Strategic Management Journal, Vol. 17, 21-38 (1996)

LOCAL SEARCH AND THE EVOLUTION OFTECHNOLOGICAL CAPABILITIESTOBY E. STUARTGraduate School of Business, The University of Chicago, Chicago, Illinois, U.S.A.

JOEL M. PODOLNYGraduate School of Business, Stanford University, Stanford, California, U.S.A.

The assumption that 'local search' constrains the direction of corporate R&D is central inevolutionary perspectives on technological change and competition. In this paper, we proposea network-analytic approach for identifying the evolution of firms ' technological positions. Theapproach (I) permits graphical and quantitative assessments of the extent to which firms 'search behavior is locally bounded, and (2) enables firms to be positioned and groupedaccording to the similarities in their innovative capabilities. The utility of the proposedf ramework is demonstrated by an analysis of strategic pa rtnering and the evolution of thetechnological positions of the 10 largest Japanese semiconductor producers fro m 1982 to 1992.

INTRODUCTION

A common assumption- of evolutionary perspec­tives on industrial innovation is that 'local search'significantly constrains the direction of corporateR&D (Nelson and Winter, 1973; Dosi, 1988;Teece, 1988; Cohen and Levinthal , 1989). Thecharacterization of search as ' local' or 'problemis­tic' (Cyert and March , 1963) implies that organi­zations initiate new R&D projects that share tech­nological content with the outcomes of their priorsearches. It seems uncontroversial to assert thatthe notion of 'local search' is relative: the termlocal presumes a broader context of inventiveactivity forming the backdrop against which thesearch behavior of a focal firm can be referenced.However, while the qualifier local has meaningonly when it is paired with the specification ofa broader search context, the literature has yet toprovide a generalizable approach for characteriz­ing this technological landscape and the position sof firms within it.

In this paper, we propo se a network-analytic

Key words: local search, evolutionary theory, techno­logical change, innovation

eee OI43-2095/96/SI002l -1 8© 1995 by John Wiley & Sons, Ltd.

methodology to measure the technological land­scape that is produced by the simultaneous searchactivities of a group of high-technology firms. Afirm' s position or niche in this landscape derivesfrom the overlap of its inventi ve activities withthose of its competitors. In our approach, a firmengages in search when its niche shifts acro sstime periods, and the manifestation of ' localness'is equivalent to the amount of its niche shift.We propo se this approach because it enables asystematic assessment of the extent of interfirm,intertemporal, or interindustry differences in the'localness ' of search.

Our primary objective is to illustrate the meth­odology' s capacity to describe changes in firms'technological positions. However, the approachwe present is relevant to other areas of research,particul arly theories of the resource-based viewof the firm and of strategic groups. In our analy­sis, firms' technological position s der ive from onecompetence that partly shapes their competitivesuccess: the ability to innovate in particular tech­nological subfields. Specifically, we propose arelational construction of technological positionssuch that firms that have developed portfoliosconsisting of similar technologies are located nearto one another. Assuming that firms' abilities to

22 T. E. Stuart and J. M. Podolny

develop technologically similar invention s revealproximities in their underlying 'innovative capa­bilities,' then firms' technological position s in thispaper reflect their innovative capabilities. More­over, clusters of firms with adjacent technologicalpositions cohere because their members havesimilar innovative capabilities, and so they canbe seen as strategic groups. In the discussion, wesuggest that sociologists' notion of a role lendsto our construction of technological position s atheoretical bas is and repre sents a compellingapproach to measuring clusters of firms. Thebenefit of this approach is that sociologists havedeveloped well-establi shed techniques for measur­ing role equivalencies in a network.

The paper is organized according to the follow ­ing plan . The first section discu sses the literaturethat elaborates diverse organizational causes oflocal search. The next section develops the meth­odology for repre senting local search within abroader context. The third section discusses adata source, patent citations, which is used tomeasure firms' technological niches and nicheshifts. The fourth section introduces the empiricalsetting and the sample- the Japanese semicon­ductor industry during a IS-year period. The fifthsection contains the maps of technological po­sitions of the sample members, and it relate sposition in these maps to the market share s,number of patents, and a measure of the innov­ativeness of the sampled firms. The sixth sectionis a discussion that draw s parallels between theapproach developed in this paper and theresource-based view of the firm. The final sectionelaborates implications of and extensions to thisresearch.

LOCAL SEARCH IN R&D

The literature s on evolutionary economics, themanagement of technology and organizationaltheory , all posit that R&D is history dependent.In other words, organizations search for noveltechnologies in area s that enable them to buildupon their established technological base. Thislocal search results from individual and organiza­tional level proce sses, as well as from the natureof the firm' s innovative capabilities.

At the level of the individual decision maker,bounded rationality engenders loca l search whenorgani zation al members fail to consider the uni-

verse of possible applications of R&D funds and,instead, look to the firm' s previous developmentdecisions for guidance. The management of R&Dinvolves investment decisions that must be madein the context of uncertain technical, economic,and social environments in which the actions ofcompetitors are particularly difficult to anticipate(MacKenzie, 1992; Tushman and Rosenkopf,1992). In such ambiguous and uncertain settings,a heavy reliance on historical experience is thenorm (March, 1988). In other words, the resultsof past searches become natural starting point sfor initiating new searches (Nelson and Winter,1982).

At the organizational level, local search is pro­duced by the smooth functioning of organizationalroutines (Cyert and March, 1963; Nelson andWinter, 1982). A routine is defined by Nelsonand Winter (1982: 96 ) to be a pattern of activ itythat is repeatedly invoked. In Nelson and Winter' sschema, routine s generate similar organizationalrespon ses to frequently encountered stimuli, andare therefore the source of continuity in organiza­tional behaviors. The upshot of this conceptionof the firm is that organizational behaviors likeR&D are delimited by the routine s that evolvein a firm. Even when environmental conditionshave decreased the attractiveness of a particularactivity to a firm in possession of a given skillset, intraorganizational politic s and historical pre­cedent can prevent or slow managers from aban­don ing a particul ar technical undertaking(Burgelman, 1994).

Another reason that search is likely to be localis that organizations have a higher likelihood ofsuccessful technology development in areas inwhich they have prior experience. Organizationallearning is a cumulative activit y that is facilitatedby concentrating it in areas of prior knowledgeaccumulation. The competence to innovate in aparticular domain follow s consistent investmentsto develop the facilit ies, personnel , intellectualproperty, intero rganizational relations, and tacitorgani zational knowledge to successfully innovatein that technological area (Teece, 1988). Thismeans that the knowledge stock a firm hasaccumulated in a technological subfield conditionsits return s to R&D investments in that subfield(Cohen and Levinthal , 1989). Therefore, it isnatural to expect that R&D will produce superiorresult s when it is concentrated in the areas of afirm' s establi shed competencies.

Historical, case study, and other empiricalresearch provide scattered evidence to support thehypothesis of local search in many technologicalareas. Even when a major shift in technologystrategy is desired, the literature proposes a num­ber of reasons why firms may have a limitedability to make rapid adjustments. Lee and Allen(1982) showed that one firm required a numberof years to integrate new technical staff, suggest­ing that it may take a considerable amount oftime for organizations to acquire and assimilatenew technological knowledge by augmenting ormaking substitutions in their staff of technol­ogists. There is also evidence to show that high­tech firms do not capriciously shift the marketniches in which they participate. In a study ofthe semiconductor industry , Boeker (1989) foundthat entrepreneurial firms typically maintained thestrategies that they had at the time of founding .Podolny and Stuart (1995) found that semicon­ductor technologies in crowded technologicalareas were the ones most likely to be elaboratedin later periods because they were within reachof the search areas of many firms.

