Boon Witteloostuijn Caroll (2002)

8/3/2019 Boon Witteloostuijn Caroll (2002)

1/25

Resource Distributions and Market Partitioning: Dutch Daily Newspapers, 1968 to 1994Author(s): Christophe Boone, Arjen van Witteloostuijn, Glenn R. CarrollSource: American Sociological Review, Vol. 67, No. 3 (Jun., 2002), pp. 408-431Published by: American Sociological AssociationStable URL: http://www.jstor.org/stable/3088964

Accessed: 06/10/2010 12:12

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at

http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless

you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you

may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at

http://www.jstor.org/action/showPublisher?publisherCode=asa.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed

page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact [email protected].

American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to

American Sociological Review.

http://www.jstor.org
http://www.jstor.org/action/showPublisher?publisherCode=asahttp://www.jstor.org/stable/3088964?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/action/showPublisher?publisherCode=asahttp://www.jstor.org/action/showPublisher?publisherCode=asahttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/3088964?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=asa


2/25

RESOURCE DISTRIBUTIONS AND MARKETPARTITIONING: DUTCH DAILY NEWSPAPERS,1968 TO 1994

CHRISTOPHE BOONE GLENN R. CARROLLUniversity of Antwerpen, Belgium Stanford University

ARJEN VAN WITTELOOSTUIJNUniversity of Groningen, the Netherlands

Resource-partitioning theory explains how, in heavily concentrated populations,specialist organizations arise and proliferate. The theory also addresses the processof market concentration itself although far less attention has been devoted to thetheoretical claims in this area. In this analysis, the theory is used to explain general-ist concentration through the distribution of environmental (market) resources. It isargued that the higher the homogeneity and concentration of relevant environmentalresources, the higher the concentration of large generalist organizations competingon the basis of scale. Using data on the Dutch daily newspaper industry from 1968to 1994, statistical analyses show that concentration among generalist (national)newspapers occurs morefully in province-level markets where the readership baseconsists of relatively homogeneous sets of individuals in terms of age, religion, poli-tics, and education. At the same time, these concentrated markets prove to be fertileareas for the operation of specialist papers, at least when resources are not fullyhomogenized. The analysis thus provides a more complete model of the resource-partitioning process among organizations in a population.

S OCIOLOGISTS have long been inter-sted in economic or market concen-tration and its consequences for society. Al-though technological, political, cultural,andclass-based theories have all been offered toexplain concentration, most explanationscannot account for the emergence of small

Direct all correspondence to ChristopheBoone, University of Antwerpen, Faculty of Ap-plied Economics, Prinsstraat 13, 2000Antwerpen, Belgium ([email protected]). This research was supportedby the Insti-tute of Industrial Relations, University of Cali-fornia, Berkeley, the Warwick Business School,United Kingdom, and the Dutch Science Foun-dation. We appreciate helpful comments fromWilliam P. Barnett, Bert De Brabander,StanislavDobrev, Michael T. Hannan, Anand Swamin-athan, and the ASR's editors and referees. Wethank Walter Hendriks for his assistance in datacollection and management.

specialist organizations in certain highlyconcentrated industries (e.g., banking, air-line passenger service, and film production).In fact, most theories of concentration denythe possibility of such a development(Sutton 1991).By contrast, resource-partitioning theoryuses spatial imagery to explain the simulta-neous occurrence of both marketconcentra-tion and specialist organizationproliferation(Carroll 1985). Specifically, in markets withstrong scale advantages, the theory assumesthatlarge organizationsaim to maximize de-mand for their products or services. To doso, they target the areas of the market withthe greatest numberof consumer resources.Given a particulardistribution of resourcesin the market, such targetingleads these or-ganizations to adopt generalist market pos-tures, thatis, they offer productsdesigned toappeal to many different types of consum-

