A Taxonomy of Manufacturing Strategies

Jeffrey G. Miller • Aleda V. RothSchool of Management, Boston University, 621 Commonwealth Avenue, Boston, Massachusetts 02215

Kenan Flakier Business School, University of North Carolina, Chapel Hill, North Carolina 27599

This paper describes the deveiopment and analysis of a numerical taxonomy of manufacturingstrategies. The taxonomy was developed with standard methods of cluster analysis, and is

based on the relative importance attached to eleven competitive capabilities defining the man-ufacturing task of 164 large American manufacturing business units.

Three distinct clusters of manufacturing strategy groups were observed. Though there is anindustry effect, all three manufacturing strategy types are observed in various industries. Thetwo main dimensions along which the manufacturing strategy groups differ are the ability ofthe firms in them to differentiate themselves from competition with their products and services,and the scope of their product lines and markets, A general method for mapping manufacturingstrategies on these dimensions is described. For each manufacturing group, the relationshipsbetween the competitive capabilities (which describe the manufacturing task), the businesscontext (the business unit strategy), manufacturing activities (manufacturing strategy choices),and manufacturing performance measures are explored and compared.{Manufacturing; Strategy; Taxonomy; Typology)

1. IntroductionThe strategic view of manufacturing as a competitiveweapondatesbackat least to Miller and Rogers (1956).They did not differentiate between a business strategyand a manufacturing strategy. Rather, they saw man-ufacturing policies as necessary ingredients of businessstrategy. The notion of manufacturing strategy as a sep-arate but related functional component of a businessunit strategy is of more recent vintage (Skinner 1978,Hayes and Wheelwright 1984),

Two core elements are central to the definition of amanufacturing strategy as a functional substrategy. Thefirst element is a statement of "what the manufacturingfunction must accomplish" (Skinner 1978). This state-ment, commonly referred to as the "manufacturingtask," is defined in terms of the capabilities the man-ufacturing unit must have in order for the firm to com-pete given its overall business and marketing strategy.Lists of critical competitive capabilities typically includequality, cost/efficiency, delivery/responsiveness, andflexibility. Recent lists also include innovation and cus-tomer service as important capabilities (Giffi et al. 1990).

Hill (1989) has linked the manufacturing task to cus-tomer needs by defining the manufacturing task in termsof those capabilities that are critical to winning customerorders.

The second element of a manufacturing strategy isdefined by the pattern of manufacturing choices that acompany makes (Hayes and Wheelwright 1984,Wheelwright 1984, Hayes etal. 1988, Hill 1989). Hayesand Wheelwright (1984) classify these strategic man-ufacturing choices into two categories. The first categoryrefers to structural or "bricks and mortar" decisionsabout facilities, technology, vertical integration, and ca-pacity. The second category of choices is concerned withmajor decisions about the manufacturing infrastruchire,such as organization, quality management, workforcepolicies, and information systems architecture.

The central theme that links the two elements of amanufacturing strategy is the notion that the pattern ofchoices followed by manufacturing must be congruentwith the manufacturing task (Anderson et al. 1989,Buffa 1984, Cohen and Lee 1985, Fine and Hax 1985,Wheelwright 1978, 1984, Hayes and Wheelwright 1984,

0025-1909/94/4003/0285$01.25Copyriphl €i 1994, The Institule of Management Sciences MANAGEMENT SciENCE/Vol, 40, No, 3, March 1994 285

MILLER AND ROTHA Taxonomy of Manufacturing Strategies

Schroeder et al. 1986, Skinner 1969, 1978, 1985, Rothet al. 1989, Hill 1989, Roth and Miller 1990, Stobaughand Telesio 1983, Swamidass and Newell 1987). Thedemand that manufacturing choices and manufacturingtasks be linked follows from the presumption that gooddesigns (such as those specified by the manufacturingchoices) meet appropriate design criteria (as defined bythe manufacturing task). The implication of this pre-sumption is that superior performance will follow forthose firms that develop congruence between theirbusiness and manufacturing strategies. Evidence tosupport this implication is growing (Skinner 1978, Buffa1984, Swamidass and Newell 1987, Roth 1989, Rothand Miller 1990, 1992),

Though the basic elements of a manufacturing strat-egy are commonly accepted, important questions remainabout how to operationalize them, and about the natureof the connections between task and choices. Batten etal. (1978), in their description of functional area re-search on operations strategy and performance noted,"In general these models deal with existing operations

and are not strategic in outlook, i.e., they are not con-cerned with changing the existing operations, nor dothey take a global encompassing view of the firm." Tenyears later, a literature review of manufacturing strategyundertaken by Anderson et al. (1989), describes agrowing body of research that attempts to take a morecomprehensive view. However, they conclude thatknowledge about key relationships among manufac-turing tasks, manufacturing choices, and business strat-egies remains disappointingly small.

The research described in this paper has two relatedpurposes. The first purpose is to identify strategic groupsof manufacturers with similar manufacturing tasks, thatis, with similar sets of competitive capabilities. Deter-mining the types of strategies extant has significant valuefor the emerging field of manufacturing strategy re-search. Taxonomies provide parsimonious descriptionswhich are useful in discussion, research and pedagogy.Moreover, finding sets of manufacturers with commonprofiles may reveal insights into underlying structuresof competition as viewed from the perspective of themanufacturing function. Are the manufacturing strat-egies of firms shaped by market or other forces? Towhat extent do manufacturing managers in differentindustries and markets share similar views on the man-ufacturing task?

The second purpose of this research is to explore thecentral theme in the manufacturing strategy literatureby determining and comparing how members of man-ufacturing strategy groups typically define their businessstrategies, manufacturing choices, and performancemeasures. This analysis illuminates the extent to whichthe Ht between strategy, actions, and measures can beobserved for groups.

This paper is organized into five sections, the first ofwhich is this introduction. Section 2 provides a reviewof the application of taxonomies in strategy research asbackground for this inquiry. In §3, the specific variablesforming the basis for the classification scheme are iden-tified, and a description of the data and methods usedin the analysis is presented. Here we report on the strat-egy groups identified by cluster analysis procedures andalso outline the results of a canonical discriminant anal-ysis. Canonical analysis aids in interpreting the under-lying factors that separate the manufacturing strategygroups. In §4, we report on how the strategy groupssystematically differ from one another in terms otherthan those used to define them in the first place. AN-OVA techniques are employed to identify the contextualvariables, manufacturing strategy choices, and perfor-mance measures which correspond to the manufactur-ing strategy groups. We conclude in §5 with a summaryof findings and conclusions, as well as opportunities forfuture research.

2. BackgroundThe determination of homogeneous groups of firmsbased upon taxonomies has been an important researchtheme in the general strategic management and orga-nization literature. (Hambrick 1983a, Fahey and Chris-tensen 1986, and McGee and Thomas 1986 provideuseful reviews). Most of the research has recognizedthat firms can be classified by multiple variables intogroups which are best characterized by the "gestalt" ofthe communalities they share (Miller and Friesen 1977).

A variety of different classification variables or taxonshave been employed to form the diverse taxonomiesthat have been developed. In some cases, the classifi-cation schemes are based on the context of the firm,cast in terms of environmental, technological, or productdescriptors. For example, the stages of a product's lifecycle have been used by many researchers to discrim-

286 MANAGEMENT SciENCE/Vol. 40, No. 3, March 1994

MILLER AND ROTHA Taxononuf of Maniifactiiriug Strategies

inate among various types of strategies and to explainbehavior (Utterback and Abernathy 1975, Thietart andVivas 1984, Horwitch and Thietart 1987), The typol-ogies of organizational theorists have been based onsuch factors as the degree of uncertainty or complexityin the environment (Duncan 1972, Galbraith 1973,Lawrence and Lorsch 1969, Thompson 1967) and ontechnology type (Woodward 1965, and others). Ham-brick and Lei (1985) provide an overview and priori-tization of the contingency variables which have beenused as taxons, as well as those that might be used.

Another line of taxonomical research attempts to cat-egorize firms in terms of strategic decision variables.The "strategic group" research of Harrigan (1985),Hatten et al, (1978), and Cool and Schendel (1987),and others has discriminated between firms on the basisof the scope of their business, as well as their resourceallocation patterns. Others, such as Porter (1980), andKim and Lim (1988) have classified firms on the basisof classical industrial organization constructs such asthe degree of competition, and bargaining power. Milesand Snow (1978) have categorized firms on the basisof their behavior with respect to competitors (e,g,, dif-ferentiators, focusers, defenders) or new opportunities(prospectors vs, reactors).

Clearly, the basis for strategic group formation hasvaried considerably from one study to another (McGeeand Thomas 1986, Fahey and Christensen 1986), Pegelsand Sekar (1989) conclude that the most appropriateway to form strategic groups depends on what the re-searcher intends to accomplish. We believe that thecompetitive capabilities which define the manufacturingtask are the appropriate grouping criteria for the pur-poses of this research. The manufacturing task as re-vealed through ratings of these capabilities indicatesthe "strategic intent" (Hamel and Prahalad 1989) ofmanufacturing, and provides a basis for testing whetherthe business strategy and manufacturing strategychoices are consistent with this intent,

Stobaugh and Telesio (1983), who have developedthe only other taxonomy of manufacturing strategygroups that we have found, have also used the man-ufacturing task to define their strategic groups. Theirtaxonomy can best be referred to as a "conceptual" tax-onomy, or typology, since it was inferred from theirreview of case studies describing over 100 multinationalfirms. They hypothesized the existence of cost, tech-

nology, and market driven manufacturing strategygroups on the basis of their a posteriori empirical reviewof these cases. The cost-driven group emphasized thecapability to produce at low cost, the technology-drivenfirms emphasized flexibility to introduce new products,while market-driven firms focused on quality and de-livery. For each manufacturing strategy group, Stobaughand Telesio identified critical manufacturing decisionsand choices.