The constraint of local search is also impliedby conceptual frameworks that have highlightedthe difficulties experienced ' by incumbent firmsin adjusting their technology strategies to majorenvironmental changes (Abernathy and Clark,1985; Tushman and Anderson, 1986; Hendersonand Clark, 1990). These studies have discussedand documented the effects of 'competencedestroying' technical changes, which are definedas major technological changes that obviate thetechnical competencies of established firms. Afinding of these studies is that when radical tech­nological developments shift the basis of compe­tition, the path-dependent nature of firms' capa­bilities prevents them from responding quickly.Importantly, such observations do not suggest thatthere is no variation in a firm's technologicaldevelopments, but they strongly imply that afirm' s technical developments do not follow sud­den and unanticipated changes.

This review of the literature has been devotedto establishing the widespread prevalence of theassumption of local search. Nevertheless, itremains the case that the primary empirical evi­dence to support this assumption comes from in­depth case studies of individual organizations orindustries (Abernathy and Clark, 1985; Burgel­man, 1994; Helfat, 1994; Rosenberg, 1969; Sahal,

Local Search 23

1985). Qualitative studies with the firm as theunit of analysis (e.g., Burgelman, 1994) havedocumented the history-dependent quality of cor­porate R&D, while those concentrating on theindustry or technical field (e.g., Sahal, 1985) havetraced the path-dependent nature of industry- orfield-level technical change. Although these stud­ies richly describe organizational learning andtechnological evolution in specific historical per­iods, the methodologies that they employ do notlend themselves to a systematic assessment ofinterfirm, intertemporal, or interindustry variancein the scope of search. I Without a generalizablemethod allowing for such a systematic assess­ment, it is difficult to (I) identify which membersof a group of competitors have been the mostlocally bounded in the outcomes of their R&D, (2) measure the extent to which the searchtrajectories of the members of a group of firmsconverge or diverge over time, or (3) test basichypotheses of how a firm's technological positionat one point in time is contingent on its priorposition and search trajectory.

NICHE OVERLAP AND EVOLVINGTECHNOLOGICAL POSITIONS

Clearly, the appropriate place to look to assessthe degree of path dependence in corporate inno­vation is the actual technological knowledge cre­ated by a firm. In this section, we develop amethodology in which all of the recent inventionsof a group of firms serve as a reference pointfor identifying relative technological shifts ofindividual members . We propose that companieswhich shift technological positions relative totheir competitors are the ones that have movedthe greatest technological distance from the po-

I An alternative approach has been to explore the implicationsof local search in simulation studies (Nelson and Winter,1982; Winter, 1984). Simulation s have typically situated cor­porate search in the context of an abstract space identifiedby standard economic variables, such as input coefficientmagnitudes (Winter, 1984). This approach entails defining thecontext of search as a probability distribution that representsa set of input coefficients in the neighborhood of a firm'scurrent production techniques. In other words, the terrainover which search takes place is an 'economic space ' of inputcoefficients, and the degree to which localness is built intothe model is reflected in the parameters of the search distri­bution. Given the assumptions of this approach, it is simpleto assess interfirm distances and the rate and direct ion of afirm' s movement in 'economic space'.

The community matrix will be denoted At .The measure of overlap that we use produc;sasymmetric competition coefficients in each pairof firms. The elements of the At". matrix for theijth dyad at time t; are defined to be:

where v denotes a technological antecedent, andp indexes the total number of distinct antecedentsthat were foundations for the sampled firms attime 1m • The value aivI is coded I if antecedentv served as a foundation for the inventions offirm i at time 1m, and 0 otherwise; similarly,a j vl

mis coded I if antecedent v is a foundation

for the inventive activity of firm i at time and 0otherwise. Two firms, i and j, produce both anij and a ji cell in the Atm matrix. The ijth cellresults from counting the number of commonantecedents of i and j's inventions at time t.; andthen dividing this sum by the total number ofdistinct technological precursors of firm i'sactivity. Similarly, the jith element results fromtaking the same numerator, but in this case divid­ing by the total number of antecedents of j'sactivity. Clearly, the ijth and jith cells in theAt matrix generally will not be equal to onean~ther; although the numerator is common toboth cells, the denominator in almost all caseswill differ (one exception is if there is no overlapin the antecedents of i and j, in which case bothcells equal zero).

Given that the ijth cell represents the degreeto which firm j is in firm i's niche, it should beclear that row i specifies the extent to which allother firms are in i's niche, and column i specifiesthe degree to which firm i is present in the nichesof all other firms. (Returning to the hypotheticalcommunity matrix of Figure I, row I specifiesthe occupants of firm A's niche and column Iregisters A's presence in the niches of its alters).Taken together, row i and column i define the

24 T. E. Stuart and J. M. Podolny

sitions that they had previously occupied. Theseare the companies that have deviated from locallycircumscribed research.

Our methodology allows each firm to occupya 'technological niche' that emerges from thedistribution of technological antecedents of thefirm's current technology developments (Stuart,1995; Podolny, Stuart and Hannan, 1996). Wedefine the technological overlap between themembers of a pair of firms in terms of the extentto which they build on the same foundations fortheir current inventions. We will use the notation(Xij to denote the proportion of firm i's niche thatis occupied by another firm j: (Xij represents theproportion of inventions built upon by firm ithat are also foundations for the inventions of j.Therefore, (Xij is bounded by zero and one: atzero, two firms are completely differentiated; atone, j fully occupies i's niche.

For a system of N innovators, complete infor­mation about interfirm technological overlaps canbe expressed in an asymmetric matrix of orderN x N (McPherson, 1983; Hannan ard Freeman,1989). The elements of this matrix are called'competition coefficients,' and the matrix itself isknown in the literature as a 'community matrix.'The competition coefficients are simply the (Xij'

(Xji for (j = I, 2, "', N; i = I, 2, ' '' , N; i =;e j).Figure I depicts a hypothetical technological net­

work including three firms, denoted A, B, andC. The figure includes arrows, which representtechnological building relations at the level of thediscrete invention. The arrows are directed fromthe firms to a number of inventions that belong tounidentified actors; each arrow represents the actof building on a discrete invention. For example,four inventions were foundations for A's techno­logies. In the hypothetical network, (XAB is 0.5because B builds on two of the four inventionsthat are foundations for the technologies of A.

Figure I also illustrates the corresponding com­munity matrix for the three networked firms . Thefirst row of this matrix registers the degree towhich each of the companies in the sample occu­pies the niche of firm A. Thus, B occupies 50percent and C occupies 75 percent of A's niche.The first column indicates the extent to whichfirm A occupies the niches of the other membersof its network. Thus, A overlaps with 100 percentof B's and 37.5 percent of C's niche . The maindiagonal has no significance and so it is setto missing.

p

2: a;Vlm a j vlm

>=1

p

2: aiVlm ajvlm

>=1

(1)

(2)

Local Search 25

Corresponding Community Matrix :

A 0 .50 .75

B 0

C .375 .25 0

Figure I. Hypothetical technological network for three firms. The figure illustrates the level of competitivecrowding among three hypothetical firms, A, B, and C. In the figure, the objects of the arrows emanating fromthe firms, the numbered boxes, represent existing inventions . The lines with arrows represent technologicalbuilding relations. For example, firm B has developed inventions that built on inventions 3 and 5. Firm A'srow in the community matrix registers the percentage of its niche filled by Band C. Firm A's column in the

matrix is the percentage of the niches of Band C that it occupies

technological position of a focal finn with respectto all other firms at a particular time tm • In effect,the entries in row and column i define a globalposition for finn i in a 2N - 2 dimensionalspace, where N is the number of firms in thecommunity matrix .