408 AMERICAN SocIOLoGIcAL REVIEW, 2002, VOL. 67 (JUNE:408-43 1)


3/25

MARKET PARTITIONING 409ers. Scale competition among such largegeneralist organizations in this type of re-source space leads to concentrationof theseorganizations in the market. Almost para-doxically, increased concentration also en-hances the opportunities for specialist orga-nizations. Because the few surviving gener-alists are squarelypositioned in the center ofthe resource space, specialists can findnewly opened peripheral space, given thereis sufficient heterogeneity (Carroll andHannan 1995, 2000). Resource-partitioningtheory, therefore, contains two components:The first deals with the behavior of general-ist organizations (competition for those ar-eas of resource space with the greatest num-bers of consumers), and the second dealswith the implications of the structural out-come of the first part (i.e., concentration ofgeneralists) on the life chances of specialistorganizations.Organizational sociologists report sub-stantial empirical support for the secondcomponent of the theory. Studies of a widevariety of populations show that as general-ist concentration increases, the foundingrates of specialist organizations rise and/ortheir mortality rates decline. The forms oforganizationsexamined so far include news-paper publishing (Carroll 1985; Dobrev2000), automobile manufacturing (Dobrev,Kim, and Hannan 2001; Torres 1995), beerbrewing (Carroll and Swaminathan 2000),film production (Mezias and Mezias 2000),microprocessormanufacturing Wade 1996),early telephone companies (Barnett andCarroll 1987), medical diagnostic imagingproducers (Mitchell 1995), and winemaking(Swaminathan 1995), as well as services ofbanking (Freemanand Lomi 1994; Li forth-coming), airline passenger travel (Seidel1997), investment banking (Park andPodolny 2000), legal counsel (Jaffee 2000),and financial auditing (Boone, Brocheler,and Carroll 2000). More generally, the riseand proliferation of specialists often spursinnovation and increases product diversityand consumer choice (Carroll and Swamin-athan2000; Peterson and Berger 1975).The first component of resource-partition-ing theory, the part dealing directly withcompetition among large generalists, hasbeen neglected in empiricalresearch(butseeCarroll and Swaminathan2000). So a core

part of the theory still requires testing.Equally important,this neglect occasionallycauses confusion about how the theoryworks. In particular,analysts sometimes donot see that the predictions aboutthe effectsof the concentration variable on the vitalrates of specialist organizations rest on aconflation of ideas about (1) the shape of re-source distributionsand (2) the behavior ofgeneralists. Although this conflation oftenproves helpful in conducting empirical re-search on specialist viability, it does shoveinto the recesses of assumption (implicitlyand often wittingly) the centralrole of envi-ronmentalresource distributions.

Ourmain goal here is to test the first com-ponent of the theory.We do so by modelinghow the distribution of environmental re-sources (in the present case, the distributionof customer tastes) drives concentrationamong generalists. Specifically, we predictthat the competitive struggle among gener-alists for the center of the market's resourcespace will be most pronounced when envi-ronmentalresourcesarehighly (but not com-pletely) concentrated.To our knowledge, nopriorstudy in the ecological traditioninves-tigates the relationshipbetween environmen-tal resources and generalist concentration.Moreover, we believe that this type of re-source-basedanalysis mightbe widely appli-cable to the study of concentration in mar-kets and other contexts.Following the second component of thetheory, we predict that specialist organiza-tional forms will flourish when market con-centration increases. We expect the special-ists to benefit even when the underlying re-source distribution does not change, pro-vided it initially contained sufficient hetero-geneity. A unique aspect of our study per-tains to the specialist outcomes. While mostprior tests of resource partitioning analyzethe founding or mortalityof individual (par-ticularly specialist) organizations,we exam-ine here: (1) a well-established business per-formance measure (i.e., the combined mar-ket share for specialist forms of organiza-tion), and (2) a more conventional ecologi-cal measure (i.e., growth of organizationalmass for the specialist form).Our empirical analysis examines, at theprovince level, Dutch daily newspapersfrom1968 to 1994, an attractivecontext for sev-


4/25

41 0 AMERICAN SOCIOLOGICAL REVIEWeral reasons. First, we can identify andmea-sure the major contoursof the resourcespacewithin which these organizations operate.Specifically, Dutch paperscompetitively po-sition themselves along dimensions of place(national or local) as well as dimensions ofage, education, religion, andpolitical prefer-ence. Second, despite the presence of na-tional newspapers (and thus a national mar-ket), the province-level (local) markets n theNetherlands are relatively distinct. Roughlycomparable with respect to many resourcedimensions, the provinces also have clearand distinguishable local identities. For in-stance, some provinces are predominantlyProtestant,while others aremainly Catholic.In addition, provinces also varyin theirratesand types of resource realignment over thestudy period. This variability gives us con-siderable research leverage because we canstudy the impact of multiple (11 to be exact)environmental resource distributions.Third,there are substantial differences in the levelof newspaper market concentration acrossthe provinces, the initial dependentvariable.