In contrast to the work of Stobaugh and Telesio, thetaxonomy developed in this paper borrows techniquesfrom the biological and social sciences to develop a"numerical" taxonomy using primary as opposed tosecondary sources of information. The taxonomy is de-veloped by applying multivariate statistical proceduresand grouping algorithms to measures of the perceptionsof manufacturing managers. The research reported inthis paper goes beyond the formulation of a numericaltaxonomy by attempting to identify the constructs thatunderlie its formation, and by observing the apparentrelationships between group membership, businesscontext, manufacturing choice, and manufacturing per-formance measures.

3. Methods

3.1. The SampleThe data for this study were obtained from the 1987Manufacturing Futures Project (MFP) Survey (Millerand Roth 1988), The annual Manufacturing FuturesProject Survey is specifically designed to gather infor-mation on competitive factors pertinent to leadingmanufacturing business units. The underlying logic ofthe survey instrument was established in the first Man-ufacturing Futures survey administered at Boston Uni-versity in 1981 (Miller 1982). Since 1983, the surveyhas also been conducted internationally on an annualbasis (DeMeyer et al. 1989, Ferdows et al. 1987, Milleret al. 1989, Roth et al. 1989). The duster analysis re-ported in this paper is restricted to the 1987 NorthAmerican respondents.

A manufacturing business unit (MBU) is defined bythe level in the organization where a manufacturingstrategy was formulated, ln the 1987 study, the domi-nant type of manufacturing business unit was a divisionor group. However, not all firms are divisionalized. Insome firms, business unit considerations are pulled to-

MANAGEMENT SCIENCE/VOI. 40, No, 3, March 1994 287

MILLER AND ROTHA Taxmomxi of /Urim /̂zn/Krujy Stratcgit

gether at the plant, or firm level. Table 1 highlights de-scriptive statistics characterizing the 1987 manufactur-ing business units.

The 1987 sample was pooled from two sources: alongitudinal panei of businesses whose responses wereobtained in previous years, and a probability sample ofbusinesses derived from the 1986 Diin'i^ Business Rank-ings Director}/. This sample was stratified to ensure thatall types of manufacturing business would be repre-sented. The longitudinal panel was originally drawnfrom the 1980 Fortune 500 listing of the largest indus-trial corporations, supplemented by various other listsof senior manufacturing executives. (See Roth andMiller 1987 for sample details.) In 1987, the responserate for the longitudinal panel was about 40 percent;and from the probability sample, 23 percent. An analysiswas performed to determine whether there were anysignificant differences between the longitudinal panel

Table 1 Respondent Profile Statistics

Industry Mix' '

Machinery

ElectronicsConsumer

Industrial

Basic '

Type of Manufacturing Unit

Division/group ' '

Entire company '

Plant

Other .* 1

r-' i. ' ' •

Overall Performance . • „•

Annual sales revenue (median)

Net protit (mean % of sales)

1986 unit growth rate (mean)

Market share of primary product (mean)

Manutacturing

Manufacturing costs (mean)

Capacity utilization (mean)

Direct labor employees (mean)

Number cf plants (mean)

27.7%23.9%12.8%

21.8%

13.8 %

100.0 %

1

54.3 %

30.3 %

• 13.8%

.. 1.6%

100.0 %

:-* . ,ii

$200 million

8.3 % of sales

9.9 % growth

33.4 % share

57.4 % of saies

69,7 % of capacity

2800 people

9.9 plants

* Unless otherwise indicated."%" refers to the percent of the sample

respondents. • , ' • . :

and the probability sample. The hypothesis that thedistribution of manufacturing tasks of the two panelswas drawn from different populations was easily re-jected at the 0.05 level of significance.

Of the 195 surveys returned in 1987, 188 useabJesurveys were initially eniployed in this analysis. Thoseexcluded were duplicate responses and substantiallyincomplete questionnaires. A comparison of the finan-cial and market performance of the respondents withaggregate industry data suggests a nonresponse bias.Respondents were more likely to be market share leadersand financially mt)re successful.3.2. The Respondents

A single respondent from each sample MBU, typicallyholding the title of Vice President or Director of Man-ufacturing, completed the survey instrument. Socialscientists have long puzzled over the problem of "com-mon method variance" due to the bias associated withusing a single informant. Asking a single informant tomake complex social judgements nbuut organizationalcharacteristics may increase the subjective propensityof respondents to seek out consistency in their responsesand increase random measurement error, It is not clearwhether these random error components result fromthe reporting process, knowledge deficiencies, inade-quate measures, or some combination of these and otherfactors. It is generally concluded that strong assessmentsof convergent or discriminent validity cannot be madewhen there is a single informant. However, the costassociated with gaining both participation and consen-sus from several individuals from large numbers of or-ganizations is very high. Numerical taxonomies are bestdeveloped from large samples. Therefore, we have em-ployed data from single respondents while attemptingto minimize the extent of common method variance.

Research suggests that greater attention to informantselection can help to overcome the common methodvariance problem when practical considerations requiresingle respondents. Phillips (1981) indicates that "highranking informants tend to be more reliable sources ofinformation than their lower ranking counterparts . . .and . , , large multidivisional firms may have morewell-established competitive intelligence systems thansmall organizations." The high level of the respondentsin the MFP survey and the size of their business unitshelp to moderate the mono-respondent problem.

288 MANACEMEhJTSciENCE/Vol. 40, No. 3, March 1994

MILLER AND ROTHA Taxoiwiiiu of Miuiufticturin;^ Strategies

We also took several other steps to reduce responsebias and measurement error due to mono-informants.The survey instrument was long and variables in eachsection of the survey were randomly placed, reducingthe chance that respondents could cross check for theirown internal consistency. Internal and external checksfor the reliability and validity of taxons, as described inthe next section, were also made.

3.3. The InstrumentThe questionnaire focused on four broad categories ofquestions. The first category determined the profile ofthe company or business unit. The second category ad-dressed the competitive capabilities the respondentsplanned to pursue, and thus the manufacturing task.In the third section, the respondents were questionedon the performance measures employed in manufac-turing, the business unit and in the company overall.The fourth and largest category of questions probed forthe key action programs the respondents and theirbusiness units intended to invest in over the ensuingtwo years. This fourth category of questions attemptedto tap the pattern of choices in the manufacturingstrategy.

The 1987 survey instrument consisted of over 100questions and over 300 separate data elements. Someof the questions relating to contextual data sought ob-jective answers, such as the percent market share orprofit margin. The majority, including those whichsought to determine the relative importance of theeleven competitive capabilities and those indicating theimportance of various action programs or performancemeasures, required subjective estimates.

The following list describes the 11 competitive ca-pabilities delineated in the survey that were used astaxons in this research. Manufacturing executives in thesample were asked to rate each competitive capabilitymeasure separately on seven-point, self-anchoringscales. They indicated the relative importance attributedto each capability that the manufacturing firm chose toemphasize in appealing to customers and competing inthe marketplace, where "1 = not important" and "7= critically important." The first eight capabilities clearlyhave an impact on manufacturing, and are similar tothose suggested by Buff a (1984), Skinner (1969, 1985),and Hayes and Wheelwright (1984). The last three ca-

pabilities have been added to provide some insights intothe relationship between the first eight manufacturingcapabilities and marketing capabilities, and because de-velopment of these capabilities can have important im-plications for manufacturing choices.

' • *

Taxons

CompetitiveCapability

Low PriceDesign Flexibility

Volume Flexibility

CunformancePerformance

SpeedDependability

After Sale ServiceAdvertising

Broad Distribution

Broad Line

Defined as:

The capability to compete on price.The capability to make rapid design changes

and/or introduce new products quickly.The capability to respond tu swings in

volume.The capability to offer consistent quality.The capability to provide high performance

products.The capability to deliver products quickly.The capability to deliver on time (as prom-

ised).The capability to provide after sale service.The capability to advertise and promote the

product.The capability to distribute the product

broadly.The capability to deliver a broad product

line.

The industry pooling problem referred to by Hattenet al. {1978), was avoided by wording the questions sothat respondents prioritized their capabilities relative totheir particular competitive situation, rather than usingindustry in general as a reference set. It is important tonote that the internal reliabilities of the capability ques-tions in the survey have been shown to exceed 0.90(Huete and Roth 1987). Other researchers have alsoreported high levels of measurement reliability and va-lidity using similar competitive capability variables withdifferent populations and using different researchmethods (Nemetz 1990, Wood et al, 1990).

3.4. Identifying Strategy TypesCluster analysis was employed to identify the manu-facturing strategy types from the respondent capabilityprofiles. The SAS AceCLUS (Approximate CovarianceEstimation for CLUstering) procedure, a variation ofthe Art et al. (1982) algorithm, was used to obtain es-

MANACEMENTSCIENCE/VOI, 40. No. 3, March 1994 289

MILLER AND ROTHA Taxononiy of Manufacturing; Strategies

timates of the pooled within-duster covariance matrixof firms, using the competitive capability scores. Thepooled within-cluster covariance matrix from AceCLUSwas processed by the SAS Fastclus algorithm. Fastclusis particularly well suited for large data sets (i.e., n> 100). The Fastclus procedure uses a variation of thefc-means method for nonhierarchical cluster analysiscalled nearest centroid sorting. Limitations of the kmeans procedure include its sensitivity to outliers andthe requirement for fairly stable seed points (Milligan1980). Of the 188 observations entered into the clus-tering procedure; six were dropped due to missing data;18 other observations were not assigned to a cluster.An a priori analysis of the frequency distribution of thedistances of each point to the nearest cluster seed in-dicated that these observations would have a distortingimpact on cluster definition.