This conception of a finn's global position asa function of the proximities of its technologicaldevelopments to those of the members of a groupof competing firms serves as our point of depar­ture for measuring the technological distancebetween firms in each period of time. In addition,we will use this measure of position to definethe extent of a finn's technological movementacross time periods. Specifically, we define thedistance between i and j at time t.; in terms ofthe degree to which i and j have a similar patternof niche overlap with all other firms k. Formally,the (Euclidean) distance between finn i and j fora given time t.; is defined to be:

]

1/2

+ (akitm - a kjtJ2] ,k "" i.] (3)

where the alphas are the (asymmetric) competitioncoefficients for the ikth and jkth dyads at time t.;Notice that the distance between firms i and j inEquation 3 is a function of the level of the dissimi­larity of their patterns of niche overlap with eachof the other (N - 2) firms in the sample. Thus, (aki/ - akjt ) is the difference in the extent to whichfim';'s i and j occupy the niche of a third finn k,and (aiklm - ajktJ is the difference in the extent towhich the niche of k overlaps the niches of i and j .

Similarly, it is possible to quantify the intertem­poral shift of finn i's technological niche in termsof the degree to which its pattern of niche overlapchanges over time. Formally, we define the shiftin finn i's technological niche from time t, to tm as:

26 T. E. Stuart and J. M. Podolny

where 8 equals I if i = j and I ~ m, and 0otherwise. According to Equation 5, the morethat firm i's pattern of niche overlap with itscompetitors in period t.; is similar to firm j'spattern of overlap with its competitors in period

The expression inside of the sum operator registersthe extent to which firm i's niche overlap with theother (n - I) firms changes between time periodst[ and t.; The more that the pattern of i' s overlapchanges between t[ and t.; the larger will be thesummed expression of Equation 4.2

Equation 3 represents the distance between dif­ferent firms within a single time period. Equation4 yields the amount of a single firm's niche shiftacross time periods (i.e., the distance betweenthe positions occupied by firm i in period t[ andthe same firm in period tm ) . Finally , to representthe distance between different firms in differenttime periods , we construct a symmetric matrix,D, where cell itJtm registers the difference in thepattern of overlap between firm i at time t[ andfirm j at time tm • Formally, we define the elementsof the matrix D:

d itV lm == d jlmill =(: =~r (i [(Ctikll - CtjklJ2k=1

t[, the lower will be the value of cellsd it,j'm' d jl mil(

Equation 5 incorporates the specifications ofEquations 3 and 4. Specifically, when 8 = I, t[

~ tm and i = j, Equation 5 reduces to Equation4, and when 8 = 0 and t[ = tm , Equation 5reduces to Equation 3. Assuming that all firmsare present for all time periods, the dimensionsof the symmetric D matrix are N * T rows byN * T columns, where N is the number of firmsand T is the number of time periods. Given thenested equations, Equation 5 identifies a matrixthat includes three types of information: (i ) thedistance between all firms within time periods,(ii) the distance between each firm and itselfacross time periods, and (iii) the distance betweendifferent firms across different time periods.

Readers familiar with the social network litera­ture will recognize the Euclidean distances ofEquations 3, 4 and 5 as continuous measures ofstructural equivalence. Structural equivalence is ameasure of the extent to which two actors areclosely situated in their network because theyhave similar ties to the other network members.As Burt (1987) observed, the more similar arethe relational patterns of two network members,the greater is their structural equivalence andtherefore the more that one member could substi­tute for the other member in its role relations . Ineffect, our approach defines the context of a focalfirm's search by the technological undertakingsof competing firms. Search can be considered tobe a structural property in that a focal firm'schange in position across periods of time can bedefined by its niche shift between times t, and t.;

Using conventional multidimensional scaling(MDS) routines, it is possible to convert the infor­mation in the D matrix to a graphical representationof interfirm distances. However, it is first necessaryto construct the competition coefficients (the At",matrices) from which technological distances canbe derived. To do this, we use the patent citationsmade by a sample of semiconductor firms.

(5)

(4)

2 The sum in Equation 4 is multiplied by «n - 2)/(n - I»so that the metric is comparable to that in Equation 3. Whenthe Euclidean distance between i and j is measured, thecomparison is across n - 2 other actors. However, when theEuclidean distance between i at time period I, and i at I,. isassessed, there are n - I comparisons. Since there are morecomparisons when i is compared to itself across time period,we deflate the distance by (n - 2)/(n - I) so that the distancethat a firm shifts over time is comparable to the distancesbetween firms within a particular time.

PATENTS AND TECHNOLOGICALLINEAGE

Patents identify inventions because they are onlygranted to products, processes, or designs that areindustrially useful and nonobvious to an indivi­dual who is knowledgeable in the relevant techni­cal field. An important component of the patent

application procedure is the 'prior art' provision .In the United States, previous U.S. patents thatare identified as technological precursors to thecurrent invention are referred to as 'prior art .'The citation process is legally important becauseit limits the claims of a pending patent: legalprotection is awarded only to the technologicalclaims that are not anticipated by the prior art.A number of scholars have noted that patentcitations trace out technological building relation­ships among inventions (e.g., Jaffe, Trajtenbergand Henderson, 1993).

Given that patent citations identify the techno­logical antecedents of a firm's current inventions,we use patent citations to quantify technologicalniche overlaps among a community of innovatingfirms, Recall that a ij was used to represent theextent to which the inventions of firm j sharedantecedents with firm i. Because patent citationsidentify technological building relations, we mea­sure the aij as the proportion of patents cited byi that are also cited by j. For example, if i cites100 patents and j cites 50 of those patents, aijequals 0.5.

SETTING: THE JAPANESESEMICONDUCTOR INDUSTRY

To illustrate the utility of this methodology foridentifying search trajectories, we map the tech­nological positions of the largest firms in theJapanese semiconductor industry. A number ofconsiderations motivated the choice of this set­ting. First, the semiconductor industry is one thatis still very much technology-driven. Semiconduc­tor production involves tremendously complexprocesses (Langlois et al., 1988), and technicaladvances have incessantly driven down the priceand increased the performance of semiconductordevices throughout the history of the industry.For this reason, R&D expenditures are quite high(routinely exceeding 10% of revenues for manyincumbents), and firms' decisions about whichtechnological area( s) to target are critical factorsin determining organizational performance.Second, the Japanese industry underwent radicalchange during the period of the analysis. Ourdata span the period from 1978 to 1992. Althougha few Japanese firms began semiconductor pro­duction in the I950s, they were comparativelyminor players in the global marketplace until the

Local Search 27

late 1970s and early 1980s. Therefore, we observethe evolution of the industry during the intervalin which it achieved global prominence. Finally,a number of detailed books on the Japaneseindustry offer a yardstick against which to com­pare the results of this analysis.

The sample includes the 10 largest Japanesesemiconductor manufacturers: Fujitsu, Hitachi,Matsushita, Mitsubishi, NEe, Oki Electric, Sanyo,Sharp, Sony, and Toshiba. These firms were verti­cally integrated, and there was substantial overlapamong them in their participation in end-use mar­kets. For example, all of these companies pro­duced computers and consumer electronics prod­ucts, and most had telecommunicationsoperations.

The data for the analysis are the U.S. semicon­ductor patents held by each of the 10 sampledJapanese producers. The United States is theworld's largest technology marketplace, and forthis reason non-U.S.-based firms routinely submitpatent applications in the United States. Each ofthe 10 sampled firms are among the largest U.S.patent holders for semiconductor device, design,and process innovations. The semiconductor pat­ents held by these 10 firms were collected forthe period from 1978 to 1992, inclusive .'