We focus on a single profit-orientedorga-nizational form for research design reasons,not because the theory is limited in this way.The many prior tests of the second compo-nent of resource-partitioning theory make itobvious that the theory applies to a varietyof for-profit organizational forms. Morebroadly, we believe that if the appropriateconditions exist, resource-partitioningideasare potentially applicable to many nonprofitorganizational contexts, including politicalparties (Peli 2001), voluntary associations(PopielarzandMcPherson 1995), sport clubs(Fort and Quirk 1995), religious bodies, andsocial movements. Beyond organizations, ithas been suggested that (parts of) the theoryapplies to productdesigns (Wade 1996) andweb sites on the Internet (Peli andNooteboom 1999). What seems generallycommon across these disparateapplicationsis that the units involved (organizations,products, parties, etc.) compete over re-sources (consumers, members, ideas, etc.) inlarge part through their locations and nichewidths (broador narrow)in a space definedby characteristics of the resources.We develop theoretical argumentsin twosteps, first general and then specific. In thenext section, we explain and develop re-

source-partitioning theory, focusing on therole of resource distributions. We advanceseveral general theoreticalpropositions.Thefollowing section then applies these propo-sitions to the Dutch daily newspaper indus-try. It formulates specific testable hypoth-eses. The next section discusses our researchdesign, measures,and estimation models. Fi-nally, we presentthe empirical findings, dis-cuss them and suggest some directions forfuture research.RESOURCE-PARTITIONINGTHEORYResource-partitioning theory characterizesorganizations in terms of their niche width.A generalist organization targets a broadrange of consumer tastes in the market bymaking products or services with a broadappeal; a specialist organization targets asmall range of very specific customer tastes(Hannanand Freeman 1977).Broadly speaking, the theory explains thesimultaneous occurrence of two organiza-tional trends within a market:generalistcon-centrationandspecialist proliferation.It doesso by positing a specific process of scalecompetition that drives generalistconcentra-tion, and another (related) process of nichediscovery that drives specialist proliferation.Owing to scale competition, large generalistorganizationscome to dominate the market.This occurs because some aspect of productor service delivery in the marketpossesses ascale advantage,meaningthat it is less costlyor less difficult (on a per unit basis) to pro-duce on a largescale. Owing to niche discov-ery, empty spots in the resource space defin-ing the marketbecome populated by new ormobile organizations, implying that the re-source space is saturatedor tightly packed.GENERALIST CONSOLIDATIONLike much organizational sociology, re-source-partitioning heory assumes a marketconsisting of a finite set of heterogeneousresources. For generalist consolidation, akey issue concerns the resource distribution.To clarify, suppose that there are n salientenvironmentalresourcedimensions. In somecontexts, these might be thought of as di-mensions of customer taste preferences (Peli


5/25

MARKET PARTITIONING 411and Nooteboom 1999). Let the joint distri-bution of these n dimensions be unimodaland well behaved.1 In other words, a singlepeak characterizes (roughly) the joint re-source distribution, representing the place atwhich most resources accumulate. We callthis peak the market center.An organization in this space attemptingto maximize its size would find the area nearthe peak attractive because of its resourceabundance. However, effective operation inthe marketcenter requires a particulargener-alist posture: This firm must develop andmaintain appeal at the highest point of eachof the dimensions simultaneously. Yet thegeneralist organization hat does secure a toe-hold in the dense center may obtain an ad-vantage over other competitors. This is be-cause the higher scale of activities sustain-able in this location leads to lower per-unitcosts, which can be passed along to custom-ers in the form of lower prices or better prod-ucts or services. So, in a unimodal resourcespace, generalism andscale go hand in hand.Accordingly, competition among generalistorganizationsin the center often triggers anescalating scale-driven war for resources.Consider the level of dispersion aroundthe peak. Imagine two hypothetical spaces,one with resources tightly dispersed aroundthe peak, and a second one with resourcesbroadly dispersed. It seems clear that in the