One thorny problem with cluster analysis is the de-termination of the most appropriate number of clusters.Three criteria were employed to determine the finalnumber of clusters to be used in subsequent analysis.First, we were guided by Lehmann's (1979) suggestionthat the number of clusters be limited to between n/30 to n/60, where n is the sample size. Thus, onlymodels with between three and six clusters were con-sidered. Second, we looked for pronounced increasesin the tightness of the clusters, as measured by the R^and pseudo-F statistic {Milligan and Cooper 1985). Thethree cluster model provided the best fit. Third, wesought managerial interpretability of the clusters on thedefining variables using (a) ANOVA and (b) the Scheffepairwise comparison tests of mean (centroid) differences(Harrigan 1985). The three cluster model best satisfiedthese criteria. An overall multivariate test of significanceusing the Wilks Lambda criterion and the associated Fstatistic indicated that the null hypothesis that the threeclusters are equal across all defining variables could berejected (p < 0.0001).

The three resultant manufacturing strategic groupsare described in Table 2 in terms of their respectivegroup centroid (mean) scores and their relative rankingin the set of 11 competitive capability variables. Theprobability that one or more of the cluster means dif-fered from another is also depicted for each competitivecapability. The clusters differed from each other on six

of the eleven variables at the 0.05 level of significanceor less.

The interpretation of the three manufacturing stra-tegic groups, which we have named "caretakers,""marketeers," and "innovators" respectively, is givenbelow. These interpretations are predicated upon: (a)whether there are significant differences on the clustermeans of the competitive capability variables at the 0.05level or less, and (b) the relative ranking of the impor-tance of a competitive capability within a cluster. It ispossible that a high ranking capability within a clustermay actually exhibit a relatively low numerical score.

Cluster 1: Caretakers. We label cluster 1 the "care-takers" because their low relative emphasis on the de-velopment of competitive capabilities appears to preparethem for the minimum standards for competition. Pricewas not significantly different in importance among thegroups. However, based upon its relative rank, priceappears to be the dominant competitive capability forthe members of cluster 1. The relative ranks of the time-based competitive capabilities of delivery dependabilityand speed are high (2nd and 4th respectively), thoughthe ability to meet delivery schedules and the ability tomake fast deliveries is also important to the membersof other clusters. Conformance quality, while rated sig-nificantly below the importance given by both cluster2 and 3 members, is the third most important compet-itive priority for the caretakers. On the other hand,cluster 1 members ascribe significantly less importanceto after sales service and high performance products.The 18 members of cluster 1 represent 11 percent ofthe cases in all three strategic groups.

Cluster 2: Marketeers. The marketeers distinguishthemselves from their cluster 1 and 3 counterparts onseveral key market oriented competitive capabilities.They seek to obtain broad distribution, to offer broadproduct lines and to be responsive to changing volumerequirements. Top ranked priorities within the marke-teer cluster were conformance quality, dependable de-liveries, and product performance. On each of thesetop-ranked competitive capabilities, except for depend-ability, the magnitude of importance was markedlyhigher in comparison to the caretakers but were notdistinguishable from cluster three scores. Furthermore,marketeers reflected some price consciousness. The

290 MANAGEMENT SCI ENCE/Vol. 40, No. 3, March 1994

MILLER AND ROTHA Taxonomy of Mamifacluniig Slralegies

Table 2 Competitive Capabilities By Group*

Competitive Capabilities

Low PriceCluster mean* . , , ,Rank**Standard Error" ' .

Design FlexibilityCluster mean"Rank" , ..Standard Error '"

Volume FlexibilityCluster mean*Rank**Standard Error***

ConformanceCluster mean'Rank"*Standard Error""'

PertormanceCluster mean"Rank"Standard Error '"

SpeedCluster mean'Rank**Standard Error'"" , . . . , .

DependabilityCluster mean'Rank"Standard Error"""

After Sale ServiceCluster mean'Rank""Standard Error '" —

AdvertisingCluster mean*Rank*'Standard Error'*'

Broad DistributionCluster mean*Rank'*Standard Error*"

Broad LineCluster mean'Rank"Standard Error""'

Caretakers{n = 13)

Group 1

6il610.21

5,1150.35

4.1790,37

5.8330.17

4,179QM

5,5540.25

6,0020,20

2,94100,25

4.5580.35

4.7a ,

60 33

Manufacturing Strategic Group

Marketeers{n = 31)

Group 2

5,7040,13

(3) 5,0790,15

5.04100.15

{2,3) 6,7810.05

(2.3) 6,0730.09

MM

0.12

6.1120,09

(2,3) 5.2780.16

4.30110.14

5.6850.13

-V 5.3070,13

(3)

{3)

(1)

(1)

(1)

{3}

Innovators(n = 65)

Group 3

5.4660.18

5.9840,12

4.0610

'0.16

6,7510,05

6.3826.06

«.4O7

3

' am"5.5250.17

•4 :

,m3,92110.22

4,7880,19

(1.2)

(2)

(1)

(1)

(1)

(2)

F = Value[p = probability)

F= 1,68, p = 0,190

F= 10,07p < 0.0001

F = 9,42p < 0,0001

F = 32,59p < 0.0001

F - 46,76p < 0.0001

' F = 0,584p = 0,666

- f = 1,465p " 0,234

F = 25,92p = 0.0001

F = 0,239p = 0,787

F = 26,12p < 0,0001

F = 2,59p = 0.078

' Represents ttie average degree of importance attactied to each competitive capability by cluster. Importance is measured on a sevenpoint self ancfioring scale, (Interval scale 1 - 7 1 = very unimocrtant and 7 = very important)," The rank order ot importance of this competitive capability within the group." ' T h e standard error of the estimate of the mean for the group.

Note. The numbers in parentheses indicate the group numOers from which this group was significantly different at the 0.05 level asindicated by the Scheffe pairwise comparison procedure. Numbers in bold indicate the highest group centroid tor that measure. Group1 = caretakers, group 2 = marketeers, group 3 = innovators. The observed f-statistics were derived trom one-way ANOVAs and thep-values are associated with the observed f-statistics.

MANAGEMENT SCIENCE/VOI, 40, No, 3, March 1994 291

MILLER AND ROTH

A Taxonatity of Matiufaclitring Strategies

ability to offer low price was fourth in overall impor-tance lo this group. Finally, marketeers were more likelyto note after sales service as important than the caretakergroup. The marketeers are the largest group, accountingfor about 49 percent of the cases.

Cluster 3: Innovators. Labelled the "Innovators,"cluster 3 members are differentiated by the relative em-phasis placed upon their ability to make changes in de-sign and to introduce new products quickly. The in-novators share certain characteristics with the marke-teers. In both groups, conformance and performancequality hold top ranked spots. Dependability is also im-portant to the innovators. There were no statisticallysignificant differences between the innovators and themarketeers on after sales service. As with the caretakers,the innovators share lower degrees of emphasis on theability to carry a broad product line and on volumeflexibility. In comparison to the other clusters, price isleast important to this group. The innovators form thesecond largest cluster, comprising 40 percent of the

cases.

4. Analysis and Discussion4.1. Underlying DimensionsThe 11 taxons representing the importance of the com-petitive capabilities succeeded in generating a numericaltaxonomy somewhat consistent with that hypothesizedby Stobaugh and Telesio (1983). However, the 11 tax-onomic variables used to form the manufacturing strat-egy groups are correlated with each other. Therefore,we adopted multivariate statistical methods to extendour exploration. Multiple group discriminant analysis,with each of the taxonomic groups as criterion variablescoded into 3 - 1 ^ 2 dummies and with the 11 com-petitive priority taxons comprising the predictor set, wasperformed using the SAS Candisc procedure (canonicaldiscriminant analysis). Canonical discriminant analysis,a more general approach to discriminant analysis, is adimension-reduction technique related to principalcomponent analysis and canonical correlation (Green1978, Fomell 1978, Kendall and Stuart 1968, Dillon andGoldstein 1984),

With this approach, we obtained standardized esti-mates for both the canonical structure loadings and forthe canonical coefficients. The canonical structure load-

ings can be interpreted like the factor loadings in prin-ciple components analysis; i.e., they represent the cor-relations of the original variables with an underlying,unobserved dimension. For substantive interpretations,the canonical loadings are useful as indicators of whichoriginal variables are most correlated with each canon-ical variate in the spirit of factor analysis (Dillon andGoldstein 1984), The standardized canonical coeffi-cients are analogous to beta weights in regression, andcan be used to predict cluster membership. To furtherverify our results, multiple discriminant classificationanalysis and cross-validation techniques were adopted.