Following the preceding discussion, at a time1m the matrix of competition coefficients for thenetwork formed by the 10 sampled firms is alOx lOin which each element registers the extentto which the row firm overlaps with the columnfirm in its patent citations. One measurementissue encountered in computing the At matricesis the length of time during which the c~mpetitioncoefficients specified by Equations 1 and 2 arecalculated. It is unreasonable to define an organi ­zation's technological focus at time t; only bythe inventions that it had patented during theprevious year. We therefore chose to create theAt m matrices from the patent cocitations made by

'Semiconductor patents were retrieved from the MicropatentCD series. This series contains all patents granted in theUnited States since 1976. When a patent is granted. the patentexaminer assigns it to a primary class and subclass. Thepatent is also typically cross-referenced in a number of otherclasses. We identified approximately 2400 patentclass/subclass combinations that included semiconductordevice. design. or process inventions. and we included in thedataset all semiconductor patents held by the sampled firmsthat were either primary-classed or cross-referenced in anyone of these locations . Details of the dataset and a list of the2400 classes are available from the first author.

28 T. E. Stuart and J. M. PodoLny

the firms in the sample during five-year, movingwindows." Our analysis spans the period from1978 to 1992, inclusive. Using a 5-year windowallows us to construct three communitymatrices-each one derived from nonoverlappingyears of data-during the 15-year span of ouranalysis. Thus, we constructed an AS2 matrixthat is a 10 x 10 computed from all of the U.S.semiconductor patents awarded to the sampledfirms during the period from 1978 to 1982,inclusive. Similarly, the An matrix was generatedfrom the patent cocitations in the sample duringthe years from 1988 to 1992.

ANALYSIS

Although assessing search trajectories requiresthat we consider all years simultaneously, webegin by examining each year separately. AsEquation 3 specifies, we construct a separatedistance matrix, D, , for each of the years 1982,1987, and 1992. The three panels of Figure 2show the MDS configurations for each of theseyears. The coordinates for these plots were gener­ated by the MDS procedure in SAS, version 6.09.In all cases, the number of dimensions was setto two , which resulted in reasonably good stresslevels."

Evolution of the technological landscape

As anticipated by the arguments about the path­dependent quality of organizational innovation,the figures suggest a significant degree of stabilityin the relative positions of the firms in the period

4 We selected 5 years because it is roughly the duration ofthe product life cycle in the semiconductor industry. For manytypes of products. five years understates the time interval inwhich the product is manufactured (e.g.• each successivegeneration of computer memory, 64K, 256K, etc., has beenin production for about a decade ). However, 5 years mayoverstate the time period during which a particular product,design, or process is on the leading edge of the technologyin the industry (e.g., the next generation of computer memorychip has arrived approximately every 2.5 years).5 In MDS, stress is a normalized, residual sum of squaresthat suggests the degree to which the resultant configurationagrees with the n-dimensional distance matrix. Stress is oftenknown as a 'badness-of-fit' criterion because higher valuessuggest worse fits. For the figure representing 1982, thebadness-of-fit criterion was 0.086 for two dimensions; for1987, the badness-of-fit criterion was 0.092; and for 1992.the stress level was 0.036. These stress levels are consideredfairly good (Kruskal, 1964).

of analysis. In all three figures, the leading-edgesemiconductor producers are located to the eastof the less technically advanced firms. The firmswith the greatest percentage of their electronicsend-use business concentrated in consumer elec­tronics products are positioned toward the westend of the figures . This is clearest in panel C,which suggests a two-tier structure. The gen­eralists and technological leaders in 1992 wereToshiba, Hitachi, NEC, Fujitsu, and Mitsubishi.In panel C of Figure 2, these firms occupy po­sitions that appear to be differentiated from theirconsumer electronics-oriented competitors.

Although Figure 2 suggests a great degree ofstability in the structure of the industry, visualcomparisons of the panels of the figure are diffi­cult because absolute locations in each of thepanels are meaningless and the range of theaxes of the panels differ. Therefore, the apparentinterfirm distances are not constant across thethree panels. To make intertemporal comparisons,we apply MDS to a pooled distance matrix asspecified by Equation 5.

Quantifying the amount of niche shift

The result of the MDS of the pooled distancematrix is shown in Figure 3, which confirmsthat the pattern of niche overlap in the Japanesesemiconductor industry was indeed quite stable.Evidence of stability comes from the fact thatthe position of a firm at time tm is generally quiteclose to its position at times tm- 5 or tm+5 • Forexample, Sanyo in 1987 is relatively near toSanyo in 1982 and Sanyo in 1992. It is possibleto quantify the amount of movement in a firm'sposition by assessing the change in its column(or row) of the distance matrices specified byEquation 3 at different points in time. One suchmeasure follows this reasoning: if firm i did notchange positions over time, then its distance fromother firms will be relatively stable (the ith col­umn in the distance matrices at two points intime will be highly correlated).

Following Burt (1988), we assess the stabilityof a firm's position across time periods by con­structing a firm-specific covariance matrix inwhich each cell represents the covariance betweena firm's vector of distances to its competitorsacross two time periods (therefore, for 10 firmsand three time periods, we construct a total of10 3 x 3 covariance matrices ). The more similar

Local Search 29

oToshiba

oFujitsu

oNEe

oSharp

Okio

oMitsubishi

Matsushitao

PANEL A: 19820.2.."....-------------------,..-------------,

0.15

0.1 00.05 Sony

o 0-0.05 Sanyo

-0.1

-0.15

-0.2

-0.25-0.2 -0.1 0 0.1 0.2 0.3 0.4

PANEL B: 19870.2

S~yo0.15

0 00.1

NEeFujitsu

0.05

0 Sonyo 0 0

-0.05 Sharp oMitsubishi ToshibaO .

-0.1Matsushita

0OkiO Hitachi

-0.15 I I-0.2 -0.1 0 0.1 0.2 0.3 0.4

PANEL C: 19920.08 ,...

0.06Sanyo

NEe0

Hitachi0.04 Sh~ 0

0.02 0Matsushita

Fujitsu0

0-0.02

-0.04 0 0 0-0.06 Sony Mitsubishi Toshiba

-0.08

-0.1O~i

-0.12-0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4

Figure 2. Technological positions of Japanese semiconductor firms

are a finn's distance vectors across all three timeperiods, the higher is the percentage of variancecaptured by the first factor of a principal compo­nents analysis of each 3 x 3 finn-specific matrix.The results of this analysis are reported in TableI. The findings show that Toshiba experiencedthe least movement: 92 percent of the variancein its position vectors for the 3 years is capturedby the first principal component. On the otherhand, Mitsubishi and Fujitsu were the companiesthat moved the most: the first principal componentcaptured about 75 percent of the variance in the

positions of these two firms, Of all of the sampledfirms, Mitsubishi is the only one for which oneof the three finn years-1992-had a negativeloading on the first principal component. In otherwords, Mitsubishi moved significantly between1987 and 1992.

