This is the four-group Miles and Snow solution used as the benchmark.∗∗ p < 0.001; n.s., not significant improvement at p < 0.10.

K = 2, 3, 4, and 5 types (clusters). Based onnested model Pseudo F -tests on � for the vari-ous numbers of strategic types, we find that fourstrategic types result from this analysis, and sub-sequently increasing the number of strategic typesfrom 4 to 5 does not produce a significant increasein the overall goodness-of-fit measure, �. As canbe seen in Table 1, the four strategic group solutionappears to render consistent fits over each of thefour batteries of items (0.689, 0.819, 0.771, 0.881)and a highly significant (p < 0.01) improvementof 0.790 overall.

Table 1 produces the corresponding fits for thetraditional Miles and Snow (1978) classificationscheme for the purpose of comparison. Here, weclustered just the 11 strategic type items into fourclusters, which resulted in very clear P, A, D,and R groups. As shown in the table, the 0.999fit value for the strategic type items in this anal-ysis approaches 100 percent, which indicates anear-perfect separation of the resulting strategictypes when these 11 items are considered solely.However, one can see the price paid for stick-ing to this traditional approach in terms of verylow fit values for the environment and strate-gic capability item batteries (0.376 and 0.396).This implies low statistical relationships or asso-ciations between the traditional Miles and Snowtypology and environmental conditions and strate-gic capabilities of these firms, which is clearlycounter to theory. Even the performance battery fitin the Miles and Snow taxonomy is much lower(0.570) than what we witness for the derived four-group solution. Of particular note is the fact thatthe overall goodness-of-fit value is substantiallyhigher for the derived four-group solution obtainedfrom NORMCLUS when applied to this dataset.In fact, as has been depicted in Table 1, the Milesand Snow four-group solution is even dominated

statistically by the NORMCLUS two-group solu-tion in this instance. As will be shown shortly,there are substantial gains and insights obtainedfrom utilizing the proposed strategic type frame-work in a contingency sense together with otherconceptually meaningful variables compared to thetraditional Miles and Snow typology considered inexclusion.

Note that Table 1 also shows the comparable fitacross the various variable batteries for a four-cluster traditional KMEANS clustering solutionwhere one concatenates all the variables togetherin all four batteries together and treats the entireset as one large battery. The corresponding over-all fit is comparable to that derived from theMiles and Snow typology, but is also consider-ably inferior to the four-group NORMCLUS solu-tion. Thus, traditional clustering methodologies arenot substitutes for the NORMCLUS procedure inthis setting.

Characteristics of the four derived strategictypes

Table 2 indicates how each of the four derivedgroups correspond to the P-A-D-R Miles and Snowframework based on the 11-item Conant et al.(1990) strategic type questionnaire. Each respon-dent was classified as one of the four Miles andSnow strategic types (see procedure above in theMeasures section), and then assigned to eitherProspectors, Analyzers, Defenders, or Reactors.Table 2 shows the cross-tabulation relationshipsbetween the derived four-group solution and theMiles and Snow typology:

• Group 1 is composed of about 52 percent Pros-pectors and 32 percent Analyzers.

• Group 2 is about 55 percent Defenders, with theremaining 45 percent Reactors.



Table 2. Correspondence between empirically derived strategic types and Miles and Snow typology

Strategic types

Miles & Snow: Group 1 Group 2 Group 3 Group 4 Total

Prospectors (% in group) 120 (52.2%) 0 (0%) 27 (26.0%) 87 (45.8%) 234Analyzers (% in group) 73 (31.7%) 0 (0%) 45 (43.3%) 102 (53.7%) 220Defenders (% in group) 37 (16.1%) 102 (55.1%) 28 (26.9%) 1 (0.5%) 168Reactors (% in group) 0 (0%) 83 (44.9%) 4 (3.8%) 0 (0%) 87

Total 230 185 104 190 709

Chi-square tests:

Value d.f. Asymp. sig. (2-sided)

Pearson chi-square 535.543 9 0.000Likelihood ratio 648.056 9 0.000

• Group 3 is a hybrid of Analyzers (43%), Pros-pectors (26%), and Defenders (27%).

• Group 4 is split between Prospectors (46%) andAnalyzers (54%).

In sum, Groups 1 and 4 are primarily Prospec-tor /Analyzer groups, Group 2 is primarily a Defen-der /Reactor group, and Group 3 is a mix of Ana-lyzers, Prospectors, and Defenders. While the chi-square test indicates a clear dependence betweenthese two classifications, the derived four-groupsolution is not isomorphic or cleanly mapped intothe P-A-D-R framework.

The four derived strategic types and strategiccapabilities

Table 3 presents additional information (mean res-ponses) that allows us to characterize the fourderived strategic types in terms of their predomi-nant firm strategic capabilities. Looking at Table 3,Groups 2 and 4 have significantly higher mar-keting capability levels (i.e., customer and com-petitor knowledge, marketing activity integration,skill at segmenting and targeting, effective pric-ing, and advertising) than do the other two groups.Groups 2 and 4’s means on the marketing capa-bility scale items range from 3.97 to 4.67 (ona 10-point scale), while those of Group 1 rangefrom 1.29 to 2.19. By contrast, Group 1 has thehighest technology capabilities (NPD and technol-ogy development capabilities, manufacturing pro-cesses, predicting technological changes, and pro-duction facilities), followed by Groups 4, 2, and3 in that order. In general, Group 1’s means on

the technology capability scale items are around8, while those for Group 3 are about 1. Interest-ingly, the groups’ means on the information tech-nology scale items (IT systems for new products,for functional integration, for technical and marketknowledge creation, and for internal communica-tion) also occur in the same order: Group 1 hasthe highest IT capabilities (scale item means arealmost 9 on a 10-point scale), followed by Groups4, 2, and 3 in that order.

Groups 2 and 3’s market linking capabilities(market sensing, customer linking, channel bond-ing, customer retention, and supplier relationshipbuilding) are generally significantly higher thanthose of the other groups. Group 2’s scale itemmeans range from 2.28 to 3.15, while those ofGroup 3 range from 1.36 to 3.08; these are inalmost every case significantly higher than the cor-responding means for Groups 1 and 4. On the man-agement capability scale items (integrated logisticssystems, cost control capabilities, financial man-agement skills, human resource management capa-bilities, profit and revenue forecasting, and market-ing planning process), Group 4’s means were sig-nificantly higher (ranging from 6.17 to 7.22), andGroup 1’s means were significantly lower (rangingfrom 4.55 to 6.04).

Note the differences between the two groupsmade up primarily of Prospectors and Analyzers(Groups 1 and 4). Group 4 is highest among allgroups in both marketing and management capa-bilities, while Group 1 is the lowest. Group 1is highest among all groups in technology andIT capabilities, but Group 4 is second highest.Finally, in market linking capabilities, Group 4



Table 3. Means of strategic capability items for the empirically derived strategic types

Group 1 Group 2 Group 3 Group 4

Marketing capabilitiesKnowledge of customers 2.19 3.94 2.25 4.23Knowledge of competitors 1.29 3.97 2.17 4.19Integration of marketing activities 1.52 4.41 2.00 4.53Skill to segment and target markets 1.71 4.40 1.93 4.29Effectiveness of pricing programs 1.70 4.62 3.48 4.37Effectiveness of advertising programs 1.80 4.67 3.50 4.72Technology capabilitiesNPD capabilities 7.90 6.24 1.00 7.15Manufacturing processes 8.01 6.51 1.00 7.34Technology development capabilities 8.34 6.84 1.23 7.61Predicting technological changes 7.86 6.14 0.84 7.28Production facilities 8.26 6.66 1.35 7.51