Table 3 contains the results of the canonical discrim-inant analysis used to investigate the relationship be-tween the 11 taxons and cluster membership. In theseanalyses, the f statistic for Wilk's Lambda = 0,19 in-dicated a significant overall multivariate relationship(p < 0,0001), Two significant canonical functions fordefining manufacturing strategy clusters resulted(K, - 0.76, p < 0.0001 and K-2 '= 0,74, ;; < 0.0001),applying Rao's (1973) approximate f-statistics. Takentogether, the two canonical functions retain 23 percent

Table 3 Results of Canonical Oiscriminant Analysis

Canonical

Function Eigenvalue

Correlation or Root

1 1,33

2 1,19

Predictor Set

Low Price

Design Flexibility

Volume Flexibility

Contormance

Performance

Speed

Delivery Dependability

After Sale Service

Advertising/Pro mo

Broad Distribution

Broad Line

Significance of

Canonical

Correlation-flf

0,76

0,74

Canonical 1

Function

1

-0.1639

0,1761

0,0608

0.6980

0.7996

-0.0347

0.1220

0.6529

-00713

-0,0665

-0,0136

Loadings

Function

2

0,0965

-0.4146

0,4346

0,1343

-0,0631

0,1095

-0,1312

-0,0042

-0,0104

0.6677

0.2390

Canonical

Correlation

0,0001

0,0001

Squared

Canonical

0,58

0,55

Canonical Coefficients

Function

1

-0.2954

-0,2169

-0,0652

0.6837

0.7960

-0,1002

- 0 1151

0.5788

-02799

-02204

0,2636

Function

2

0,0262

-0.7404

0.4377

0 3944

-0,1270

0,0085

-0,3813

0,0554

0,4903

1.4484

-0.4694

Bold numbers indicate fiigh loadings (weights) in canonical functions ± 10,401.

292 MANAGEMENT SciENCE/Vol. 40, No, 3, March 1994

MILLER AND ROTH

A Taxonomy of Manufacturing Strategies

of the variance in the data as indicated by their averagecommunality (Fornell 1978), The canonical correlationis a measure of the relative strength of the relationshipbetween the set of competitive capabilities and themanufacturing strategy group membership. We em-ployed 7(, as the measure of multivariate association(MVA) {Cramer-Nicewander 1979). In the case of ca-nonical discriminant analysis, 7̂ can be defined as theaverage squared multiple correlation which is equivalentto the average squared canorucal correlations. Therefore,about 56 percent of the variance in manufacturingstrategy group membership is explained by the set of11 taxons, and vice versa, since the measure is sym-metric. The redundancy index (Green 1978) was lessappropriate because only a few of the observed variablesdetermined the structure matrix (Fornell 1978) and be-cause the criterion is a single categorical variable, i.e.,manufacturing strategy group membership (Cramer-Nicewander 1979),

Our interpretation of the underlying manufacturingstrategy dimensions is based upon canonical variateanalysis of the standardized canonical loadings. The ca-nonical loadings are also helpful in labelling the derivedcanonical variates of the predictor set. Using these, weinterpret canonical function 1 to be a "market differ-entiation" dimension that depicts the perceived re-quirement of the manufacturing firm to distinguish itselffrom competitors by attributes of its products and ser-vices. Here the largest correlates of cluster membershiphave to do with the relative importance given to productperformance, conformance quality and after-sales ser-vice. The relatively large canonical coefficients for per-formance, conformance and service suggest that a firmplacing a greater priority on these capabilities will fallat the high end of the market differentiation dimension;and those with lower emphasis will be less likely todifferentiate products/services on these attributes.

We call function 2 "market scope." This canonicalvariate reflects the magnitude of the customer baseserved by the business unit. Broad distribution and vol-ume flexibility are positively correlated with function2, while design flexibility is negatively associated. Thestrong positive coefficients for broad distribution andvolume flexibility imply that a firm placing a high em-phasis on these capabilities will be assigned to the highend of the spectrum. Those manufacturers with an em-

phasis on specialty distribution channels and designflexibility are likely to be assigned to the low end. Oneinterpretation is that the low end competitors are com-peting in specialty markets with a broad range of tastes(hence the need for design flexibility), while those atthe high end of this scale are competing in high volumemass markets with more stable product lines.

Figure 1, derived from the canonical coefficients,characterizes the responding MBUs strategic positionagainst the two manufacturing task constructs of marketdifferentiation and market scope. The numbers on theplot, indicate the manufacturing strategy group assign-ment designated by the clustering procedure (1 = care-takers, 2 ̂ marketeers, and 3 = innovators). It seemsclear that the caretakers are less likely to have or tovalue differentiating products and services. Therefore,it is not surprising that the caretakers compete mainlyon price, and are on the low side of the market differ-entiation scale. The innovators, like the marketeers, aremore apt to give weight to product attributes, and thusappear on the high end of the market differentiation

Figure 1 Plot of Respondent Business Units and Group Centroids on

Canonical Functions

Letters in circles indicate cluster centro ids. The letters are: Caretakers ( C ) ,

Marketeers ( M ) , and Innovators ( I ) ,

->,« •I.M

FUMOtOHl

MariiM Dtrre

Key: 1 = Caretakers; 2 = Marketeers: 3 = Innovators

M A N A G E M E N T SCIENCE/VOI , 40, No, 3, March 1994 293

MILLER AND ROTHA Taxonomy of Mamifacturing, Strategies

scale. However, unlike the marketeers who as a grouphave very broad distribution channels, and thus are onthe high end of the market scope scale, the innovatorsemphasize responsiveness to product change throughdesign flexibility, which is associated with the low endof the market scope scale.

These findings suggest that in general, the manufac-turing task is a multivariate, multidimensional constructthat reflects the needs of widely different market en-vironments and relative market positions. They alsosuggest that these contextual factors represent deeperand more fundamental drivers of firm behavior thancost, technology, or markets as proposed by Stobaughand Telesio (1983). Here, we see that a cost (price)emphasis is associated with a lack of differentiation,that technological change in terms of flexibility to in-troduce new products is associated with specialty mar-kets, and that marketeer oriented behavior is associatedwith broad distribution and product and service differ-entiation.

On the other hand, the market structure orientationof these underlying constructs reinforces the views ofSkinner (1978) and Hill (1989) who relate the manu-facturing task to fundamental market positioning (andorder winning) strategies. The factors are also closelyallied with those proposed by Porter (1980), He cate-gorized firms by their propensity to differentiate them-selves or to compete on price, and by the degree towhich they focus their efforts on particular markets orproducts,

4.2. Statistical Cross-validationTo determine the stability of the estimates, a cross-loading analysis was first conducted over the canonicalstructure variates. The cross-loadings on the originaldefining variables appeared stable as shown in Table 3.Each of these loadings exceeded |0,30|, giving us noreason to change our original interpretation. Next, theperformance of the discriminant taxons was evaluatedby estimating the probability of misclassification in theclassification of future observations through cross-val-idation procedures. We could not employ split-sampletechniques due to the small size of cluster 1. Using ajackknife procedure, a discriminant function with the11 taxons as predictors was computed for u - I out ofn cases, and that function was used to classify the one

Table 4 Number of Observations and Percent Cross Validated

To/From Cluster 1 Total

Caretakers Marketeers Itjnovators

1 13(72%) 3(17%) 2(11%) 18(100%)? . 0(0%) 80(99%) 1(1%) 81(100%)3 • 0(0%) 11(17%) 54(83%) 65(100%)

Error Rates from:

Cross-validation 0,28 001 0,17 0,10Posterior

Probabilities 0,34 0,05 0,28 0,18

observation held out (See Table 4). This process wasrepeated for each of the » cases, and the proportion ofhold-out observations assigned to each group was de-termined (Lachenbruch and Mickey 1968), Cross-val-idation error rates reflect the percentage of hold-outcases misclassified. Another set of error-rate estimateswere computed based upon the posterior probabilityestimates based on cross-validation procedures de-scribed above. The cross-validation posterior prtibabilityestimators have been shown to possess both low biasand low variance (Hora and Wilcox 1982). The cross-validation results are as follows.

Cross-validation classification analysis suggests thatthe overall discrimination power of the taxons is quitegood, with 72 percent of cluster 1, 99 percent of cluster2, and 83 percent of cluster 3 correctly classified. Thetaxons do a superior job of predicting actual membershipin cluster 2, but are more prone to misclassify membersof clusters 1 and 3,

4.3. Industrial MixTable 5 illustrates the relationships between groupmembership and broad categories of industry mem-bership, where industr)' is described in terms of fivegroups developed for the Manufacturing Futures Surveybased upon aggregation of three digit Standard Indus-trial ClassificaHon (SIC) codes (See Miller and Vollmann1984 and Roth and Miller 1987). A chi-square test in-dicates that cluster membership is associated with theindustry in which the MBU resides (;J < 0.01). For ex-ample, the electronics (computer, instrument, and elec-tronic equipment manufacturers) and machinery (ma-chine tools, transportation equipment, machinery

MANAGEMENT SCIENCE/VOI . 40, No. 3, March 1994

MILLER AND ROTHA Taxonomy of Manufacluri/ig Strategies

Table 5 indusUy Representation By Strategic Groups

(Number of Respondents)

Consumers

Industrial

Basic

Machinery

Electronic

Frequency

Percent

Caretakers

6

7

3

1

1

18

11%

Marketeers

11

17

16

20

17

81

49%

Innovators

3

13

4

25

20

65

40%

Frequency/%

20

12%

37

23%2314%4628%3823%

164

100%

x ' = 24,78 d,f, = 8 p < 0 , 0 1 ,

groups) industries are more likely to populated by in-novators. Conversely, the industrial (parts, components,intermediate goods producers), basic (chemical, paper,primary metals firms) and consumer packaged goods(foods, cosmetics, pharmaceutical manufacturers) aremore likely to be marketeers. The caretakers are mostlikely to be found among the industrial goods and con-sumer packaged goods firms.