For two reasons, it is remarkable that the pat­tern of interfirm niche overlap remained so stableduring the interval of our study. First, the natureof semiconductor technology is such that a semi­conductor device generation change is typicallyaccompanied by significant changes in product

30 T. E. Stuart and J. M. Podolny

0.3 -r---- - - - - - - - - --- ------------------------,

0.2o

Hitachi82 Sony82o

Matsushita82NEC82 0

o

oSanyo87

oFujitsu87

OMitsubishi82Hitaehi87o 0 / 0 oSanyo82

Fujitsu82 ..,/ Sharp87ONEC92 ..,/ 0 Sanyo92 0

M· b" h'92 0 <l / Sony87itsu IS I ••••. .• . . . . . . .. . .. . •. . • ...o " 6"............... I>QSharp92

Fujitsu92 NEC87 Mits~b;~h·i·87·0o

Matsushita87 0Matsushita92 0 Sharp82

oHitaehi92

oo Toshiba87

oToshiba92

0.1

-0.10

Sony92Oki82 0

0aki92

0Oki87

0 0.1 0.2 0.3-0.1-0.2

oToshiba8 2

-0.3-0.4-0.3 +-...,....-..,.-~.,..-.-..--~,........,---r~~~~_.__~-.-...,....-..,.-.,.-.,..-...--r-......,---.~~~_.__........~-.-...,....-.__~.,........--,.....,

-0.5

-0.2

Figure 3. Technological positions of Japanese semiconductor firms: 1982, 1987, and 1992

Table I. Results from factor analysis to quantify the amount of firms' niche shifts

Firm Factor I" Loading: 1982b Loading: 1987b Loading: 1992b

Fujitsu 0.75 0.96 0.41 0.79Hitachi 0.91 0.91 0.88 0.98Matsushita 0.80 0.95 0.88 0.84Mitsubishi 0.75 0.95 0.92 -0.70NEe 0.78 0.93 0.82 0.89Oki 0.84 0.92 0.90 0.92Sanyo 0.90 0.97 0.94 0.94Sharp 0.85 0.94 0.85 0.93Sony 0.86 0.95 0.94 0.89Toshiba 0.92 0.98 0.93 0.95

"The first facto r indicates the stability of firm i' s position . The higher the factor . the more corre lated are i's distance vectorsacross time period s."Facror loadin gs indica te the extent of a firm ' s relat ive movement in a give n year.

design s, processes, materi als, and manufacturing(i t has even been the case that new device gener­ations have required more complex productionequipment due to the tighter design rule s of moreadvanced chips ). In addition to the fund amentalchange in the technology from the first year ofour data to the last year, we follow the Japanese

semiconductor industry during the period when itgrew from a comparatively small size to one ofglobal prominence. During the first year of theanal ysis, the value of their combined productionwas under $2 billion; by the last year, the 10sampled firms generated $30.25 billion in semi­conductor sales. Despite the dramatic changes in

the scope of the sampled firms and in the indus­try's technology , the pattern of interfirm techno­logical overlap has remained relatively stable.

Interpreting the configurations

The discussion of movement in different direc­tions in Figure 3 begs the question: What are theimplications to a finn of being located in differentregions of the technology space? Descriptiveaccounts of the Japanese industry (e.g., Kimura,1988; Langlois et al., 1988) help to interpretdifferent neighborhoods of the configurations. Thesemiconductor operations of Matsushita, Sony,Sharp, and Sanyo were catered to their consumerelectronics products businesses. Thus, these com­panies focused on linear integrated circuits anddiscrete devices , and so it is not surprising tofind that they cluster in one neighborhood ofthe configurations (see Figures 2 and 3). Thetechnological leaders of the sample were NEC,Hitachi, Toshiba, and Fujitsu . These firms wereall broad-line semiconductor producers, but theyconcentrated on complex devices such as logiccircuits and MOS memories to support their oper­ations in computing. In Figure 3, these firmsappear to be differentiated from the consumerelectronics product s companies.

Oki and Mitsubishi are interesting casesbecause they do not fit neatly with the technologi­cal leaders or the consumer electronics productsfirms. Oki manufactured telecommunicationsequipment and an array of peripheral equipmentfor data-processing systems and computers.Therefore, its end-use businesses were close tothose of NEe. Nevertheless, Oki possessedneither the breadth nor the level of leading-edgetechnology of NEC, Hitachi, Toshiba, or Fujitsu.For these reasons, Oki occupied a relatively iso­lated position in the technological structure ofthe industry: although the foci of its operationsparalleled those of the technology leaders, itplayed a more peripheral role in the evolution ofthe industry's technology ."

6 For this reason, Oki appears to be less of an isolate inMDSs of a correlation matrix (instead of a Euclidean distancematrix) because correlat ions eliminate scale effects (i .e., thecorrelation between the elements of two firms' rows andcolumns of the commun ity matrix does not reflect difference sin their means). However, we believe that it is undesirableto generate the configurations from correlat ion matricesbecause scale is an important attribute of firms' positions.

LocaL Search 31

Because Mitsubishi was the finn that sold thegreatest percentage of its semiconductor pro­duction on the merchant market (according toKimura, 1988, Mitsubishi consumed only 30% ofits semiconductor production in the mid-1980s),its semiconductor focus was not as strongly tiedto its production of electronic end-use systems.During the 1970s, Mitsubishi focused on discretedevices and integrated circuits for consumer elec­tronics products. However, following a strategicassessment near the end of the decade, Mitsubi­shi targeted semiconductors for computer andindustrial applications and it augmented its capitaland R&D expenditure s (Langlois et al., 1988).Around this time, Mitsubishi moved into theDRAM market, developed complementary MOStechnology, and began to second source Intel'smicroproces sors. Figure 3 suggests that the com­pany succeeded in its strategy. Mitsubishi wasthe single company to exit the group of consumerelectronic s products firms and join the technologi­cal leaders. In Figure 3, the trajectory of Mitsubi­shi' s position shift is highlighted by the arrowsthat display its movement between each of thetime periods.

It is a simple extension of the methodologythat we propose to generate 'egocentric' represen­tations of each finn ' s position. Figure 4 illustratesegocentric perspectives of Mitsubishi' s positionfor each of the three time periods. To generatethis figure, we constructed three 9 x 9 matricesfor the years 1982, 1987, and 1992. The 'space'that these matrices represent spans only Mitsubi­shi's patent citations. In other words, the con­figurations are representations of how the nineother firms in the sample are distributed throughthe areas of Mitsubishi's inventive activities ineach of these years. In the data matrices forFigure 4, the distance between the firms compris­ing any particular dyad (e.g., Sharp andMatsushita) is a function of the level of coci­tations among those two firms, subject to thelimitation that the cocitation must have been ofa patent that was also cited by Mitsubishi. Clus­ters in the panels of Figure 4 represent concen­trations of firms that overlap with Mitsubishi ' sniche in a similar fashion (e.g. , they overlap withMitsubishi in similar technological areas) .

The successive panels in Figure 4 illustratethe significant amount of change in Mitsubishi ' srelative position. In the first panel (1982), NEC,Oki and Sony are isolates: they have no overlap

32 T. E. Stuart and J. M. Podolny

PANEL A: 19820.6-r---------------------------,

0.4

0.2

o

-0.2

-0.4

-0.6

Matsushita 0

NEe00 0 Sony

Sharp 0

Toshiba 0

Fujitsu 0

Sanyo 0

Hitachi 0

1.20.80.60.4

Sony00 0

o 0.2I

-0.2-0.8 I

-0.6 -0.4PANEL B: 19870.6-r----------------..,;::--()-------------,

0.4

0.2 Matsushita 0

0Fujitsu (

NEe 0

-0.2) Sanyo Hitachi 0

-0.4 Sharp 0 Toshiba 0

-0.6-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8

PANEL c. 19920.15

SanyoONEeO

0.1Hitachi 0

0.05 Matsushita0

Sharp00 FujitsuO

-0.05

-0.1 Toshiba0

-0.15 Sony0Oki 0

-0.2I I I I

-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8

Figure 4. Competitors in Mitsubishi's technological space: 1982, 1987, 1992

at all with Mitsubishi, and so they cluster in thefigure. All of the other firms have some overlapwith Mitsubishi , but by and large they do notform any discernible pattern based on finncharacteristics. By the time of the second panel(1987 ), all of the nine companies have someoverlap with Mitsubishi, so there are no longerany isolates. However, there are still no salientcompetitive groupings and companies are rela­tively dispersed in the space. In contrast, in theconfiguration for 1992 the consumer

electronics/broad-line producer distinction is evi­dent along the east-west axis in panel C ofFigure 4. By 1992 the egocentric snapshot ofMitsubishi' s position reveals two general group­ings of competitors: on one side are the consumerelectronics products-focused firms (Sanyo, Sharp,Sony, Oki and Matsushita) and on the other arethe broad-line producers (NEC, Hitachi, Toshiba,and Fujitsu).