1 The originalniche-width heoryof Hannanand Freeman 1977) assumes hat organizationsface a singleresourceconditionat any pointintime. Intertemporally,nvironmental esourcesand conditionsare assumed o be disjointedorhighlydissimilar Peli 1997). So, in this theory,organizationsacehighlydissimilar nvironmen-tal resourcestates that alternateover time. Be-cause of the dissimilarity in environmental con-ditions, a generalist organization straddling twodifferent resource pockets or conditions pays aprice in terms of overheador excess capacity. Asdiscussed here, resource-partitioning theory usesa different assumption about environmental re-sources. In this theory, organizations confrontseveral resource conditions at the same point intime, and these conditions are not assumed to bedissimilar. When environments are not so dis-similar, generalists may not be burdened by thestraddle-across-resourceconditions, as in origi-nal niche-width theory. In fact, they may actuallybenefit from participationin more than one envi-ronmentalresource location, as we arguehere.

first case, a generalist positioned near thecenter need not cover as wide a range ofspace along the various dimensions toachieve the same potential scale as would ageneralist in the second case. In the narrowlydispersed resource space, a generalist cangain, and possibly protect, a large quantityof resources by staying within the narrowpeaked area. In the widely dispersed re-source distribution, a generalist would haveto cover much more resource space in orderto gain the same scale.Much organizational theory suggests thatthis second generalist faces a much moredaunting task. The market appeal it makesmust be broader; he associated tensions andcosts will likely be higher, making its returnlower. It also confronts a greater number ofplaces where challenging competitors mighttake hold. So a more homogeneous or con-centrated resource base creates the condi-tions for a generalist to accumulate evenstronger advantagesand thereby begets con-centration,which gives:Proposition 1: Given a unimodaldistribution

of environmental resources, the higherthe homogeneity and concentration ofresources, the higher the concentrationof generalist organizations competingon the basis of scale.Anecdotal evidence suggests that in news-paper markets satisfying the conditionsspecified in Proposition 1, generalist behav-ior often conforms to the expected patterns.For example, in the American local newspa-

per industry, most cities evolved over thepost-WorldWar II period into one-newspa-per industries, where the paper gaining anearly size advantage was able to ride it toeventualdominance(Rosse 1980). This trendcoincides with an increased homogenizationin cultural tastes (Peterson and Kom 1996).And in modem-day Miami, a city among themore ethnically diverse, an opposing trendhas recently appeared:The dominantMiamiHerald has experienced problems due inlarge partto thedifficulty of adequatelyserv-ing its diverse potentialreadershipbase.2 AsSwartz (1999) explains:2 It is noteworthy that other large ethnicallydiverse cities still have major competing dailies(e.g., New York andChicago), or did until recent


6/25

412 AMERICAN SOCIOLOGICAL REVIEWIt is generally believed that over the last de-cade the Herald ... has lost both its nationalstature and much of its influence in the localcommunity. Circulationhas dropped by overa hundredthousand.... [T]he Herald's big-gest problem was a lack of creativity and vi-sion, which was most obvious in the paper'sattempts to engage Miami's immigrantpopulation, the largest of any Americancity.(P. 37)