Market-linking capabilitiesMarket-sensing capabilities 1.51 2.40 2.83 1.49Customer-linking capabilities 1.77 2.59 2.13 2.43Durable relationship with suppliers 0.92 2.28 2.41 1.49Ability to retain customers 1.44 2.34 1.36 1.39Channel-bonding capabilities 1.58 3.15 3.08 2.22

Information technology capabilitiesIT systems for NPD projects 9.45 7.52 6.19 8.63IT systems for functional integration 9.48 7.66 6.57 8.82IT systems for tech knowledge creation 8.99 7.77 5.68 8.56IT systems for mkt knowledge creation 8.70 7.13 6.01 8.19IT systems for internal communication 8.97 7.20 5.90 8.53

Management capabilitiesIntegrated logistics systems 6.04 6.70 5.99 7.22Cost control capabilities 4.55 5.79 6.57 6.17Financial management skills 5.29 6.52 7.54 7.06HR management capabilities 5.29 6.35 6.59 6.81Profitability and revenue forecasting 5.23 6.35 6.49 6.71Marketing planning process 5.14 6.42 6.86 6.81

For complete wording of questions, see Appendix 3.Interpretation: The highest mean for each scale item is highlighted in bold; the lowest mean for each scale item is underlined. (Ifthere are no significant differences between two or more means, all are highlighted.)

usually outscores Group 1 (though Group 2 is sig-nificantly higher than both). Thus, Groups 1 and4 are Prospector /Analyzer groups with differentprofiles: Group 4 is strong in marketing, manage-ment, technology, and IT. Group 1 relies on itscapabilities in technology and IT, and is in factthe weakest group in marketing, market linking,and management.

Group 2, the Defender /Reactor group, has rel-ative strengths in marketing and market linkingcapabilities, but is relatively weaker in technol-ogy and IT capabilities. Group 3, comprising Ana-lyzers, Defenders, and Prospectors, has relativestrength in market linking and management capa-bilities but is the weakest or among the weakest inall other capabilities.

The four strategic types and environmentaluncertainty

Table 4 provides additional insights by comparingthe four derived strategic types’ mean responseson the environmental uncertainty factors. Group 2firms face the most uncertain marketing, compet-itive, and technological environment, followed byGroup 4. Average scores on environment uncer-tainty scale items are in the range of 6.12 to6.59 (on a 10-point scale) for Group 2, and5.18 to 6.66 for Group 4, where a higher scoremean indicates greater perceived uncertainty. Bothother groups face relatively lower uncertainty intheir marketing, competitive, and technologicalenvironments.



Table 4. Means of environment items for the empirically derived strategic types

Group 1 Group 2 Group 3 Group 4

Market environmentPreferences change through time 3.97 6.38 4.32 5.18Customers look for new products 4.04 6.27 4.39 5.33Price relatively unimportant 4.07 6.25 4.26 5.28Product-related needs are different 3.97 6.34 4.35 5.37Cater to many of the same customers 3.53 6.59 3.90 6.06Difficult to predict marketplace changes 4.01 6.23 4.48 5.31

Competitive environmentCompetition is cutthroat 3.68 6.40 4.02 5.72Many ‘promotion wars’ in industry 3.74 6.33 3.95 5.90Competitors can match offers readily 3.80 6.24 4.03 5.78Price competition in industry 3.77 6.29 4.07 5.85New competitive moves every day 3.20 6.58 3.48 6.66Competitors are relatively weak 3.74 6.25 4.11 5.78

Technological environmentTechnology changing rapidly 4.12 6.28 3.92 5.49Tech change provides opportunities 4.10 6.32 3.69 5.71Difficult to forecast technology 4.16 6.12 3.80 5.64New product ideas from technology 4.06 6.18 3.77 5.69Tech developments are minor 3.55 6.52 3.16 6.47Technological changes are frequent 4.11 6.10 3.97 5.54

For complete wording of questions, see Appendix 3.Interpretation: The highest mean for each scale item is highlighted in bold; the lowest mean for each scale item is underlined.

Table 5. Means of performance items for the derived four-group solution

Performance measure Group 1 mean Group 2 mean Group 3 mean Group 4 mean Overall mean

PROF 29.13 7.21 3.20 33.12 20.68ROIPEC −1.66 19.52 0.06 16.80 9.07ROI 1.82 6.11 1.97 5.63 3.98ROA 1.65 5.90 1.70 5.96 3.92RMS 1.85 5.83 1.83 5.69 3.92CUSRET 2.27 6.42 1.73 6.13 4.31CUSRET2 2.21 6.41 1.97 6.32 4.37SALESGR 1.19 6.90 1.64 7.91 4.55PERF1 1.99 6.35 2.63 6.16 4.34PERF2 1.98 6.01 2.63 5.75 4.14PERF3 1.67 5.38 1.69 5.55 3.68

Legend: PROF = profit margin (i.e., total revenue- total variable costs)/total revenue; ROIPEC = average return on investment inthis business unit over the past 3 years (in %); ROI = return on investment; ROA = return on assets; RMS = relative market share;CUSRET = overall customer retention; CUSRET2 = retention of major customers; SALESGR = sales growth; PERF1 = overallprofit margin relative to the objective for this business unit; PERF2 = overall sales relative to the objective for this business unit;PERF3 = overall return on investment relative to the objective for this business unit.See Appendix 3 for measurement procedure for each of these performance measures.Interpretation: The highest mean for each scale item is highlighted in bold; the lowest mean for each scale item is underlined.

The four strategic types and performance

Table 5 presents the means for the 11 perfor-mance variables for the four derived strategictypes. It is evident from Table 5 that Groups 2and 4 are the highest-performing groups on almost

every performance variable. Average 3-year ROI(ROIPEC) for Groups 2 and 4 are 19.52 and16.80 respectively, while for Groups 1 and 3 thecomparable means are −1.66 and 0.06 (i.e., Group1 actually shows a negative 3-year ROI). Similarresults are obtained for 1-year ROI (Groups 2 and



4 = 6.11 and 5.63 respectively; Groups 1 and 3 =1.82 and 1.97 respectively), return on assets, rela-tive market shares, customer retention, retention ofmajor customers, sales growth, profit margin rel-ative to objective, sales relative to objective, andROI relative to objective. Group 1 showed strongprofit margin performance (means for Groups 4and 1 = 33.1 and 29.1 respectively, Groups 2 and3 = 7.2 and 3.2 respectively); Group 3 laggedbehind in performance on all measures.

The four strategic types cross-tabulated bycountry and industry

Finally, to further characterize the four derivedstrategic types, they were cross-tabulated bycountry (Table 6) and by industry (Table 7).As shown in Table 6, both Groups 2 and 4are relatively evenly balanced across the threecountries. Group 2 is 40.0 percent Japanesefirms, 23.2 percent U.S. firms, and 36.8 percent

Table 6. Cross-tabulation of the four derived groups by country

Country Group 1 Group 2 Group 3 Group 4 Overall

Japan (% in group) 110 (47.8%) 74 (40.0%) 0 (0%) 64 (33.7%) 248U.S. (% in group) 12 (5.2%) 43 (23.2%) 104 (100%) 57 (30.0%) 216China (% in group) 108 (47.0%) 68 (36.8%) 0 (0%) 69 (36.3%) 245

Total 230 185 104 190 709

Chi-square tests:



Table 7. Cross-tabulation of the four derived groups by industry

Industry Group 1 Group 2 Group 3 Group 4 Overall

1. Chemicals and RelatedProducts (% in group)

32 (13.9%) 33 (17.8%) 11 (10.6%) 22 (11.6%) 98

2. Electronic and ElectricalEquipment (% in group)

27 (11.7%) 21 (11.4%) 12 (11.5%) 20 (10.5%) 80

3. Pharmaceuticals, Drugs, andMedicines (% in group)

27 (11.7%) 4 (2.2%) 13 (12.5%) 23 (12.1%) 67

4. Industrial Machinery andEquipment (% in group)

19 (8.3%) 23 (12.4%) 6 (5.8%) 13 (6.8%) 61

5. TelecommunicationsEquipment (% in group)

23 (10.0%) 22 (11.9%) 15 (14.4%) 14 (7.4%) 74

6. Semiconductors andComputer-Related Products(% in group)

23 (10.0%) 22 (11.9%) 16 (15.4%) 18 (9.5%) 79

7. Instruments and RelatedProducts (% in group)

24 (10.4%) 20 (10.8%) 9 (8.7%) 26 (13.7%) 79

8. Others (% in group) 55 (23.9%) 40 (21.6%) 22 (21.2%) 54 (28.4%) 171

Total 230 185 104 190 709

Chi-square tests:





Chinese firms; comparable figures for Group4 are: 33.7 percent Japanese, 30.0 percent U.S.,and 36.3 percent Chinese. By contrast, Group1 is almost entirely composed of Asian firms(47.8% Japanese, 47.0% Chinese, 5.2% U.S.), andGroup 3 is 100 percent U.S. firms. As shownby the chi-square tests provided in Table 6, thedifferences across countries are significant.

Table 7, which shows cross-tabulations byindustry, has less obvious results. Considering thetwo top performing groups, Table 7 shows thatGroup 2 contained relatively more chemical firms,industrial machinery and equipment manufactur-ers, telecommunications equipment manufacturers,and semiconductor and computer-related products,while Group 4 had comparatively more pharma-ceutical and drug makers and instrument manufac-turers. Other than these observations, little of notewas obtained from the industry cross-tabulation.

DISCUSSION

The four groups in our solution bear some resem-blance to the familiar four-group (P-A-D-R) typol-ogy of Miles and Snow. We do, however, find dif-ferences across groups that are explainable in termsof the strategic capabilities, performance, and envi-ronmental factors which would have been other-wise obscured using the classic Miles and Snowtypology for this dataset. For example, Groups1 and 4 are both primarily Prospector /Analyzergroups, but differ in terms of marketing, manage-ment, and some market-linking capabilities. Group4 also faces more challenging environments. Simi-larly, Groups 2 and 3 have a bigger representationof Defenders, but Group 2 significantly outper-forms Group 3 in marketing, market linking, tech-nology, and IT capabilities, despite facing a morechallenging environment. Our results also suggestthat Analyzers do not necessarily constitute a sepa-rate group, but rather tend to be ‘like’ Prospectors(Groups 1, 3, and 4) or ‘like’ Defenders (Group 3).Hence, we believe our solution clarifies and com-plements the familiar Miles and Snow typology inthat the relationships between strategic types, capa-bilities, and environmental uncertainty are explic-itly included in the empirical classification schemederived. Whereas Miles and Snow found Prospec-tors, for example, we find Prospectors who arestronger in marketing, market linking, and man-agement, Prospectors who are significantly weaker

in marketing and market-linking capabilities, andAnalyzers that share characteristics with both typesof Prospectors. As noted above, we also found thatcertain types of environmental uncertainty wereassociated with strategic type selection as well.

At this point one can combine all the resultspresented in Tables 2–7 to compile descriptivecharacteristics of all four derived strategic groups:

• Group 1: Asian-based prospecting firms withtechnology strengths. These firms, comprisingmostly Miles and Snow Prospectors and Ana-lyzers, seek to maintain their competitive edgeby prospecting using their strengths in technol-ogy and IT capabilities. They possess relativeweaknesses in marketing, market linking, andmanagement, which would seem to limit theirability to respond quickly to market changes;however, they operate in relatively uncertainmarkets, competitive and technological environ-ments, which may mitigate the need for strengthin market linking. These firms are overwhelm-ingly Asian (about 95% total Japanese and Chi-nese), and tend to do well in 1-year profit per-formance, though not on any of the other per-formance measures.

• Group 2: Defensive firms with marketing skills.This group is slightly over one-half Defenders,and the remaining firms (about 45%) are Reac-tors. These firms stay competitive by defendingtheir established positions through superior mar-keting, market linking, and management capa-bilities. They are among the weakest in technol-ogy and IT capabilities, however. Though theyoperate in relatively uncertain markets, compet-itive and technological environments, they areamong the leaders on almost all performancemeasures. This group includes a mix of firmsfrom all three countries.

• Group 3: U.S.-based firms with market linkingand management strengths. This group is a mixof Miles and Snow types, including roughlyhalf analyzers and the rest approximately evenlysplit between Prospectors and Defenders. Nota-bly, all Group 3 firms are U.S.-based. Thesefirms are relatively strong in market linking andmanagement capabilities, but among the weak-est in marketing, technology, and IT. Theseweaknesses contribute to the uniformly lowperformance of this group on all performancemeasures, despite the fact that these firms face



relatively low market, competitive, and techno-logical uncertainties.

• Group 4: Balanced prospecting firms. This groupis about evenly split between Prospectors andAnalyzers and also includes a relatively evenmix of Japanese, Chinese, and U.S. firms. Thesefirms have the highest marketing and manage-ment capabilities and are also among the lead-ers in technology and IT capabilities (hence thename ‘balanced’), though they have a potentialweakness in market-linking capabilities. Thesefirms perform extremely well on all performancevariables, despite facing relatively high market,competitive, and technological uncertainties.

Here, we can make some overall observationsabout the four groups and how our findings corre-late with the expectations of the Miles and Snowmodel. There are two clear paths to higher per-formance, and both are pursued by U.S. as wellas Asian-based firms. Group 4 firms are Prospec-tors and Analyzers who have balanced market-ing, technology, IT, and management capabilities.The Miles and Snow model would suggest thatProspectors require strengths in technology andIT, yet also need to receive information on cus-tomer needs efficiently to drive product develop-ment: this would be consistent with the ‘balanced’set of capabilities possessed by Group 4 firms.Group 2 firms are Miles and Snow Defenders (andReactors that behave like Defenders). Miles andSnow would predict that these firms would par-ticularly require strategic capabilities in marketingand market linking for optimum performance; thisis indeed consistent with our findings for Group2. There were two paths to lower performanceand, interestingly, one was predominant amongAsian firms and one among U.S. firms. Asian low-performance firms (Group 1) tended to be strongin technology and IT but deficient in marketing,market linking, and management. One could inferthat they attempt to compete on the basis of theirtechnological strengths, but lack the marketing andmanagement skills to be successful. U.S. low-performance firms (Group 3) are in some ways anopposite to Group 1: their relative strengths lie inmarket linking and management, but their techno-logical and marketing capabilities are lacking. Inshort, both Groups 1 and 3 are deficient in severalof the capability sets, while Groups 2 and 4 showfew if any glaring capability weaknesses.

CONCLUSIONS

In this article, we revisit the long-established Milesand Snow (1978) strategic typology in the lightof recent theoretical discussions as to inclusion ofconceptually relevant variables as well as throughthe prism of a new classification methodology.The modified NORMCLUS constrained, multi-objective, classification procedure was utilized toderive a typology empirically from data pertainingto strategic type, strategic capabilities, environ-mental uncertainty, and performance. This method-ology is well suited to obtain a parsimonious andflexible grouping of entities in accordance withestablished empirical and statistical criteria. Therelative goodness-of-fit of the derived classifica-tion scheme is compared with that of the Milesand Snow grouping (used as a benchmark), andis found to more accurately capture the data asso-ciations and better explicate the interrelationshipsamong the variables.