Despite these broad industry associations with man-ufacturing strategic group membership, examination ata more detailed level showed that it was common forat least one competitor in a particular three digit SICindustry to compete on a substantially different basisthan its primary competitors. We could identify 17 in-stances in which two or more of the respondents to thesurvey were competing in the same industry. In 12 ofthese instances, at least one competitor was in a differentmanufacturing strategic group from the other(s). Forexample, in the micro computer industry, three of thecompetitors in the sample were innovators, but two re-spondents were clearly classified as marketeers. In an-other industry, three competitors each occupied a dif-ferent strategic position; one was a caretaker, anothera marketeer, and the third an innovator. These examplesare consistent with findings from other strategic groupresearch that demonstrates that a broad range of strat-egies is available to competitors within an industry (forexample. Porter 1980).

4.4. ContextAfter developing the clusters with the procedures de-scribed above, the factors associated with strategic groupmembership were systematically examined to fulfill thesecond purpose of the study. Using ANOVA andScheffe post hoc pairwise comparisons of group meandifferences, industry, environmental, and contextualvariables, as well as manufacturing strategy choices andperformance measures, were explored to see if theyshowed significant differences across clusters. In makingstatements about relationships, we have designated ana priori level of statistical significance of 0,05 or less.Table 6 describes how each cluster of manufacturersdiffers across the context variables. This table is typicalof several that follow. With the exception of the Scheffemultiple comparisons tests, no adjustments were madefor the problem of simultaneously testing multiple hy-potheses on the same data set due to the exploratorynature of our inquiry. Thus, the true ;?-value may ac-tually be much greater than the nominal level shownin the tables, and future research is required to confirmthe associations found.

In Table 6 we see evidence that the clusters tend todiffer both in terms of the stage of their product lifecycle, and by the degree of product standardization.The caretakers derive significantly more revenues from"maturity stage" sales, as a percent of total sales, thaneither of their counterparts. The marketeers lie in be-tween the caretakers and innovators in terms of bothgrowth and maturity stage revenues. The innovatorshave the least standardized (most customized) productswhile the caretaker's products are most heavily stan-dardized.

As might be expected the innovators invest moreheavily in R&D than their counterparts, especially thecaretakers. They were also more apt to report a strongerfunctional influence by Engineering/R&D in setting thelong run goals and strategies of the business unit. In-novators place more emphasis on increasing marketshare by developing new products for both old and newmarkets. Marketeers may develop new products for ex-isting markets but they are less apt to emphasize en-tering new markets with new products.

Further scrutiny of the data shows some evidencethat the clusters exhibit the "product-process" matrixcharacteristics suggested by Hayes and Wheelwright

MANAGEMENT SciENCE/Vol, 40, No. 3, March 1994 295

MILLER AND ROTHA Taxonomy of Manufacturing Strategies

Table 6 Strategy and

Context/Strategy Variable

R&D/Sales %

Cluster mean"

Standard Error" ,,

Export Sales %

Cluster mean'

Standard Error"

Product Standardization

Cluster mean***

Standard Error**

Maturity Stage Revenues %

Cluster mean*

Standard Error**

T Engineerifjg/R&D Influence

Cluster mean" ' . . . , . . . . . •.

,, • Standard Error"

Increase Mkt. Share

Cluster m e a n ' " . .,•.Standard E r r o r "

New Products/Old Mkts.

Cluster mean ' "

Standard Error*'

1 New Products/New Mkts.

: _ Cluster mean***

Standard Error**

Context By Group

Caretakers

{n - 18)

Group 1

2,41

0,86

1.91

0,82

4,44

0.41

73.44

6,18

3,83

0,37

5,00

0,32

4,72

0,27

3,72

0,44

Manufacturing Strategic Group

(3)

(2,3)

(2,3)

(3)

(3)

1k

(3J

(2,3)

(3)

Marketeers

[n = 81)

Group 2

3,88

0,40

14.89

2,04

4,15

0,19

57,59 , •

2.84 ,

4.61

O.t5

5.58 '0.1?

5.75

0.11

4,43

0,19

(1)

(1)

(1)

1

(1)

Innovators

(n = 66)

Group 3

4.97

0.60

14,88

2,52

3,30

0,21

55,15

3,04

. . ,

5.31

0.15

5.78

0,14

5,39

0.16

4.95

0,22

(1)

(1)

(1)

(1)

(1.2)

(1)

(1)

(1)

F

F

P

F

P

P-

F--

= Value

probability)

= 3,06

= 0,051

= 4 24

= 0.017

= 5,66

c 0.004

= 3,91

- 0,022

F = 10.92

p < 0.0001

P-

P =

F -.

P =

= 3,10

-- 0 048

•• 7 6 4

•- 0,001

4,06

0,019

II

V

1 ,

*'

1

* Represents the mean value of the context measure reported by each strategic group,

** The standard error of the estimate of the mean for each group.

" * Represents the average values measured on seven point self anchoring scales (interval scaie 1-7).

See Roth and Miller 1987 (or a detailed definition of the variables, ' ' '

Note Ttie numbers m parentheses indicate the group numbers (rom which this group was significantly different at

the 0,05 level as mdicaled by the Scheffe pairwise comparison procedure. Numbers in bold indicate the highest

group centroid for that measure. The observed f-statistics were derived from one-way ANOVAs and the p-valuesare associated with each of the observed f-statistics.

(1984). We asked the executives to rate the degree towhich the business unit's products are standardized, ona scale from "1 = highly customized" to "7 = highlystandardized," The product/process theory would

2% "

suggest that the lower the standardization score, themore likely the process is characterized as a job shop.The higher the score, the more likely the process is con-tinuous. The caretakers display a strong proclivity for

MANAGEMENT SciENCE/VoI. 40, No. 3, March 1994

MILLER AND ROTHA Taxonomy of ManiifacturiDg Strategic?

producing more standardized products, in contrast tothe innovators and marketeers. Weaker differences inthe degree of product customization were reported bythe innovators and marketeers. The mean values forthese two groups indicate greater customization for theinnovators and higher standardization for the marke-teers.

We asked a question concerning the dominant pro-duction process of the manufacturing unit, v '̂here a scoreof " 1 " indicates a continuous flow process and a scoreof "5," a job shop with no dominant flows. While theevidence is weaker {p = 0.08), the data suggest thepredicted pattern. The innovators are at the near endof the job shop continuum, while the caretakers are atthe opposite extreme (more continuous processes); themarketeers lie in between.

4.5. Action ProgramsIn the survey, respondents were given a list of 36 keyaction programs to improve the effectiveness of theiroperations over the following two years. (See Millerand Roth i988 and Roth and Miller i 987 for a discussionand complete list of the action programs in the survey.)The manufacturing executives were asked to rate from"1 = unimportant" to "7 = very important" the degreeof emphasis that would be placed on each action pro-gram over the next two years. The action programs in-dicate the intended emphasis on the respondent man-ufacturing choices. Roth et al. (1989), Roth and Miller(1990, 1992), and Ward etal. (1988) have shown thatthese action plans tap important underlying structuraland infrastructural directions in a manufacturing strat-egy. In testing the common theme In manufacturingstrategy that actions (choices) should be congruent withthe manufacturing task, ten strategy choice variableswere found to be significant at the 0.05 level in ourdata.

The innovators' manufacturing strategy choices placesignificantly more emphasis on programs that promiseto shorten total product cycle times. Innovators also fo-cus on computer aided design (CAD), and emphasizedeveloping new processes for their new products. And,innovators plan to embark upon manufacturing pro-grams that reduce their manufacturing lead-times. Thispattern of choices appears to be congruent with a man-ufacturing task that prizes the ability to change designsand introduce new products.

Table 7 suggests that the marketeers plan onstrengthening their manufacturing operations throughinfrastructural changes, particularly those that will cutcosts and improve quality. They intend to emphasizechanging labor/management culture and streamliningtheir workforces through headcount reductions andplant closings. Besides reducing the size of the workforceand improving productivity, the marketeers hope to at-tack quality problems through zero defect programs andstatistical process control. These infrastructural changesappear to be congruent with their more standardizedproducts and continuous processes. They are also con-sistent with the relatively high priority they attached tothe quality, dependability and price capabilities.

Notably, the caretakers, who are competing first onprice, tend to place relatively lower emphasis than theircounterparts on each improvement program measured.Recall that this group had the most mature products.At worst, it appears that the caretakers, as a group, haveno manufacturing strategy and are not renewing theirmanufacturing function. At best, their manufacturingstrategy is passive. Does the caretaker group representthe end of the life cycle? The data suggest that this maybe so. However, there are several among the caretakergroup that have high profits, market shares and eco-nomic returns. Further work must be done to examinehow a firm can use manufacturing to be successful inmature markets.

4.6. MeasuresMeasuring manufacturing performance is becoming anarea of concern for U.S. manufacturers. In particular,there is a need to understand how nonfinancial mea-sures of performance are viewed in the manufacturingunit. Nannietal. (1988), and McDougall (1988) havecalled for performance measures relating to the businessstrategy and the key improvement programs in man-ufacturing, Richardson et al. (1985) have suggested thatmeasures should correspond specifically to the strategiccapabilities which define the manufacturing task. Roth(1989) has shown that the content of a manufacturingstrategy is correlated with manufacturing perfonnancemeasures that have a business unit impact.