Returning to the 3-year configuration rep­resented in Figure 3, it is possible to draw axes

Local Search 33

oSanyo87

o Mitsubishi82

oSanyo82

Sanyo920o

Sony87

oOkin

Toshga82-0.2

0.3Market Share

'"_cco8 '=.. ~Il.O

00.2 Hitachi82

tl

~e,

0.1

00 Toshiba87 Mitsubishi92 0

Fujits'i?92

0 0Toshiba92

Fujitsu87

-0.1

0.30.20.1o-0.1-0.2-0.3-0.4-0.3-+-~~"""""",,-,,--"--""--'--'--r-...,...-...,....-...--.,..L-..---,---.-!---.--,-.....,...J.""'-""""-~,........,---.--.---.--r--r--.-...,--...--,.--,.........--..--.--l

-0.5

Figure 5. 'Regions' in the technological map of the industry

through the figure that associate characteristics ofthe firms with their positions in the configuration.Specifically, directions in an MDS configurationcan be interpreted by regressing variables overthe coordinates of the configuration. Figure 5adds to three regressions over the coordinates;one is for patents, a second is for market share,and a third is for patent citations .' Precisely, thefollowing three regressions were estimated:

Share i'm = (31 . Diml itm + (32 . Dim2 itm

Patents i1m= (31 . Dim 1i lm

+ (32 . Dim2<1m

7 The number of citations received by a patent is a commonlyemployed measure of the commercial and technical importanceof that innovation (Albert et al., 1991). Therefore, the fre­quency at which a firm's patent portfolio is cited is a com­bined (i.e ., summed) indicator of the importance of its individ­ual inventions. All of these variables (sales, patent cites, andtotal patents) are measured as proportion s to prevent escalationin their values simply as a function of time. Thus, sales areincluded as market share, patents as the proportion of allpatents awarded to a focal firm, and cites as the proportionof all citations that are received by the patent portfolio to afocal firm.

where Diml and Dim2 are the MDS coordinatesof firm i at time tm . All three regressions hadsignificant F-values and high coefficients of mul­tiple correlation.

The regression analysis corroborates that thefirms with the largest number of patents, thehighest market share, and the most technologi­cally important inventions are located in the west­ern half of the configuration, angled slightlytoward the north. The proportion of all patentsreceived by the firm is the axis with the gentlestslope relative to the horizontal plane; the pro­portion of patent citations received by a firm isthe adjacent axis; and market share is the steepestaxis. Bisecting each of the regression lines witha perpendicular divides the configuration intohalves. We use this technique to bifurcate thecompetitive space into two regions, which arequite consistent with the descriptive accounts thatdistinguish the technological leaders from theconsumer electronics products firms. The 'core'

34 T. E. Stuart and J. M. Podolny

segment is the one that includes the most inno­vative firms and those with the largest marketshare. We delineate the two-tiered structure bythe bold-faced line in Figure 5 (the market shareaxisj-s-this line segments the industry such thatthe firms with the highest market share arenorthwest of the bold-faced market share axis.

Strategic positions and interfirm alliances

A number of scholars have suggested that thecompetitive position occupied by a firm influencesits strategic behavior. Specifically in the domainof technology strategy, Kimura (1989) arguedthat technological position may explain variationacross firms in their foreign direct investmentactivities. Eisenhardt and Schoonhoven (1996)and Shan (1990) hypothesized that the techno­logical position of firms affects their incentivesand propensities to engage in interfirm strategicalliances.

We briefly consider the relationship betweencompetitive position and alliance behavior. Dur­ing the period of the analysis, an exhaustiveliterature search uncovered 35 alliances involvingsome type of technology exchange among thesemiconductor operations of the firms in the sam­ple." In Figure 6 we illustrate the pattern ofalliances as it relates to the technological po­sitions of the sampled firms. In the figure, thepositions of two firms in 1982 were connectedwith a line if they formed an alliance during theperiod from 1982 to 1986 (e.g ., NEC82 andOki82). Similarly, two firms in 1987 were linkedif they established an alliance between 1987 and1991 (e.g ., Toshiba87 and Hitachi87). Companiesthat formed a partnership in 1992 were connectedfor that year, the last year for which we possessthis data. Bold lines join firms that engaged intwo or more alliances during a time period.

A number of findings emerge from Figure 6.First, it is remarkable the degree to which the'core' firms-NEC, Fujitsu, Hitachi, andToshiba-are central in the alliance network. Intotal, 31 of the 35 partnerships involved one (ortwo) of those firms. In other words, the patternof ties is nearly exclusively core-to-core or core­to-periphery. Moreover, each of the four alliances

8 We coded patent license and cross-licen se. second source,joint ventures, joint product development, and technologyexchange agreement s for this analysis.

among the noncore firms included Mitsubishi in1987. This was the year just prior to the timethat Mitsubishi moved into the region of theconfiguration occupied by the core producers. Arelated observation about the pattern of intercor­porate alliances is that NEC in 1982 and Mitsubi­shi in 1987, two firm-years that were near thecore-periphery border, were particularly activeparticipants in the alliance network. These twofirms participated in the greatest number of part­nerships among all of the sampled firms in allthree years.

Clearly, there is a relationship between positionin the configuration of Figure 6 and the decisionof a firm to participate in the alliance network.In addition to the fact that alliances appear tobridge the core-periphery border or to join corefirms, change in position has a clear relationshipto active participation in the recorded technology­exchange and technology-development alliances.For the first period, the correlation between thenumber of alliances formed by a company andthe amount it moved from 1982 to 1987 is 0.13(not significant). However, for the period from1987 to 1992, this correlation is 0.71 and statisti­cally significant." The high magnitude of thiscorrelation suggests a positive associationbetween the propensity of a firm to form alliancesand the degree to which it innovates in techno­logical fields that are not directly related to thosein which it has developed technologies in thepast (Stuart, 1995, presents more systematic evi­dence of this). The decision to branch out froma firm's existing fields of innovation is the mostlikely source of its movement in the configu­rations.

DISCUSSION: TECHNOLOGICALPOSITIONS, INNOVATIVECAPABILITIES, AND STRATEGICGROUPS

The core imagery that underlies this work is theconception of the technological base of an indus-

9 In a comprehensive data base on strategic technologyalliances. Hagedoorn (1993) reported that the MitsubishiGroup had the highest total number of technology partnershipsamong all firms worldwide . Hagedoorn found that Mitsubishiformed 157 alliances in the 1980-84 period. and 293 alliancesin the 1985-89 period. Hitachi and Toshiba were also amongthe world' s 10 most frequent technology partners .