Our framework augments the scope of the orig-inal Miles and Snow model by considering theroles of three batteries of variables: strategic firmcapabilities, environmental uncertainty, and per-formance. All of these have been discussed inpast research (Hambrick, 1983; Zajac and Shor-tell, 1989; Conant et al., 1990), but have neverbeen included together when deriving or testing atypology. In their original study, Miles and Snow(1978) suggested that there might be a complexframework of interrelationships among firm capa-bilities, environmental uncertainty, and strategy.However, they did not explicitly model the roleof environmental factors or strategic capabilitiesin the shaping of strategic types (Hambrick, 1983).The Miles and Snow model implies that it is theSBU’s strategic type that shapes its capabilities(i.e., Prospectors keep on prospecting). Hambrick(1983) suggested that a more complex relationshipamong all of these variables exists. Our results sug-gest that capabilities, and environmental factors, doin fact interrelate with strategic type, and under-standing this framework of interactions is impor-tant to managers in that it does have significantimpact on SBU performance.

As detailed in the above sections, we findthat including strategic capabilities, environmentaluncertainty, and performance results in a some-what different classification that varies from Milesand Snow in the compositions of the four derivedgroups. They do not, however, negate the Miles



and Snow grouping. In fact, the groups identifiedhere can be viewed as second-order derivativesof the pure and conceptually distinct P-A-D-Rgroups being considered as first-order ‘primitives.’Our study highlights a general point that strategictypes as empirically derived from field samplestend to be highly context-dependent and do notneatly fall into the tight Miles and Snow group-ings. As an obvious example, in our study, Groups1 and 4 both behave much like classic Miles andSnow Prospectors and Analyzers, but one grouphas superior capabilities in marketing and man-agement, and clearly dominates the other in termsof performance. The empirically derived strategictypes provide a more accurate representation ofstrategic behavior for the industries under consid-eration, and provide better insights into the dynam-ics of strategy with respect to how firms copewith environmental uncertainty using their avail-able capabilities (see Discussion section above).

We expect that in different contexts, differ-ent numbers and/or compositions of groups willemerge. For example, given a different set ofindustries, it is possible that five groups mightbe found, including two groups that are essen-tially Prospectors (but with different capabilitiesor facing different environments), and one eachof Analyzers, Defenders, and Reactors. Withoutincluding the capability and environment variablesin the analysis, one would miss the complex inter-relationships between these variables and strate-gic type, and the performance implications. Thederived strategic types capture the context-specificconditions that shape strategic decisions within agiven set of industries, and therefore add a layerof understanding on top of the classic Miles andSnow strategic types. Our findings suggest thatSBU managers need to consider both environmentand capability when developing strategy, as thereis a clear relationship between these batteries ofvariables and SBU performance.

In addition, we would also expect empiricallyderived strategic types would differ if alterna-tive classification methods were utilized in such acontingency-based approach. We recommend thefurther use of the NORMCLUS methodology forderiving the strategic types as it provides con-siderably more sophistication and flexibility thanraw conceptualization or standard cluster analy-sis in providing the ability to embed applicationconstraints reflecting a priori knowledge, havinga heuristic to test for the appropriate number

of clusters, and building a framework to com-pare alternative classifications with respect to thesame objective function and input variables. Withthese advantages, NORMCLUS is ideally suited toincorporating all the batteries of variables requiredin our capability-environment framework, includ-ing performance outcomes. We have demonstratedthe superiority of NORMCLUS over traditionalKMEANS cluster analysis for this application.

This study can be extended in a number ofways. Further studies should seek to validate thestrategic typology using data sets obtained from adifferent set of industries using the NORMCLUSmethodology as described above. It is expectedthat the derived strategic types are very context-dependent, so different groups might be found indifferent contexts. However, it would be intriguingto determine whether the groups derived from dif-ferent industry settings bear some resemblance tothe original Miles and Snow strategic types. Ourderived strategic types can be easily understood asvariants of classic Miles and Snow types, viewedwithin the context of the capability–environmentframework. If empirically derived strategy types inother settings are also second-order derivatives ofthe Miles and Snow types (even if different groupsare obtained), it supports our belief that the Milesand Snow strategic typology, viewed in conjunc-tion with the capability–environment framework,is a powerful model of strategic behavior with realimplications for SBU performance.

Future research can also extend our cross-cultu-ral findings. We found four strategic groups acrossthe United States, Japan, and China, and deter-mined that the two strategic types that led tohighest performance contained firms from all threecountries. Thus, we have some evidence that man-agers from these three countries ‘think alike’ whenreacting to similar capability and environmentalsettings. These two top-performing groups alsoresemble classic Miles and Snow groups: Group2 resembles Defenders, while Group 4 is a combi-nation of Prospectors and Analyzers. We thereforehave some preliminary evidence that the Miles andSnow model generalizes across the United States,Japan, and China: managers from top perform-ing firms in those three countries choose strate-gies not too different from those of the Miles andSnow model. It is unknown, however, whether thisobservation holds true in industries other than theones we studied, or in other countries. Possibly,the business environment in other countries might



be so different as to make the Miles and Snowmodel inapplicable. Future studies could develophypotheses regarding business or cultural environ-mental conditions in other countries that wouldaffect strategic choice, and determine whether thebest-performing firms in those countries choosestrategies that resemble the Miles and Snow strate-gic typology. It would be worthwhile to determinewhether firms in different business environments(the European Union, for example, or emergingmarkets such as Eastern Europe or Southeast Asia)employ significantly different strategic types. Afertile area for future research here would beto conduct separate NORMCLUS analyses withineach country and then compare the results acrosscountries using the Miles and Snow typology asa benchmark. In addition, estimating overlappingclusters or strategic types in comparison with tra-ditional partitions in NORMCLUS would prove tobe an interesting direction.

Finally, given the cross-sectional nature of thisempirical study, it was not possible to observe theeffects of strategic adjustments on SBU perfor-mance through time. Much insight on the interre-lationships among the variables would be obtainedby conducting a longitudinal analysis. One couldobserve SBUs through time, examining how theirenvironment evolves, how their capabilities devel-op, and whether adjustments in strategy yield pay-offs in terms of long-run performance.2

ACKNOWLEDGEMENTS

The authors wish to acknowledge the financialsupport for data collection provided by the Facultyof Technology Management, Eindhoven Universityof Technology, The Netherlands.

REFERENCES

Bharadwaj AS, Bharadwaj SG, Konsynski BR. 1999.Information technology effects on firm performance asmeasured by Tobin’s Q. Management Science 45(7):1008–1024.

Churchill GA Jr. 1979. A paradigm for developingbetter measures of marketing constructs. Journal ofMarketing Research 16: 64–73.

Cohen JA. 1960. A coefficient of agreement for nominalscales. Educational and Psychological Measurement20: 37–46.

2 We thank the reviewers for this excellent suggestion.

Conant JS, Mokwa MP, Varadarajan PR. 1990. Strategictypes, distinctive marketing competencies and orga-nizational performance: a multiple-measures-basedstudy. Strategic Management Journal 11(5): 365–383.

Covin JG, Slevin DP. 1989. Strategic management ofsmall firms in hostile and benign environments.Strategic Management Journal 10(1): 75–87.

Davis FD. 1986. A technology acceptance modelfor empirically testing new end user informationsystems: theory and results. Doctoral dissertation,Massachusetts Institute of Technology.

Davis FD. 1989. Perceived usefulness, perceived ease ofuse, and user acceptance of information technology.MIS Quarterly 13(3): 319–339.