From a list of 29 manufacturing performance indi-cators, the respondents to the 1987 Manufacturing Fu-tures Survey were asked to indicate on a self-anchoring

MANAGEMENT SciENCE/Vol. 40, No. 3, March 1994 297

MILLER AND ROTHA Taxonomy of Maiiufactunug Strategies

Table 7 Future Improvement Programs By Group

Manufacturing Strategic Group

Caretakers Marketeers Innovators( " = 1 8 ) ( / ) -81 ) (/I = 65)

F - ValueP^Qfl^^^s Group 1 Group 2 Group 3 (^^probability)

Lat)or/Mgmt Relationships ' " •'" . i

Cluster mean- .,,,.,..,... 4.28 i ;1« (3) 479 ( j j ^ = 5 53

Standard Error" 0.37 . d.1« 0.22 p = 0,005

Zero Defects ' '

Ciuster mean- 4.28 (2,3) 5.43 (1) 5.35 (1) f = 4 35Standard Error" 0.39 0.17 0.1S p = 0.014

MFG Lead Time ReductionI-

Cluster mean- 4.33 (3j 5,16 5.S6 (1) f = 5 8 5Standard Error" 0.36 0,16 0.15 p ^ 0,004

C40 . , , ,.

Cluster mean' „ . , 3.50 (2.3) 4,99 (1) 5.14 (1) f = 6 5 6Standard Error" 0.38 0^0 0.21 p = 0,002

New Process/New Product

Cluster mean- 4.0$ ,' tm 5.05 ' f = 316

Standard Error" 0 , * ' 0.18 0,i8 p = 0,045

Closing Plants

Ciuster mean' 2.00 2.91 (3) 211 (2) f = 3 8 6

Standard Error" . . . . . . . . 0.33 ' 0,15 0.20 p = 0.023

SPC (Process) . "'

Cluster mean- . . . . . . i..v 4,61 (2) 5.79 (1,3) 5,11 (2) f = 6 5 4

Standard Error" 0,36 0,15 0,20 p = 0.002

SPC {Product)

Cluster mean* 4,39 (2) 5.43 (i) 4,89 f = 4 53

Standard Error" 0,36 0,16 0 20 p = o'oi2

New Product Introductions

Cluster mean- 376 (2,3) 5,06 (1) 5.42 (1) f = 6 56

Standard Error" 0,44 o,2O 0,20 p = 0.002

Reducing Workforce Size

Ciuster mean- , . . . . . . . . , 3,83 4.83 (3) 3,89 (2) f = 6 61Standard Error" ;. 0,47 0,20 0,22 p = 0,004• Represents the average degree of importance attached to each program. Importance is measured in a seven point selfanchoring scaie where 1 = very unimportant and 7 = very important," The standard error of the estimate of the mean for each group.

Note. The numbers in parentheses indicate the group numbers from which this group was significantly different at the0,05 ievei as indicated by the Scheffe pairwise comparison procedure. Numbers in bold indicate the tiighesi group centroidlor that measure. The observed f-statistics were derived from one-way ANOVAs and the p-vaiues are associated witheach of the observed f-statistics.

298 M A N A G E M E N T SciENCE/Vol, 40, No , 3, M a r c h 1994

MILLER AND ROTHA Taxouottiy of Mafnifiicturiiig Stratef^ics

Table 8 Pertormance Measures By Group

.

Performance Measures

Manufacturing Lead Time

Cluster mean'

Standard Error"

Changeover Setup Time

Cluster mean'

Standard Error"

Headcount

Cluster mean'

Standard Error"

Ratio White/Blue Collar

Cluster mean*

Standard Error"

Outgoing Quality

• - ' Cluster mean'

, • Standard Error"

Number of Grievances

Ciuster mean"

Standard Error"

% New Products Ontime

Cluster mean'

Standard Error"

Caretakers

{n - 18)

Group 1

4.88 (3)

0,30

4,58

0.35

4,83

0.41

4.44

0.30

5,83 (2,

0.29

3,72

0,29

4,33 (2,

0,23

Manufacturing Strategic Group

Marketeers

{n - 81)

Group 2

5,51

0,12

5,35

0,15

5,54

0,12

4.94

0,15

3) 6.63

0,06

4.42

0.21

3) 5,24

0,14

(3)

(1)

(3)

(1)

Innovators

[n = 85)

• Group 3

5.63

0,13

4.83

0,20

521

0,14

4,37

0.18

6.78

0.06

3,61

0.06

5.64

0.15

(1)

(2)

(1)

(2)

(1)

F = Value

(p = probabiiity)

p = 0,039

p = 0,039

F = 3,39

p - 0.036

F - 3.32

p - 0,039

f = 15,38 'p = 0,0001

f = 5,38

p = 0,006

F = 8,39

p = 0.0003

• « )

•

* Represents the average degree of importance attached to each measure, importance is measured on a seven point self

anchoring scaie where 1 = very unimportant and 7 = very important,

" The standard error of the estimate of the mean for each group.

Note. The numbers in parentheses indicate the group numbers from which this group was significantly different at the

0,05 ievei as indicated by the Scheffe pairwise comparison procedure. Numbers in bold indicate the highest group centroid

for that measure. The observed f-statistics were derived from one-way ANOVAs and the p-values are associated with

each of the observed f-statistics.

scale ranging from 1 to 7, the relative degree of impor-tance the manufacturing management team placed oneach measure, (See Roth and Miller 1987 for a completelist: of performance measures.) Eight performance in-dicators separated the strategy groups at the .05 leveland are shown in Table 8.

The innovators differ from their counterparts by at-taching more importance to performance measured bythe percent of new products/models introduced on

time. They are less apt to be concerned with personnelproblem indicators, such as the ratio of white collar toblue collar workers, headcounts, or the number ofgrievances. Personnel measures are more important tothe marketeers as are changeover and setup times, andwork in process inventories. Innovators place about thesame relative degree of importance as the marketeerson manufacturing lead times, and the level of outgoingquality. Interestingly, the caretakers tend to give lower

M A N A G E M E N T SCIENCE/VOI . 40, No. 3, March 1994 299

MILLER AND ROTHA Taxonomy of Manufacturing Strategies

ratings overall to each manufacturing performance in-dicator. This is further evidence that the caretakers areplaying out an endgame in the product Hfe cycle.

5. Summary and ConclusionsThere is consensus among researchers ab.out the keyelements of a manufacturing strategy. The commontheme that permeates the growing literature is that themanufacturing task, as measured by the importancegiven to competitive capabilities like quality, flexibility,delivery, cost, must be linked to manufacturing strategychoices, and that both should link to the business strat-egy. This research has identified three types of manu-facturing strategies, and has empirically addressed howthis common theme is played out in the choices andbusiness strategies of each,

5.1. Manufacturing TaxonomyThree distinct types of manufacturers can be identifiedby the importance they place on competitive capabilities;caretakers, marketeers, and innovators. Our taxonomyprovides analytical evidence that generally supports thetypology hypothesized by Stobaugh and Telesio (1983),but it is different in some significant and intriguing ways.Our marketeers and innovators are similar to their mar-ket and technology-driven groups. However, there is asubstantial difference in their conception of a "cost-driven" group and our caretakers. While both the cost-driven firms and caretakers emphasize price competi-tion, their cost-driven firms are assumed to have adaptedthis as part of a coordinated market, manufacturing andbusiness strategy. Our caretakers are notable for thelow levels of importance they ascribe to manufacturingcapabilities and choices, and for their seeming lack ofcongruence between these capabilities and choices.Second, we have shown that the underlying structureof the manufacturing task consists of multivariate con-structs. Combinations of capabilities provide better in-sight into the manufacturing task than statements aboutindividual capabilities.

Third, we also found that more than one multivariatedimension is necessary to determine the manufacturingtask. The differences between categories of strategy wereshown to be related to two underlying market charac-teristics: the degree of market differentiation and market

scope. Innovator firms may not be simply "driven" bytechnology, per se, as suggested by a univariate per-spective. There appears to be an interplay between fo-cusing on markets where technology offers a greateropportunity to differentiate and on markets whose nar-row and more specialized customer base creates pressureto change products frequently. More research, howeveris required to determine whether technology drives in-novator firms into niche markets or whether the re-quirements of specialized markets propels the firms to-ward technology. The marketeers may be motivated tomaintain market share through their broad distributionsystems, and respond to demand opportunities withtheir volume fiexibility. The lack of differentiation couldbe behind the caretaker's primary focus on price.

Fourth, we have developed evidence that there is arelationship between broadly defined industry types andthe manufacturing task. However, at a more detailedlevel, we see that this effect does not preclude variationsin the manufacturing task between the firms competingin the same industry.

The taxonomy developed here has much in commonwith other taxonomies developed with entirely differenttaxons and by researchers from substantially differentdisciplines. The similarity to Miles and Snow's (1978)taxonomy is apparent, though ours yields fewer cate-gories. Clearly, their "prospectors," "differentiators"and "defenders," are similar to the innovators, mar-keters, and caretakers here. Their "focusers" and "re-actors" are likely to be variants of the large mass mar-keter and innovator categories. To the extent that theirtaxonomy identifies the same general categories of stra-tegic behavior, this research provides evidence on howmanufacturing strategies can be linked to the businessunit. Similarly, we see a reflection of the industrial or-ganization oriented taxonomies of Porter (1980) andKim and Um (1988), the product process theories ofHayes and Wheelwright (1984a), and the technologytheory oriented taxonomies of Hambrick (1983) andHorwitch and Thietart (1987), in this taxonomy ofmanufacturing strategies. These similarities suggest thestrength of the underlying competitive factors whichseem to explain much of industrial behavior. They alsosuggest that the further development of manufacturingstrategy research will be enhanced by synthesizingfindings and approaches from a number of disciplines.