Local Search 35

0.3.,....------------------------------__-----,

0.2

0.1

o

-0.1

-0.2o

Toshiba82

oSony92

oSanyo82

0.30.20.1o-0.1-0.2-0.3-0.4-0.3 +-~.,...._,........,,........,---r--.--..,.........,...._r_...,...-~.,....-,.L,..........---r--.--..,.......,...........-...,...-~.,....- ..........---r--.--..,.......,...........-...,...-~.,....-,....--.--.---.--1

-0.5

Figure6. Technological positions of Japanese semiconductor firms and strategic alliances: 1982, 1987, and 1992

try as an evolving network. Discrete inventions,which for the most part belong to corporate inno­vators, form the nodes of this network. Techno­logical commonalities among the inventions inthe network are the ties that connect nodes. Inan ongoing research program, we suggest thatpatents and patent citations can be used to rep­resent this expanding 'technological network' fora select number of high-technology industries(Stuart, 1995; Podolny and Stuart, 1995). Col­lecting all patent s in particular technological areasand aggregating firms' inventions allows the com­putation of network-theoretic attributes of firm­level positions in this technological network . Likethe measures and methods used in this study,ideas and techniques in the network literaturelend theoretical insights to the computation andbehavioral implications of different attribute s offirms' technological positions.

In the analysis of this paper , the proximity oftwo firms' positions depends upon the degree towhich they are structural equivalents. Two actorswho are perfect structural equivalents are alsoassumed to perform the same role in the relational

structure in which equivalence is measured (i.e.,they are role equivalents). Generalizing thisinsight to the empirical context of this article,two firms that occupy structurally equivalent po­sitions in the technological network do so becausethey perform similar roles as innovators. Twosuch firms would appear very near to one anotherin the positional maps of this paper. In principle,they could substitute for one another in theirinnovati ve roles.

Assume that the ability to develop the inven­tions that form the basis for a particular inno­vative role rests on the incumbent's accumulationof difficult-to-imitate innovative skills. This is nota Herculean assumption: following the discussionof the broad literature that characterizes inno­vation as a path-dependent process, it is quiteplausible that firms' positions derive from skillsthat are in fact quite difficult for competitors toreplicate quickly . Moreover, it is also the casethat a well-honed innovative capability can be anextremely valuable resource to high-technologyfirms. In this case, the configurations of firmscan be viewed as maps of innovative capabilities.

36 T. E. Stuart and J. M. Podolny

We believe that the analysis in this paper has anobvious link to the resource-based view of thefirm: the configurations of Figures 1-6 representone approach to positioning firms on the basis ofinimitable, valuable resources that are potentialsources of sustainable competitive advantage.

In addition, the analyses of this paper arepertinent to the literature on strategic groups. Ifpositions cohere because the firms that hold themperform similar innovative roles, then clusters offirms can be viewed as grouping based on similarinnovative capabilities. To date, scholars haveusually identified intraindustry group structure bycategorizing firms according to their product mar­ket positions, or else by general descriptors oftheir corporate strategy. However, as a numberof scholars have suggested (McGee and Thomas,1986; Dierickx and Cool, 1989; Barney, 1991),one point of contact between the resource-basedview of the firm and work on strategic groups isto define intergroup mobility barriers in terms ofheterogeneities among groups of firms in theirpossession of strategically valuable resources.

The findings of this paper do suggest thatmobility barriers segregate technological po­sitions. Furthermore, as other researchers haveargued, the technological areas targeted by afirm' s inventions in large measure circumscribethe expertise that it develops in manufacturing,marketing, and other core business functions(Teece, 1988). It is therefore compelling to usesimilarity of technological position as a basis foridentifying groups whose capabilities are not eas­ily imitated .'? To identify groups in the squarematrices of interfirm technical proximity scores(the community matrices), one would apply ahierarchical clustering algorithm to partition thesample members.

10 It is important to note that intergroup mobility barriersmay be asymmetric, even when firms' group affiliations aredetermined by their innovative capabilities. For example , hier­archical cluster analyse s suggest that the firms in the regionlabeled 'CORE' in Figure 5 compri se one group , and thosein the region labeled 'PERIPHERY' form a second group.The actual technological areas that compri se the basis of theinventive activities of the 'PERIPHERY' firms (e.g.• linearICs and discrete devices ) are less complex than those thatare the focus of the 'CORE' producer s (e.g., optoelectronicsand MPUs ). In fact. most of the core firms produce linearrcs and discrete devices. in addition to more complex devices.Therefore. mobility barrier s are asymmetric: it would be easierfor the core firms to move into the periphery region thanvice versa.

CONCLUSION: DIRECTIONS FORFUTURE RESEARCH

The objective of this paper has been to developa generalizable methodology for quantifying theevolution of firms' technological positions. Ourapproach conceptualizes the context of search interms of the actual technologies developed by asample of innovators, and the outcome of searchin terms of its impact on firms' technologicalpositions. From our perspective, an important andunderemphasized component of the dynamics oftechnological change is that firms do not searchin isolation; rather, they search as members of apopulation of simultaneously searching organiza­tions. The methodology that we have suggestedin this paper implicitly recognizes that a firmmay come to occupy a differentiated technologi­cal niche not necessarily as the result of itsown R&D. but as the result of the R&D of itscompetitors. In effect. a firm's position dependsas much on the trajectories adopted by otherfirms as it does on its own trajectory.

A contribution of this research is that it offersa systematic conception of the context of search .The absence of such a method is surprising,particularly considering that the characterizationof the search process is an essential step in theconstruction of evolutionary models of industrydynamics (Nelson and Winter, 1982; Winter,1984). A direct consequence of the lack of ageneralizable approach has been the inability toempirically test the basic assumption of localsearch in a convincing manner . For example,there have been no empirical tests of the Markov­ian assumption that the innovative direction of acompany at period tm+ 1 depends critically on thestate that it occupied at period t.; but not on itsprior history (Nelson and Winter, 1982).Additionally, it has been difficult to understandhow the search environment and the history ofsearch explain current technological positions andconstrain future shifts in innovative directions.

An important influence on the findings of thisstudy was the choice of setting-the semiconduc­tor industry. Semiconductor technology is knownto be cumulative, and because of its great com­plexity the technology is notably domain-specific.With few exceptions, the fact that a firm excelsat innovating or producing in one market nichedoes not imply a similar expertise in a differentniche. With the proposed methodology, however,

it would be possible to make intersample com­parisons. Because the community matrices containinformation on the global positions of all of themembers of a system, metrics of the stability ofthe community matrices are comparable acrosssystems of the same size. For example, it wouldbe possible to compare the 10 largest Japanesefirms to the 10 largest U.S. firms during thesame interval of time to determine which groupexperienced the most change. It would also bepossible to compare, for instance, the communitymatrices representing the 50 largest semiconduc­tor firms to those representing the 50 largestpharmaceutical firms during the same time period.An analysis like this could assess the degree towhich 'localness' characterizes the search tra­jectories of firms in different industries. A priorexpectation would be that semiconductor firmsare substantially more locally bounded in theirinnovation than pharmaceuticals because semicon­ductor technologies are more cumulative than aredrug discovery techniques .

Innovation can be considered to encompass abroad array of technical and commercial func­tions, ranging from basic R&D to marketing.Our concern with firm trajectories in knowledgecreation has led to our focus on the invention­generating stages of the innovation claim . How­ever, we believe that the methodology that wehave presented can be generalized to other stagesof the innovation chain. For example, there existseveral data sources that provide information onfirms' participation in different product marketniches in the semiconductor industry. Using suchinformation and distance metrics like thoseemployed in this paper, it is straightforward tomeasure the distance between firms in productspace, just as we have measured the distancebetween firms in technology space. With thisadditional information, it would be possible tomap evolving market positions and to explore therelationship between market and technologicalpositions.