Day GS. 1990. Market Driven Strategy: Processes forCreating Value. Free Press: New York.

Day GS. 1994. The capabilities of market-drivenorganizations. Journal of Marketing 58(4): 37–52.

DeSarbo WS, Cron W. 1988. A conditional mixturemaximum likelihood methodology for clusterwiselinear regression. Journal of Classification 5:249–289.

DeSarbo WS, Grisaffe D. 1998. Combinatorial optimiza-tion approaches to constrained market segmentation:an application to industrial market segmentation. Mar-keting Letters 9: 115–134.

DeSarbo WS, Mahajan V. 1984. Constrained classifica-tion: the use of a priori information in cluster analysis.Psychometrika 49: 187–215.

DeSarbo WS, Oliver R, Rangaswamy A. 1989. Asimulated annealing methodology for cluster-wiselinear regression. Psychometrika 54: 707–736.

DeSarbo WS, Di Benedetto AC, Jedidi K, Song XM.2004. A constrained finite mixture structural equationmethodology for empirically deriving strategic types.Working paper, Marketing Department, PennsylvaniaState University.

Dillman DA. 1978. Mail and Telephone Surveys: TheTotal Design Method . Wiley: New York.

Dyer B, Song XM. 1997. The impact of strategy onconflict: a cross-national comparative study of U.S.and Japanese firms. Journal of International BusinessStudies 28(3): 467–493.

Griffin A, Hauser JR. 1996. Integrating R&D andmarketing: a review and analysis of the literature.Journal of Product Innovation Management 13(2):191–215.

Hambrick DC. 1983. Some tests of the effectiveness andfunctional attributes of Miles and Snow’s strategictypes. Academy of Management Journal 26(March):5–25.

Hambrick DC. 1984. Taxonomic approaches to studyingstrategy: some conceptual and methodological issues.Journal of Management 10(1): 27–41.

Hambrick DS. 2003. On the staying power of defenders,analyzers, and prospectors. Working paper, Pennsyl-vania State University.

Hofer CW, Schendel D. 1978. Strategy Formulation:Analytical Concepts. West: St Paul, MN.

Hrebiniak LG, Snow CC. 1980. Industry differencesin environmental uncertainty and organizational



characteristics related to uncertainty. Academy ofManagement Journal 23(4): 750–759.

Jarvenpaa S. 1989. The effect of task demands andgraphical format on information processing strategies.Management Science 35(3): 285–303.

Lawrence PR, Lorsch JW. 1967. Organization andEnvironment . Irwin: Homewood, IL.

Lawson CL, Hanson RJ. 1972. Solving Least SquaresProblems . Prentice-Hall: Englewood Cliffs, NJ.

Lin S, Kernighan M. 1973. An effective heuristicalgorithm for the traveling salesman problem.Operations Research 21: 498–516.

McDaniel SW, Kolari JW. 1987. Marketing strategyimplications of the Miles and Snow strategic typology.Journal of Marketing 51(4): 19–30.

McKee DL, Varadarajan PR, Pride WM. 1989. Strategicadaptability and firm performance: a market-contingent perspective. Journal of Marketing 53(3):21–35.

Miles R, Snow C. 1978. Organizational Strategy, Struc-ture, and Process . McGraw-Hill: New York.

Miller D, Friesen PH. 1978. Archetypes of strategyformulation. Management Science 24(11): 921–933.

Miller D, Friesen PH. 1983. Strategy-making and envi-ronment: the third link. Strategic Management Journal4(3): 221–225.

Moenaert RK, Souder WE. 1996. Context and antece-dents of information utility at the R&D/marketinginterface. Management Science 42(11): 1592–1610.

Moorman C. 1995. Organizational market informationprocesses: cultural antecedents and new productoutcomes. Journal of Marketing Research 32(3):318–335.

Narver JC, Slater SF. 1990. The effect of a marketorientation on business profitability. Journal ofMarketing 54(1): 20–35.

Porter ME. 1980. Competitive Strategy . Free Press: NewYork.

Rao SS. 1996. Engineering Optimization, Theory andPractice (3rd edn). Wiley: New York.

Robinson WT, Fornell C, Sullivan MW. 1992. Aremarket pioneers intrinsically stronger than laterentrants? Strategic Management Journal 13(8):609–624.

Ruekert RW, Walker OC Jr. 1987. Interactions betweenmarketing and R&D departments in implementingdifferent business strategies. Strategic ManagementJournal 8(3): 233–248.

Shortell SM, Zajac EJ. 1990. Perceptual and archivalmeasures of Miles and Snow’s strategic types: acomprehensive assessment of reliability and validity.Academy of Management Journal 33(4): 817–832.

Snow CC, Hambrick DC. 1980. Measuring organiza-tional strategies. Academy of Management Review 5:527–538.

Song XM, Parry ME. 1996. What separates Japanesenew product winners from losers? Journal of ProductInnovation Management 13(5): 422–439.

Song XM, Parry ME. 1997a. A cross-national compar-ative study of new product development processes:Japan and the United States. Journal of Marketing61(2): 1–18.

Song XM, Parry ME. 1997b. The determinants ofJapanese new product successes. Journal of MarketingResearch 34(1): 64–76.

Swanson EB. 1994. Information systems innovationsamong organizations. Management Science 40(9):1069–1092.

Utterback JM. 1979. Environmental analysis and fore-casting. In Strategic Management: A New View of Busi-ness Policy and Planning , Schendel DE, Hofer CW(eds). Little, Brown: Boston, MA; 134–144.

Walker OC Jr, Boyd HW Jr, Mullins J, Larreche J-C.2003. Marketing Strategy: Planning and Implementa-tion (4th edn). Irwin/McGraw-Hill: Homewood, IL.

Webster FE. 1992. The changing role of marketing in thecorporation. Journal of Marketing 56(October): 1–17.

World Bank. 2000. World Development Report 2000/2001 .

Zajac E, Shortell SM. 1989. Changing generic strategies:likelihood, direction, and performance implications.Strategic Management Journal 10(5): 413–430.

APPENDIX 1: NORMCLUSCONSTRAINT IMPLEMENTATIONOPTIONS

The parameter vector θ can include cluster mem-bership information (M), as well as other param-eters (e.g., cluster level regression coefficients(B) as in a cluster-wise regression framework).NORMCLUS can accommodate ordinary clusteranalysis where, for example, f1 (θ) can be speci-fied as a ratio of between- to within-cluster sum-of-squares to be maximized with respect to binaryθ = M indicating cluster membership. Alterna-tively, θ can include both cluster membership andcluster level regression parameters as in a cluster-wise regression approach (cf. DeSarbo et al., 1989;DeSarbo and Grisaffe, 1998; DeSarbo and Cron,1988), where f1(θ) can be specified as a resid-ual sum-of-squares to be minimized. Or, in nor-mative classification applications where costs andrevenues are readily available, f1(θ) can be anexpected profit function to be maximized. Again, avariety of optimization frameworks are accommo-dated in NORMCLUS depending upon the natureof the classification application at hand.

The remaining flexibility in NORMCLUS can bebest illustrated in terms of the user-specified con-straints that can be accommodated in Equations 2and 3 in the text. Let:

i = 1, . . ., I firms/SBUs;k = 1, . . ., K variables;r = 1, . . ., R clusters (R is user specified);



Xik = the value of the k-th variable or charac-teristic for firm i;

mir = the degree of membership of firm i in clu-ster r (0 ≤ mir ≤ 1);

Sr = the set of firms in cluster r;Ir = the number of firms (cardinality) in cluster

r;

X(r)

k = the mean of variable k in cluster r .