SPECIALIST PROLIFERATIONSpecialist organizationsemerge largely as abyproduct of generalist consolidation. Incompeting for the abundant center, largegeneralists eventually outdo their smallercompetitors because of the primacy of scaleadvantages. When the smaller generalistsfail, their target markets become free re-sources. Generalists occupying adjacent re-gions hold the best positions for securingthese newly available areas, and they typi-cally do so. The surviving generalists thusbecome larger in size and more general inscope as time passes. However, because ofthe ever-widening range of the survivinggeneralist's target area, it becomes increas-ingly difficult to secure the entire free area.This is because doing so involves uncer-tainty, might prove very difficult or morecostly than it is worth, or might entail theloss of some of the organization's existingtarget area because its identity or capabili-ties would be undermined.3These develop-ments seem especially likely in maturepopulations where generalists become verylarge, possessing extremely broad target ar-eas, and the free resource space is thin andlocated on extreme values. So, as the com-petitive struggle among generalists proceedsto its eventual small-number equilibrium,the organizationalsize and target breadth ofthe survivors increase, but the combined re-sources held by all generalist organizationsdeclines somewhat (Carroll 1985; Carrolland Hannan 1995, 2000).

A central prediction comes from compar-ing the amount of space available for discov-ery and populationby specialists when over-all concentration rises. Because concentra-tion derives exclusively from generalist con-solidation (i.e., there are no large specialistsin this resource space), this comparison canbe made by measuringthe totalresourceareaoutside generalist targets under differentstages of the generalistcompetitionscenario.As explained, this area (total combinedspace outside generalist targets) comprisesmore space when concentration is high(fewer and larger generalists), ceteris pari-bus. The theory holds that as this space in-creases, the viability of specialist organiza-tions also increases. That is, the concentra-tion process among generalists creates theconditions for the emergence and spread ofspecialists even when the underlying re-source distribution remains unchanged.So the rise of specialist organizations in-volves entrepreneursdiscovering and popu-lating the "residual"resource space that liesoutside the generalist targetareas. It is here,away from the intense competitive pressureof the dominant large generalists, that spe-cialist organizationscan find viable locations(Carroll 1985). And because resources tendto be thin in these regions, the specialists lo-cated there are typically small. Small highlyspecialized locations are also less likely tobe invaded by the ever-encroaching general-ists than arebroader ocations, and they tendto be more defensible if they areinvaded(be-cause of the generalist limitations of inertia,identity, and scope). This basic predictioncan be specified with respect to either found-ing ratesormortalityrates(the specialist seg-ment can expand as a result of changes ineither or both rates). In either case, however,the empirical implication is an interactioneffect between organizational form (special-ist-generalist status) and concentration on avital rate. This implies:Proposition 2: Under conditions for resource

partitioning, as generalist concentrationrises, the mean viability of specialist or-ganizations increases and the mean vi-ability of generalist organizations de-creases.4

decades (e.g., Los Angeles).3 An economist might characterize this situa-tion as involving no economies of scope, or pos-sibly even "diseconomies" of scope, between theactivities producing success in the center of themarket and those producing success in the pe-ripheries. 4 For the generalists, the proposition appearsto some as tautological, even when it concerns


7/25

MARKET PARTITIONING 4A13

The (necessary but not sufficient) condi-tions for the operation of resource-partition-ing processes include: (1) a finite and het-erogeneous resource environment; (2) ascale-based process of competition amonggeneralist organizations; and (3) the pres-ence of inertial forces operating on the tar-get ranges of generalists and specialists,thereby limiting adaptability.In the currentanalysis, we focus on testing the first condi-tion, which can be stated as:Scope Condition 1: The positive impact ofgeneralistconcentration on specialist vi-ability (Proposition 2) occurs only if the

resource space is sufficiently heteroge-neous.

APPLICATION TO THE DUTCHNEWSPAPER INDUSTRYORGANIZATIONAL NICHE WIDTHThe population size of the Dutch newspapermarketcomparesclosely with the metropoli-tan area of New York:In 2001, the Nether-lands' population totaled approximately 16million inhabitants. A number of regionswith clear and distinguishable identitiesmake up the Netherlands, however. By andlarge, regional identity is reflected in the 11provinces, each of which has its own admin-istration and history. As Stinchcombe(1990:136) notes, "socially organized mar-ket segments carry different information,"and the organizational structures that de-velop to serve them typically reflect thesedifferences.Daily newspapers in the Netherlands canbe meaningfully classified into two very dif-average viability. However, in empirical researchthis problem can be avoided by using a measurefor concentration that does not cover the entiregeneralist population. It is also importantto rec-ognize that the process of scale-driven competi-tion theorized actually implies different effectsfor generalists depending on where they sitwithin the size distribution-large generalistsbenefit while smaller ones suffer. Carroll andSwaminathan (2000) show that it is possible tomodel the process of generalist scale competitionmore precisely when certain types of data areavailable. For our present purposes, these issuesdo not come into play in a central way.