One of the most suggestive findings of theanalysis is that Mitsubishi 's movement into the' technological core ' was preceded by allianceswith firms in that position. This association sug­gests the possibility that considerable shifts intechnological position are facilitated by efforts toassimilate the technological developments of thefirms in the areas to which a firm seeks to move.Alliances and acquisitions represent possible strat-

Local Search 37

egies to bring about significant shifts in techno­logical focus. One possible direction for futureresearch would be to more systematically investi­gate the effect of alliance strategies and otherstrategic undertakings on the amount and direc­tion of firms' search. In effect, one could investi­gate the impact of alliances or acquisitions onthe distance of a firm's movement as specifiedby Equation 4. Similarly, it would be a simpleextension of this research to model the effectsof organizational characteristics-such as age orsize-on the stability of a firm's technologicalposition. The central question guiding this typeof analysis would be: How do variables thatproxy for the institutionalization of organizationalroutines affect the degree of inertia in the direc­tion of firms' innovation?

REFERENCES

Abern athy , W. J. and K. B. Clark (1985 ). ' Innovation:Mapping the winds of creative destruction ',Research Policy, 14, pp. 3-22.

Albert, M. B., D. Avery, F. Narin and P. McAllister( 1991). 'Direct validation of citation counts as indi­cators of industrially important patents ' , ResearchPolicy, 20, pp. 251-259.

Barney, 1. B. (1991) . 'Firm resources and sustainedcompetitive advantage', Journal of Management,17(1), pp. 99-120.

Boeker, W. (1989 ). 'Strategic change: The effects offounding and history ', Academy of ManagementJournal, 32(3), pp. 489-515.

Burgelman, R. A. ( 1994). 'Fading memories: A proce sstheory of strategic business exit in dynamic environ­ments' , Administrative Science Quarterly, 39, pp.24-56.

Burt, R. S. (1987 ). 'Social contagion and innovation:Cohesion versu s structural equivalence', AmericanJournal of Sociology, 92, pp. 1287-1335.

Burt, R. S. (1988) . 'The stability of American markets ' ,American Journal of Sociology, 88(2), pp. 356-395 .

Cohen, W. M. and D. A. Levinthal (1989). ' Innovationand learning: The two faces of R&D ' , EconomicJournal, 99, pp. 569-596.

Cyert, R. and J. G. March (1963). A Behavioral Theoryof the Firm. Prentice-Hall , Englewood Cliffs, NJ.

Dierickx, I. and K. Cool (1989). 'Asset stock accumu­lation and sustainability of competitive advantage ' ,Management Science, 35(12 ), pp. 1504-1514.

Dosi, G. (1988) . 'The nature of the innovative proce ss ' .In G. Dosi, C. Freeman, R. Nelson, G. Silverbergand L. Soete (eds.), Technical Change and Eco­nomic Theory. Pinter, London, pp. 221-238.

Eisenhardt, K. M. and C. B. Schoonhoven (1996).'Resource-based view of strategic alliance format ion:Strategic and social effects in entrepreneurial firms ',Organizational Science, forthcoming.

38 T. E. Stuart and J. M. Podolny

Hagedoorn, J. (1993). 'Strategic technology alliancesand modes of cooperation in high-technology indus­tries'. In G. Grabher (ed.), The Embedded Firm:On the Socioeconomics of Industrial Networks. Rout­ledge, London, pp. 116-137.

Hannan, M. T. and J. H. Freeman (1989). Organiza­tional Ecology. Harvard University Press, Cam­bridge, MA.

Helfat, C. E. (1994). 'Firm specificity and corporateapplied R&D', Organization Science, 5(2) , pp.173-184.

Henderson, R. M. and K. B. Clark (1990). 'Architecturalinnovation: The reconfiguration of existing prod­uct technology and the failure of established finus',Administrative Science Quarterly, 35, pp. 9-31.

Jaffe, A., M. Trajtenberg and R. Henderson (1993) .'Geographic localization of knowledge spillovers asevidenced by patent citations' , Quarterly Journal ofEconomics, pp. 577-598.

Kimura, Y. (1988) . The Japanese Semiconductor Indus­try: Structure, Competitive Strategies, and Perform ­ance. JAI Press, Greenwich, CT.

Kimura, Y. (1989) . 'Firm-specific strategic advantagesand foreign direct investment behavior of firms: Thecase of Japanese semiconductor firms', Journal ofInternational Business Studies , pp. 296-314.

Kruskal, J. B. (1964) . 'Multidimensional scaling byoptimizing goodness of fit to a nonmetric hypoth­esis', Psychometrica, 29, pp. 1-27.

Langlois, R., T. A. Pugel, C. S. Haklisch, R. R. Nelsonand W. G. Egelhoff (1988) . Micro-Electronics: AnIndustry in Transition. Unwin Hyman, Boston, MA.

Lee, D. M. S. and T. J. Allen (1982). 'Integratingnew technical staff: Implications for acquiring newtechnology', Management Science, 28, pp. 1405­1420.

MacKenzie, D. (1992) . 'Economic and sociologicalexplanation of technical change' . In R. Coombs, P.Saviotti and V. Walsh (eds.), Technological Changeand Company Strategies: Economic and SociologicalPerspectives. Harcourt Brace Jovanovich, London,pp. 25-48.

March, J. G. (1988). Decisions and Organizations.Blackwell, New York.

McGee, J. and H. Thomas (1986). 'Strategic groups:Theory, research, and taxonomy' , Strategic Manage-

ment Journal, 7(2), pp. 141-160.McPherson, J. M. (1983). 'An ecology of affiliation',

American Sociological Review, 48, pp. 519-532.Nelson, R. R. and S. G. Winter (1973). 'Toward

an evolutionary theory of economic capabilities' ,American Economic Review, 63, pp. 440-449.

Nelson, R. R. and S. G. Winter (1982). An EvolutionaryTheory of Economic Change. Belknap Press, Cam­bridge, MA.

Podolny, J. M. and T. E. Stuart (1995) . 'A role-basedecology of technological change', American Journalof Sociology, 100(5), pp. 1224-1260.

Podolny, J. M., T. E. Stuart and M. T. Hannan (1996).'Networks, knowledge, and niches: Competition inthe worldwide semiconductor industry, 1984-1991 ',American Journal of Sociolog y, forthcoming.

Rosenberg, N. (1969) . 'The direction of technologicalchange: Inducement mechanisms and focusingdevices', Economic Development and CulturalChange, 18, pp. 1-24.

Sahal, D. (1985). 'Technological guideposts and inno­vation avenues', Research Policy, 14, pp. 61-82.

Shan, W. (1990). 'An empirical analysis of organiza­tional strategies by entrepreneurial high-technologyfirms', Strategic Management Journal, 11(2), pp.129-139.

Stuart, T. E. (1995) . 'Positions in a patent network: Atheory of the rate and intensity of innovation', Work­ing paper, Graduate School of Business, Universityof Chicago.

Teece, D. J. (1988) . 'Technological change and thenature of the firm'. In G. Dosi, C. Freeman, R.Nelson, G. Silverberg and L. Soete (eds.), TechnicalChange and Economic Theory, Pinter, London, pp.256-281.

Tushman, M. L. and P. Anderson (1986) . 'Technologi­cal discontinuities and organizational environments',Administrative Science Quarterly, 31, pp. 439-465.

Tushman, M. L. and L. Rosenkopf (1992) . 'Organiza­tional determinants of technological change: Towarda sociology of technological evolution', Research inOrganizational Behavior, 14, pp. 311-347.

Winter, S. G. (1984). 'Schumpeterian competition inalternative technological regimes', Journal of Econ­omic Behavior and Organization, 5(3-4), pp.287-320.