Then, for several types of classification applica-tions, the following section discusses several typesof possible constraints representing prior infor-mation specified by the user or institutional con-straints which can be addressed (cf. DeSarbo andMahajan, 1984; DeSarbo and Grisaffe, 1998).

Types of cluster

(a) mir(1 − mir) = 0 ∀ i = 1 . . . I,

∀ r = 1 . . . R (A1)

This set of constraints restricts mir to be either 0or 1.

(b)R∑

r=1

mir = 1 ∀ i = 1 . . . I (A2)

This set of constraints, together with (a) above,provides for a non-overlapping cluster analysiswhere each firm can belong to one and only onecluster. Note that without this set of constraints(i.e., only with this (b) set), one can allow for over-lapping clusters—that is, allow for cases wherefirms can belong to more than one cluster.

(c) 0 ≤ mir ≤ 1 ∀ i = 1 . . . I,

∀ r = 1 . . . R (A3)

This set of constraints, together with those in(b) above, allow for ‘fuzzy-set’ clusters, whereobjects can be fractional members of all clusters.

(d)R∑

r=1

mir

≤=≥ci (A4)

This constraint, together with constraints in (a),restrict the number of clusters (ci) firm i canbelong to in an overlapping scheme.

Constraints concerning cluster membership

We assume that the constraints in Equation A1hold in the following discussion.

(a) msr∗ + mnr∗ = 2 (A5)

Here, one wants firms s and n to belong to thesame cluster r∗.

(b)∑i∈Tr∗

mir∗ = cr∗ (A6)

This is a generalization of constraint (a) above inthat one wants the firms in some set Tr∗, whosecardinality is cr∗, in the same cluster r∗.

(c) msr + mnr ≤ 1 ∀ r = 1, . . . , R (A7)

This constraint forbids firms s and n to be in thesame cluster.

(d) (1)

I∑i=1

mir ≥ Minr

(2)

I∑i=1

mir ≤ Maxr (A8)

These constraints allow one to restrict the numberof firms that get allocated to cluster r . Constraint(d1) states that the number of members in clusterr is to be greater than or equal to some minimumnumber Minr . Conversely, constraint (d2) restrictsmembership to be equal or less than some maxi-mum number Maxr .

(e)I∑

i=1

mir ′ =I∑

i=1

mir ∀ r �= r ′ = 1 . . . R

(A9)

This set of constraints restricts the number of firmsin any clusters r and r ′ to be equal (e.g., forperformance equalization).

(f)

∣∣∣∣∣I∑

i=1

mir −I∑

i=1

mir ′

∣∣∣∣∣ ≤ εI

∀ r �= r ′ = 1 . . . R (A10)

These constraints restrict the range or distributionof acceptable differences (ε1) in the number offirms in clusters r and r ′. This set of constraints



is basically equivalent to specifying both sets ofconstraints in (d1) and (d2) above where all theceiling values (Maxr ) and all floor values (Minr)are identical for all clusters.

Characteristics of clusters

(a) Xikmir ≥ V minkr ∀ i ∈ Sr (A11)

These constraints guarantee that all members ofcluster r possess at least V min

kr of characteristic orvariable k. Similarly, one could generalize con-straint to:

Ir∑i∈Sr

Xikmir

Ir

≤ V minkr (A12)

where the average cluster value on variable k mustbe greater than some minimum value. Similar con-straints can be constructed to insure that eachmember or cluster average be less than some max-imum value V max

kr by substituting ‘≤’ and ‘V maxkr ’

for ‘≥’ and ‘V minkr ’ respectively above. These con-

straints can insure a stable amount of homogeneityamongst the derived clusters.

(b)

∣∣∣∣∣∣Ir∑

i∈Sr ′

Xikmir −Ir ′∑

i∈Sr

Xijmir ′

∣∣∣∣∣∣≤=≥ε2 (A13)

This constraint establishes a range or distributionof acceptable differences (ε2) of characteristic k

in clusters r �= r ′ . Similarly, one can generalizethis to:∣∣∣∣∣∣∣∣∣∣∣∣

Ir∑i∈Sr

Xikmir

Ir

−

I ′r∑

i∈Sr ′

Xijmir ′

Ir ′

∣∣∣∣∣∣∣∣∣∣∣∣≤=≥ε2 (A14)

where the range of differences in mean values ofcharacteristic k in clusters r �= r′ is constrained.For example, this constraint can be gainfully uti-lized for insuring clusters will differ as to perfor-mance.

(c) |mjrmir (Xjk − Xik)| ≤ tmaxrk ∀ j, i ∈ Sr

(A15)

This set of constraints restricts the maximum devi-ation allowed (tmax

rk ) on characteristic k for any twomembers of the same cluster r . Accordingly onecould also constrain the maximum distance or dis-similarity allowed (Dr) between any two objectsin cluster r via:[

K∑k=1

(mjrmlr(Xjk − Xlk))2

]1/2

≤ Dr (A16)

(d)∣∣∣mjr(Xjk − X

(r)

k )∣∣∣ ≤ γ max

kr ∀ j ∈ Sr (A17)

This set of constraints restricts the maximum devi-ation (γ max

kr ) on characteristic k between any objectin cluster r and cluster r’s mean value on vari-able k. Similarly, one could generalize this to allvariables via:[

K∑k=1

(mjr(Xjk − X(r)

k ))2

]1/2

≤ �, ∀ j ∈ Sr

(A18)

where there is a restriction placed on the maximumdistance or dissimilarity allowed between any firmj in cluster r and the centroid of cluster r . The con-straints in sets (a) through (d) impose restrictionsthat affect the ‘compactness’ of a cluster, or thewithin sum-of-squares of a cluster. For example,such constraints can be used in a geographical-based classification scheme.

(e) |X(r)

k − X(r ′)k | ≥ Bmin

rr ′ (A19)

This constraint restricts the ‘separability’ (affectingthe between sums of squares) between the meanof variable k in cluster r and r′. This can begeneralized to the case involving all variables via:

K∑k=1

(X(r)

k − X(r ′)k )2 ≥ C min

rr ′ (A20)

where restrictions are made on the between sumsof squares between clusters r and r ′.

These and other application-specific constraintsare discussed in more detail in DeSarbo and Maha-jan (1984) and DeSarbo and Grisaffe (1998).

APPENDIX 2: ESTIMATIONALGORITHM DETAILS

For the strategic type application discussed in theEmpirical Results section, we attempt to estimate



the cluster membership indicators M = ((mir )) inorder to maximize:

� = α1f1 + α2f2 + α3f3 + α4f4, (B1)

where f1, f2, f3, and f4 are defined in Equations 5,6, 7, and 8 respectively, 0 < α < 1 is user spec-ified (as is R, the number of clusters), and con-straints A1, A2, and A8 are enforced. For thisparticular application, a modified lambda-opt com-binatorial optimization procedure (cf. Lin andKernighan, 1973) is devised. The general steps areas follows:

A. Set J = 0; select n from (1, 2, . . ., I ).Set the maximum number of iterations(MAXIT); generate a random map of thesequence 1 . . . I , indicating the order in whichcustomer cluster memberships are altered. Eval-uate �, and let �∗ = �.

B. For these n customers, change their clustermemberships randomly (i.e., alter n row vec-tors in M = ((mir )) and check for feasibility ofall constraints. Iterate until feasibility is main-tained.