300 MANAGEMENT SciENCE/Vol. 40, No, 3, March 1994

MILLER AND ROTHA TaxQitonni of Mainifacturnt^ Sfruffjjies

5.2. Consistency of PurposeGoing beyond the questions examined in previous re-search, we used the taxonomy to explore the centraltheme in the manufacturing strategy literature, that is,how major competitive orientations are related to themanufacturing task as revealed through key capabilitiesand context, manufacturing strategy choices, and man-ufacturing performance measures. We have shown inthis paper that significant differences exist in certaincapabilities, actions, and measures among the strategygroups. Moreover, these relationships are generally inthe directions one would anticipate from the underlyingtheory of life cycles, product / process hfe cycles, andother existing theories about production. However, wecannot conclude that linkage between manufacturingtask, choice measures and business unit strategy are auniversal phenomenon.

Our research supports the notion that a manufactur-ing strategy is related to product life cycles. The inno-vators, with relatively short product lives, demonstratethe characteristics of start-up firms in product life cycletheory. For example, the manufacturing strategy of thesefirms is heavily influenced by engineering and researchand development functions, These firms plan frequentmodification of the business unit's production processesand new product introduction capabilities with a specialemphasis on reduced manufacturing and product de-velopment lead times (Utterback and Abernathy 1975),

The marketeers with well established products andmarkets, pursue manufacturing strategies characteristicof a business in more mature phases of the life cycle.Marketeer manufacturing strategy choices are orientedtowards improving the reliability of the manufacturingprocess. Quality control programs were especially prev-alent, as are attempts to reduce the size of the workforce.The key performance measures of the marketeers arecongruent with their manufacturing strategy as reflectedin their focus on productivity and quality issues,

The caretakers, on average, demonstrate the char-acteristics of businesses in the declining stages of thelife cycle. The fact that this group has no clear cut patternof action programs or performance measures related tothe manufacturing task violates the assumption of theone common theme in manufacturing strategy research.There are several plausible explanations. One is thatthe sample size for this group (the smallest) is simply

too small to yield a reliable result. The second is thatthe preoccupation of these firms with price competitionin mature/declining markets leads them to think shortterm, and to merely react to the current situation ratherthan to take any coherent strategic steps for renewal.Further research will be required to determine which,if any, of the above reasons is closest to the truth, andto understand how the best performers in this groupachieve success. This research will require a larger sam-ple of caretaker firms to yield significant results.

5.3. Future ResearchThis study has several limitations which future manu-facturing strategy researchers should consider. First, asnoted above, the cases examined represent a biasedsample of U.S, industry. As a result, the proportion offirms in the caretaker group is probably underestimated,prohibiting any attempt to understand the distinctionbetween subgroups in it. Future research should attemptto obtain a broader representation of caretaker busi-nesses. Another limitation is the correlated error prob-lem that derives from the use of one respondent for allof the data gathered from each firm. Future lines ofinquiry, should consider using multiple sources of in-formation and methods to reduce this problem.

The study provides important clues for better under-standing the relationships between manufacturingstrategy and performance. Nevertheless, a criticalshortcoming is that it provides no visibility into thecausal relationships between manufacturing strategyand performance outcomes. For example, what factorscharacterize good and bad marketeers or innovators?Future research pertaining to the causal linkages be-tween manufacturing task, manufacturing choices andperformance outcomes is best studied via a longitudinalstudy, and cannot be ascertained from a cross-sectionalstudy without significant prior theory as guidance.

Finally, an important line of future research is to testthe stability of this taxonomy globally, and over time.The "laws" of business that seem to define the strategicpositioning and competitive behavior in this and othertaxonomies have all been based on observations of thebusiness environment during a short 20 year span. Ifwe suppose that intelligent competitors will use theirknowledge of the "normal" rules of battle to better de-velop new principles of competitive warfare, then we

MANAGEMENT SciENCE/Vol, 40, No- 3, March 1994 301

MILLER AND ROTH

A Taxonomy of Manufacturing Strategies

may anticipate that new manufacturing strategic groupswill be formed over time, and in different parts of theworld.* ,^^ ,

' The authors wish to thank two anonymous reviewers for their

Ihoughtful reviews of this paper.

References ' - 'Anderson, J. C . G, Cleveland and R, G, Schroeder, "Operations

Strategy: A Literature Review." /. Operations Management. 8 (April

1989), 133-158.

Ansoff, H, 1,, Corporate Strategu: An Anah/tical Approach to Business

Policy for Growth ami E.-(pansion. McGraw-Hill. N'ew York, 1965-

Art, D.. R, Gnanadesikan and R. Kettenring. "Data-based Metrics for

Cluster Analysis," Utilitas Mafhematica, 21A (1982), 75-99.

Buffa, E,, Meeting the Competitive Challenge, Dow Jones-lrwin, Home-

wood, IL 1984,

Cohen, Morris A, and 11 L, Lee, Manufacturing Strategy: Concepts

and Methods," In Paul R. Kleindorfer (Ed.), The Management of

Productivity and Technology in Manufacturing. Plenum Press, New

York, 1985,

Cool K, O, and D, Schendel, 'Strategic Group Formation and Per-

formance: The Case of the U.S, Phannaceutical Industry, 1963-

\982," Management Sci.. 9 (1987), 1102-1124,

Cramer, E. and W, A, Nicewander, Some Symmetric Invariant Mea-

sures of Multivariate Association," Psifchometrica. 44, 1 (March

1979), 43-54.

DeMeyer, A., ], G, Miller. J. Nakane and K. Ferdows, "Flexibility: TheNext Competitive Battle," Strategic Management j . . 10, 2 (1989).135-144,

Dess, G, G, and P. S, Davis, "Porter's (1980) Generic Strategies as

Determinants of Strategic Group Membership and Organizational

Performance,"/4c(J(ieH)i/ of Management /., 27 (1984). 467-488.

Dillon, W, R. and M, Goldstein, Muitivariate Analysis: Methods atid

Applications. John Wiley and Sons, Now York, 1984

Duncan, R, B., "Characteristics of Organizational Environments and

Perceived Environmental Uncertainty," Administrative Sci.

Quarterly, 17(1972). 313-327,

Fahey. L, and H, K. Christensen. "Evaluating the Research on Strategy

Context." /, Matia^enifiit. 12. 2 (1986). 167-183,

Ferdows, K,, J, G, Miller, J. Nakane and T, Vollmann, "Evolving Man-

ufacturing Strategies," liUentaiional}. Operations and Production

Management. (January 1987),

Fine, C. H, and A, C, Hax. "ManufacturingSh-ategy: A Methodology

and an Illusion." Interfaces. 15, 6 (1985), 28-46-

Fomell, C , "Three Approaches to Canonical Analysis," /. Marketing

Research Society. 20, 3 (1978), 166-181,

Galbraith. J,, Designing Complex Organizations. Addison Wesley,

Reading, MA, 1973,

Giffi, C , A. V, Roth and G, Seal, Competing in World Class Manufac-

turing: America's 21s/ Century Challenge, Business One lrwin,

Homewood, IL, 1990,

Ginsberg, A, and N, Venkatraman, "Contingency Perspectives of Or-

ganizational Strategy: A Critical Review of the Empirical Re-

search," Academy of Management Review, 10 (1985), 421-434.

Green, P. E,, Analyzing Multivariate Data, Dryden Press, Hinsdale,.IL, 1978.

Hambrick. D, C . "Some Tests of the Effectiveness and Functional

Attributes of Miles and Snow's Strategic Types," Academy of

Management}.. 1 (1983a), 5-25,

, "An Empirical Typology of Mature Industrial- Product Environ-

ment," ^fdiit'my I'/Manayciru'ii//,, 2 (1983b), 213-229.

and D- Lei, "Toward and Empirical Prioritizahon of Conhngency

Variables For Business Strategy," Academy of Management}., 28,

4(1985), 763-788-

Hamel, G, and C. K- Prahalad, 'Strategic Inleni," Harvard Business

Revieiv, i}7. 3 (May-June 1989), 64-68-

Hdrrigan, K, R., "An Application of Clustering for Strategic Group

Analysis," Strategic Management } . , 10 (1985). 55-73,

Hartigan, J, A,, Clustering Algorithms. John Wiley and Sons, New

York, 1975,

, "Statistical Theory in Cluster," /- Classtfiiation. 2 (1983). 6 3 -

76-Hatten. K, J . D, E, Schendel and A, C, Cwper. "A Strategic Model

of the US, Brewing Industry: 1952-1972," Academy of Manage-

ment } . , 4(1978), 592-609,

Hax, A- C- and N- S. Majluf, "The Corporate Strategic Planning Pro-

cess," Interfaces. 14 (1984), 47-60.

and . "The Concept of Strategy and Strategy Formation

Process," Interfaces. 18 (May-June 1988), 99-109.

Hayes, D. R, and S. C. Wheelwright, Restoring Our Competitive Edge.

John Wiley and Sons. New York, 1984.

Hayes, R, H,, S, C, Wheelwright and Kim Clark, Dytiamic Manufac-

turing. Free Press, New York, 1988,

Hill, T,. Manufacturing Strategy: Text and Cases, R, D, lrwin. Home-

wood, IL, 1989,

Hofer. C, W,. "Towards a Contingency Theory of Strategy," Academy

of Management I.. 18. 4 (1975). 784-810,

and D, Schendel. Strategy Formulation: Analytical Concepts, WestPublishing, New York. 1978.