Note, for such problems containing a cluster-wiseregression component, a next step would be toestimate the cluster-level betas. For such prob-lems where a need exists to enforce positivityconstraints, a constrained optimizer must be uti-lized in each of the R least-squares. Here, weutilize a modification of the Lawson and Han-son (1972) procedure which follows directly fromthe Kuhn–Tucker conditions for constrained min-imization. For a given r = 1, . . . , R, define:

hr = (hri ) = M

1/2ir · yi (B2)

Er = ((Er

ik

)) = M1/2ir · Xik (B3)

We can then reformulate this estimation problemin terms of r non-negative least-squares prob-lems: Minimize ||Erbr − hr || subject to br ≥ 0,for r = 1, . . ., R, (excluding such constraints onintercepts), which trivially can be shown to condi-tionally (holding M fixed) optimize B1. The algo-rithm, briefly outlined below, follows directly fromthe Kuhn–Tucker conditions for constrained mini-mization. For a given r , we form the I × K matrixof ‘independent variables,’ Er , and the I × 1 vec-tor (acting as the dependent variable) hr . In the

description below, the K × 1 vectors wr and zr

provide working spaces. Index sets Pr and Zr willbe defined and modified in the course of executionof the algorithm. Parameters indexed in the set Zr

will be held at the value zero. Parameters indexedin the set Pr will be free to take values greater thanzero. If a parameter takes a non-positive value, thealgorithm will either move the parameter to a pos-itive value or else set the parameter to zero andmove its index from set Pr to set Zr . On termina-tion, br will be the solution vector and wr will bethe dual vector.

1. Set Pr := Null, Zr : = {1, . . . , K}, and br : = 0.2. Compute the vector wr : = E′

r (hr − Erbr ).3. If the set Zr is empty or if wrj ≤ 0 for all

k ∈ Zr , go to Step 12.4. Find an index a ∈ Zr such that wra = max{wrk:

k ∈ Zr}.5. Move the index a from set Zr to set Pr .6. Let E(r)

P : = denote the I × K matrix definedby

Column k of E(r)p

:=

∣∣∣∣∣∣∣column k of Er if k ∈ Pr

0 if k ∈ Zr

Compute the vector zr as a solution of theleast-squares problem E(r)

P zr∼= hr . Note that

only the components zrk, k ∈ Pr , are deter-mined by this problem. Define zrk = 0 fork ∈ Zr .

7. If zrk > 0 for all k ∈ Pr , set br : = zr and goto Step 2.

8. Find an index v ∈ Pr such that brv/(brv −zrv) = min{brk/brk − zrk) : zrk ≤ 0, k ∈ Pr}.

9. Set Qr := brv/(brv − zrv).10. Set br : = br + Qr(zr − br ).11. Move from set Pr to set Zr all indices k ∈ Pr

for which φrk = 0. Go to Step 6.12. Next r .

On termination, the solution vector br satisfies:

brk > 0, k ∈ Pr (B4)

and

brk = 0, k ∈ Zr (B5)



and is a solution vector to the constrained least-squares problem:

E(r)

P br∼= hr (B6)

The dual vector wr satisfies:

wrk = 0, k ∈ Pr (B7)

and

wrk ≤ 0, k ∈ Zr (B8)

where:

wr = E′r (hr − Erbr ) (B9)

Equations B4, B5, B7, B8, and B9 constitute theKuhn–Tucker conditions characterizing a solutionvector br for this constrained least-squares prob-lem. Equation B6 is a consequence of B5, B7, andB9. These 12 steps are then repeated for the nextvalue of r = 1, . . ., R.

C. Set J = J + 1;D. Evaluate � in trying to improve. If there is

improvement, set � = �∗, store the M andB that resulted in that solution, and go tostep B. If no improvement, return to previousM, B, and 8∗ values and return to step B,unless J > MAXIT, in which case output bestsolution.

APPENDIX 3: ROADMAP OF SCALE,MEASUREMENT ITEMS, ANDSOURCES

I. Miles and Snow typology items (adaptedfrom Conant et al., 1990)

The following statements describe some character-istics of this selected strategic business unit/divi-sion. Please circle the description that best descri-bes this selected business unit.

1. In comparison to our competitors, the prod-ucts which we provide to our customers arebest described as: (Entrepreneurial—productmarket domain)a. Products that are more innovative, and con-

tinually changing.b. Products that are fairly stable in certain

markets while innovative in other markets.

c. Products that are stable and consistentlydefined throughout the market.

d. Products that are in a state of transition, andlargely respond to opportunities and threatsin the marketplace.

2. In contrast to our competitors, we have animage in the marketplace that: (Entrepreneuri-al—success posture)a. Offers fewer, select products which are

high in quality.b. Adopts new ideas and innovations, but only

after careful analysis.c. Reacts to opportunities or threats in the

marketplace to maintain or enhance ourposition.

d. Has a reputation for being innovative andcreative.

3. The amount of time our business unit spendson monitoring changes and trends in the mar-ketplace can best be described as: (Entrepren-eurial—surveillance)a. Lengthy: We are continuously monitoring

the marketplace.b. Minimal: We really don’t spend much time

monitoring the marketplace.c. Average: We spend a reasonable amount of

time monitoring the marketplace.d. Sporadic: We sometimes spend a great deal

of time and at other times spend little timemonitoring the marketplace.

4. In comparison to our competitors, the increasesor losses in demand that we have experiencedare due most probably to: (Entrepreneurial—growth)a. Our practice of concentrating on more fully

developing those markets which we cur-rently serve.

b. Our practice of responding to the pressuresof the marketplace by taking few risks.

c. Our practice of aggressively entering intonew markets with new types of products.

d. Our practice of assertively penetrating moredeeply into markets we currently serve,while adopting new products after a verycareful review of their potential.

5. One of the most important goals in these busi-ness units in comparison to our competitors isour dedication and commitment to: (Engineer-ing—technological goal)a. Keep our costs under control.b. Analyze our costs and revenues carefully,



to keep costs under control and to selec-tively generate new products or enter newmarkets.

c. Insure that the people, resources and equip-ment required to develop new products andnew markets are available and accessible.

d. Make sure we guard against critical threatsby taking any action necessary.

6. In contrast to our competitors, the competen-cies (skills) which our managerial employeespossess can best be characterized as: (Engi-neering—technological breadth)a. Analytical: their skills enable them to both

identify trends and then develop new prod-ucts or markets.

b. Specialized: their skills are concentratedinto one, or a few, specific areas.

c. Broad and entrepreneurial: their skills arediverse, flexible, and enable change to becreated.

d. Fluid: their skills are related to the near-term demands of the marketplace.

7. The one thing that protects us from its com-petitors is that we: (Engineering—technologi-cal buffers)a. Are able to carefully analyze emerging

trends and adopt only those which haveproven potential.

b. Are able to do a limited number of thingsexceptionally well.

c. Are able to respond to trends even thoughthey may possess only moderate potentialas they arise.

d. Are able to consistently develop new prod-ucts and new markets.

8. More so than many of our competitors, ourmanagement staff in this business unit tendsto concentrate on: (Administrative—dominantcoalition)a. Maintaining a secure financial position

through cost and quality control.b. Analyzing opportunities in the marketplace

and selecting only those opportunities withproven potential, while protecting a securefinancial position.

c. Activities or business functions which mostneed attention given the opportunities orproblems we currently confront.

d. Developing new products and expandinginto new markets or market segments.

9. In contrast to many of our competitors, thisbusiness unit prepares for the future by: (Ad-ministrative—planning)a. Identifying the best possible solutions to

those problems or challenges which requireimmediate attention.

b. Identifying trends and opportunities in themarketplace which can result in the cre-ation of product offerings which are newto the industry or reach new markets.

c. Identifying those problems which, if sol-ved, will maintain and then improve ourcurrent product offerings and market posi-tion.

d. Identifying those trends in the industrywhich our competitors have proven pos-sess long-term potential while also solv-ing problems related to our current productofferings and our current customers’ needs.

10. In comparison to our competitors, our orga-nization structure is: (Administrative—struc-ture)a. Functional in nature (i.e., organized by

department—marketing, accounting, per-sonnel, etc.).

b. Product or market oriented.c. Primarily functional (departmental) in

nature; however, a product- or market-oriented structure does exist in newer orlarger product offering areas.

d. Continually changing to enable us to meetopportunities and solve problems as theyarise.

11. Unlike our competitors, the procedures we useto evaluate performance are best described as:a. Decentralized and participatory encourag-

ing many organizational members to beinvolved.

b. Heavily oriented toward those reportingrequirements which demand immediate at-tention.

c. Highly centralized and primarily the re-sponsibility of senior management.

d. Centralized in more established productareas and more participatory in new prod-uct areas.



II. Strategic capability items

The following is a set of possible business strategic capabilities. Please evaluate how well or poorlyyou believe that this selected business unit performs the specific capabilities relative to your three majorcompetitors. Please use the following response scale: 0 = Much worse than the top three major competitorsin the industry; 10 = Much better than the top three major competitors in the industry.

Much Muchworse better

Marketing capabilitiesKnowledge of customers 0 1 2 3 4 5 6 7 8 9 10Knowledge of competitors 0 1 2 3 4 5 6 7 8 9 10Integration of marketing activities 0 1 2 3 4 5 6 7 8 9 10Skill to segment and target markets 0 1 2 3 4 5 6 7 8 9 10Effectiveness of pricing programs 0 1 2 3 4 5 6 7 8 9 10Effectiveness of advertising programs 0 1 2 3 4 5 6 7 8 9 10

Market linking capabilitiesMarket sensing capabilities 0 1 2 3 4 5 6 7 8 9 10Customer-linking (i.e., creating and managing

durable customer relationships) capabilities0 1 2 3 4 5 6 7 8 9 10

Capabilities of creating durable relationship withour suppliers

0 1 2 3 4 5 6 7 8 9 10

Ability to retain customers 0 1 2 3 4 5 6 7 8 9 10Channel-bonding capabilities (creating durable

relationship with channel members such aswhole sellers, retailers, etc.)

0 1 2 3 4 5 6 7 8 9 10

Relationships with channel members 0 1 2 3 4 5 6 7 8 9 10

Information technology capabilitiesInformation technology systems for new product

development projects0 1 2 3 4 5 6 7 8 9 10

Information technology systems for facilitatingcross-functional integration

0 1 2 3 4 5 6 7 8 9 10

Information technology systems for facilitatingtechnology knowledge creation

0 1 2 3 4 5 6 7 8 9 10

Information technology systems for facilitatingmarket knowledge creation

0 1 2 3 4 5 6 7 8 9 10

Information technology systems for internalcommunication (e.g., across differentdepartments, across different levels of theorganization, etc.)

0 1 2 3 4 5 6 7 8 9 10

Information technology systems for externalcommunication (e.g., suppliers, customers,channel members, etc.)

0 1 2 3 4 5 6 7 8 9 10

Technology capabilitiesNew product development capabilities 0 1 2 3 4 5 6 7 8 9 10Manufacturing processes 0 1 2 3 4 5 6 7 8 9 10Technology development capabilities 0 1 2 3 4 5 6 7 8 9 10Ability of predicting technological changes in the

industry0 1 2 3 4 5 6 7 8 9 10

Production facilities 0 1 2 3 4 5 6 7 8 9 10Quality control skills 0 1 2 3 4 5 6 7 8 9 10

Management capabilitiesIntegrated logistics systems 0 1 2 3 4 5 6 7 8 9 10Cost control capabilities 0 1 2 3 4 5 6 7 8 9 10Financial management skills 0 1 2 3 4 5 6 7 8 9 10Human resource management capabilities 0 1 2 3 4 5 6 7 8 9 10Accuracy of profitability and revenue forecasting 0 1 2 3 4 5 6 7 8 9 10Marketing planning process 0 1 2 3 4 5 6 7 8 9 10



III. Environmental uncertainty items

In general, how much do you disagree or agree with each of the following statements characterizing thebusiness environment or conditions in the primary markets your SBU currently serves? Please indicatethe degree to which you agree or disagree with the following statement regarding this selected businessunit (anchors: 0 = strongly disagree/10 = strongly agree)

Strongly Stronglydisagree agree

Market environmentIn our kind of business, customers’ product

preferences change quite a bit over time.0 1 2 3 4 5 6 7 8 9 10

Our customers tend to look for newproducts all the time.

0 1 2 3 4 5 6 7 8 9 10

Sometimes our customers are veryprice-sensitive, but on other occasions,price is relatively unimportant.

0 1 2 3 4 5 6 7 8 9 10

New customers tend to have product-relatedneeds that are different from those of ourexisting customers.

0 1 2 3 4 5 6 7 8 9 10

We cater to many of the same customersthat we used to in the past.

0 1 2 3 4 5 6 7 8 9 10

It is very difficult to predict any changes inthis marketplace.

0 1 2 3 4 5 6 7 8 9 10

Technological environmentThe technology in our industry is changing

rapidly.0 1 2 3 4 5 6 7 8 9 10

Technological changes provide bigopportunities in our industry.

0 1 2 3 4 5 6 7 8 9 10

It is very difficult to forecast where thetechnology in our industry will be in thenext two to three years.

0 1 2 3 4 5 6 7 8 9 10

A large number of new product ideas havebeen made possible through technologicalbreakthroughs in our industry.

0 1 2 3 4 5 6 7 8 9 10

Technological developments in our industryare rather minor.

0 1 2 3 4 5 6 7 8 9 10

The technological changes in this industryare frequent.

0 1 2 3 4 5 6 7 8 9 10

Competitive environmentCompetition in our industry is cutthroat. 0 1 2 3 4 5 6 7 8 9 10There are many ‘promotion wars’ in our

industry.0 1 2 3 4 5 6 7 8 9 10

Anything that one competitor can offer,others can match readily.

0 1 2 3 4 5 6 7 8 9 10

Price competition is a hallmark of ourindustry.

0 1 2 3 4 5 6 7 8 9 10

One hears of a new competitive movealmost every day.

0 1 2 3 4 5 6 7 8 9 10

Our competitors are relatively weak. 0 1 2 3 4 5 6 7 8 9 10



IV. Performance measures

PROFIT = gross margin (i.e., total revenue − total variable costs)/total revenue)ROIPEC = the average return on investment in this business unit over the past 3 years (in %)

Customer retention (CUSRET), sales growth (SALESGR), relative market share (RMS), and return onassets (ROA) (adapted from Naver and Slater, 1990)

Please rate how well this business unit has performed relative to all other competitors in the principalserved market segment over the past year.

0 1 2 3 4 5 6 7 8 9 100% 1–10% 11–20% 21–30% 31–40% 41–50% 51–60% 61–70% 71–80% 81–90% 91–100%

Example: If you believe that your sales growth is greater than that of approximately 45% of all competitors in yourprincipal served market segment, rate yourself a 5 for the sales growth.

Return on investment (ROI): ; Return on Assets (ROA): ;Relative market shares (RMS): ; Overall customer retention: (cusret) ;Major customer retention: (cusret2) ; Sales growth (salesgr): .

Overall performance (adapted from Moorman, 1995)

Please rate the extent to which your business unit has achieved the following outcomes during the last year. (Eleven-pointscale, where 0 = low and 10 = high)

Overall profit margin relative to the objective for this business unit (perf1).Overall sales relative to the objective for this business unit (perf2)Overall return on investment relative to the objective for this business unit (perf3)


revisiting the miles and snow strategic framework

Documents