Hora, S, C, and J B, Wilcox, "Estimation of Error Rates in SeveralPopulation Discriminant Analysis." /- Marketing Research, XIX(1982), 57-61,

Horwitch, M, and R, A, Thietart, "The Effect of Business Interdepen-dencies on Product R&D-Intensive Business Performance," Matt-agenient Sci.. 2 (1987), 178-197,

Huete, L and A. V, Roth, "Linking Manufacturing Capabilities withSBU Strategic Directions." Proceedings of the 1987 Decision SciencesInstitute. 1987,

Kendall, M. G, and A. Stuart, The Advanced Theory of Statistics. 3,Hafner Publishing, New York, 1968,

Kim, L. and Y, Lim. "Environment, Generic Strategies, and Perfor-

mance in a Rapidly Developing Country: A Taxonomic Ap-

proach," Academy of Management /,, 31, 4 (1988), 802-827,

Lachenbruch, P. A. and M, A. Mickey. "Estimation of Error Rates in

Discriminant Analysis," ri'f/ifi(iHit'/)(,-s, 10(1968). 1-10.

Lawrence, P, R- and J, W, Lorsch. Organization and Environment, lrwin,

Homewood, IL, 1969,

302 MANAGEMENT SciENCE/VoI, 40, No. 3, March 1994

MILLER AND ROTH

A Taxonomy of Manufacturing Strategies

Lehmann, D, R., Market Research and Analysis, lrwin, Homewood, IL,

1979,

McDougall, D,, "Learning (he Ropes: How to Tell When You've Found

an Effective Performance Measurement System for Manufactur-

ing," Operations Management Review, 6, 1 (1988), 38-48,

McGee, J, and H, Thomas. "Strategic Groups: Theory. Research and

Taxonomy," Strategic Management /,, 7 (1986), 141-160,

Miles, R. E. and C. C. Snow, Organizational Strategy. Structure, and

Process, McGraw-Hill. New York, 1978,

Miller, D, and P, H, Friesen, "Strategy Making in Context: Ten Em-

pirical Archetypes," /, Management Studies, 14 (1977), 253-280,

and——, "Porter's (1980) Generic Strategies and Performance:

An Empirical Examination with American Data (Part I ) , " Or-

ganizational Studies, 7 (1986a). 37-55,

-and . "Porter's (1980) Generic Strategies and Performance:

An Empirical Examination with American Data (Part II)." Or-

ganizational Studies, 7 (1986b), 255-261.

Miller, J, G,, "Report on the 1982 Manufacturing Futures Survey,"

Manufacturing Roundtabie Research Report Series. Boston Uni-

versity. Boston, MA, 1982,

, A, Amano, A, DeMeyer, K. Ferdows, J, Nakane and A. Roth.

"Closing the Competitive Gaps," In K, Ferdows (Ed.). Interna-

tional Manufacturing. North Holland. Amsterdam, 1989, 153-

168.

, A. DeMeyer and J, Nakane, Benchmarking Global Manufacturing.

Business One-Irwin, Homewood, !L, 1993.

and A, V, Roth, "Manufacturing Strategies," Operations Man-

agement Review, 6, 1 (1988). 8-20,

Miller. J, and T, Vollmann, Manufacturing Futures Sumey Report 1984,

Boston University Manufacturing Roundtabie Research Report,

1984,

Miller, S, and D, Rogers, Manufacturing Policy, irwin, Homewood, IL,

1956,

Milligan, G, W,. "An Examination of the Effect of Six Types of Error

Perturbation on Fifteen Clustering Algorithms." Psychonictrika,

45(1980), 141-160,

and M, C, Cooper. "An Examination of Procedures for Deter-

mining the Number of Clusters in a Data Set." Psi/chometrika, 2

(1985), 159-179.

Mintzberg, H,, "Patterns in Strategy Formulation," Management Sci.,

9 (May 1978), 934-948,

Nanni. A,. ]. G, Miller and T. E. Vollmann, "What Shall We Account

For?" Management Accounting (January 1988). 42-48,

Nemetz, P,, "Bridging the Strategic Outcome Measurement Gap in

Manufacturing Organizations," In ), Ettlie. M, C. Burstein and

A. Fiegenbaum (Eds,). Manufacturing Strategies, Kluwer Academic

Publishers. Boston. MA, 1990. 63-74,

Pegets, C, C, and C, Sekar, "Determining Strategic Groups Using

Multidimensional Scaling," Interfaces. 19, 3 (May-June 1989).

47-57,

Phillips. L, W,. "Assessing Measurement Error in Key Informant Re-

ports: A Methodological Note on Organizational Analysis in

Marketing," /, Marketing Research. 18 (November 1981), 395-

415,

Porter, M,. Competitive Strategy: Techniques for Analyzing Industries

and Competitors, Free Press, New York. 1980,

Rao, C, R,. Linear Statistical Inference, John Wiley and Sons, New

York, 1973,

Richardson. P. R,. A. J, Taylor and J, R, M. Gorden. "A Strategic

Approach to Evaluating Manufacturing Performance," interfaces,

l5(Nov-Dec 1985). 15-27,

Romesburg, H. C , Cluster Analysis for Researchers. Lifetime Learning

Publications. BelmonE, CA. 1984.

Roth, A. V,. "Differentiated Manufacturing Strategies for the Com-

petitive Advantage: An Empirical Investigation." Manufacturing

Roundtabie Research Report Series, Boston University, 1987,

presented at the Annual Meeting of the Decision Sciences Insti-

tute, Las Vegas, NV, 1987,

, "Linking Manufacturing Strategy and Performance: An Empirical

Investigation," Manufacturing Roundtabie Research Report Se-

ries. Boston University, presented at the Academy of Management

Annual Conference. Washington, D.C, 1989,

, A. DeMeyer and A. Amano, "International Manufacturing

Strategies: A Comparative Analysis," In K. Ferdows (Ed.), In-

ternational Manufacturing. North Holland, Amsterdam, 1989,

187-211.

and J. G. Miller, 1987 North American Manufacturing Futures

Survey Fact Book, Manufacturing Roundtabie Research Mono-

graph. Boston University, Boston. MA, 1987,

and , "Manufacturing Strategy, Manufacturing Strength,

Managerial Success, and Economic Outcomes," In J. Ettlie, M. C.

Burstein and A, Fiegenbaum (Eds.), Manufacturing Strategies.

Kluwer Academic Publishers, Boston, MA. 1990, 97-108.

and J, G, Miller. "Success Factors in Manufacturing." Business

Horizons. 35, 4 (1992), 73-81-

User's Guide: Statistics. Version 6 Edition. SAS Institute Inc.,

Cary, NC, 1985.Schendel. D, and C, W, Hofer (Eds,). Strategic Management: A New

Vieic of Business Policy and Planning, Little, Brown & Co,, Boston,

MA, 1979.

Schroeder. R, G,. J. C. Anderson and G, Cleveland. "The Concept of

Manufacturing Strategy: An Empirical Study." /, Operations

Management (August 1986). 405-415,

Skinner, W,. "Manufacturing: Missing Link in Corporate Strategy,"

Harvard Business Review. (May-June 1969), 136-145.

. Manufacturing in the Corporate Strategy, John Wiley and Sons,

New York, 1978,

, Manufacturing: The Formidable Competitive Weapon. John Wiley

and Sons, New York, 1985.

Stobaugh, R. and P. Telesio, "Match Manufacturing Policies and

Product Strategy," Han'ard Business Review (March-April 1983),

113-120,

Swamidass. P, and W, Newell, 'Manufacturing Strategy. Environ-

mental Uncertainty, and Performance: A Path Analytic Model,"

M<3Jm,(s'('»/i'ii(Sci,, 33 (1987), 509-524,

Thietart. R, A, and R, Vivas, "An Empirical Investigation of Success

Strategies for Business along the Product Ufe Cycle," Management

Sci.. 12(1984). 1405-1423.

MANAGEMENT SciENCE/VoL 40, No, 3, March 1994 303

MILLER AND ROTH

A Taxonomy of Manufacturing Strategies

Thompson. J, D,. Organizations in Action. McGraw Hill, New York,1967,

Utterback, J, M, and W, J, Abemathy, 'A Dynamic Model of Process

and Product Innovation," Omega, 3 (1975), 639-656.

Voss, C. A, (Ed.), Manufacturing Strategy. Content and Process.

Chatham & Hall London, 1992,

Ward, P., J. Miller and T, Vollmann, 'Mapping Manufacturers Con-

cerns and Actions Plans," International}. Operations and Produc-

tion Management (Fall 198S), 5-18.

Wheelwright, S. C , "Reflecting Corporate Strategy in Manufacturing

Decisions," Business Horizons (February 1978), 57-66,

, "Strategy, Management and Strategic Planning Approaches,"

Interfaces. 14. 1 (1984). 19-33.

; "Manufacturing Strategy: Defining the Missing Link," Strategic

Management!., 5. 1 (1985),

and R, H, Hayes, 'Competing Through Manufacturing," Harvard

Business Review (January-February 1985), 99-109.

Wood. C , L. Ritzman and D, Sharma. "Intended and Achieved Com-

petitive Priorities: Measures, Frequencies and Financial impact,"

In J, Ettlie, M, C, Burstein and A, Fiegenbaum (Hds), Manufac-

turing Strategies, Kluwer Academic Publishers, Boston, MA. 1990,

225-232.

Woodward, ].. Induslriat Organizathn, Theory and Practice, Oxford

University Press, London. 1965,

Accepted by L. Joseph Thomas, former Departmental Editor; received April 7. 1989. This paper has been with the authors 33 mcitiths for 3, revisions.

304 MANAGEMENT SciENCE/Vol. 40, No, 3, March 1994