ferent organizational forms. On one hand, alimited numberof national newspapers op-erate in every province. These papers seek toattractreadersfromMaastricht o Den Helder(from the extreme southeastof the country tothe extreme northwest) and from Groningento Vlissingen (extremenortheastand extremesouthwest, respectively). That is, the nationalnewspapers target the whole Dutch reader-ship audience. These papers are generallyknown for their quality andtheir political andreligious positions (or explicit lack thereof),even if they have been broadened and soft-ened over time. On the other hand, a largenumber of regional newspapers restrict theirpotential reading audiences to local residen-tial populations. Regional newspapers focuson preselected communities thatlie within oracross provinces such as De Achterhoek (aregion within the province of Gelderland),Brabant (a Roman Catholic province),Friesland (a province with a non-Dutch lan-guage [Frisian]) or Twente (a region withinOverijssel).The two types of newspapers display dif-ferentnews profiles: National papers publishgeneral national and international news,whereas the regional papers emphasize thecoverage of specific local news. So, nationalnewspapers can be seen as generalists be-cause they try to maximize their appeal tothe potential readershipall over the Nether-lands; their niches include every province(potentially providing strong scale advan-tages). Provincial or regional newspaperscan be considered specialists in thatthey tar-get smaller resource areas (i.e., the smallerprovinces or regional communities). Theydifferentiate themselves from national pa-pers by focusing on (local) news not pro-vided by the nationals. The appeal of thesespecialists outside their target region is ex-tremely low. So a Dutch daily newspaper'sgeographical breadth-national in orienta-tion or confined to a single province-is agood way to define its basic niche width.5

5 Of course, a regional newspaperadopting,forinstance, extreme positions on other resource di-mensions, such as religion and politics, would beeven more highly specialized. The fact of thematteris, however, that it is unusual to find sucha paper, as it would suffer serious disadvantagesin competition against either a general locallyoriented newspaper (e.g., agnostic to politics and


8/25

414 AMERICAN SOCIOLOGICAL REVIEWUsing data obtained from an annual indus-try survey of the Dutch NewspaperAssocia-tion (NDP), we found empirical supportforthis classification scheme in evidence per-

taining to scale economies.6 On a voluntarybasis, newspapers provided informationabout their cost structure. Simple analysisallowed us to look for scale economies in animportantcost area: press costs.7 Press costsinclude all the costs of collecting news, andthey makeupa high proportionof total news-paper costs. Journalists and editors are thekey competitive assets of any newspaper-day after day, they provide a newspaper'scontents and press costs. These data (avail-able upon request) show that the averagepress cost systematically declines as a func-tion of circulation size, thus strongly sug-gesting economies of scale. This is true forboth national andregional dailies-but aver-age press cost tends to be higher among theprovincial (specialist) newspapers. There-fore, the data reveal that the national (gener-alist) papers possess stronger scale advan-tages, as resource-partitioning theory as-sumes.RESOURCE DIMENSIONS OF THEPROVINCE-LEVEL MARKETSWhat are the relevant environmental re-source dimensions for Dutch daily newspa-pers?8 First, at the most basic level, peoplediffer by age in terms of theirability and pro-pensity to read a daily newspaper. At least atthe extremes of the very young and the veryold (Bogart 1981), the number of potentialdaily newspaper readers in a geographicalarea depends on its age distribution. Areaswith disproportionately high numbers of

children will have fewer potential readersthan areas with a similar population size butwith fewer young persons. In addition, thoseareas with age distributions skewed in onedirection or another are likely to support pa-pers with slightly different contents: Read-ers' content preferences vary with age(Bogart 1981).Newspaper readers in the Netherlandscommonly choose papersbased on theirpo-litical preferences. Some papers are knownfor their right-wing political postures, whileothers have leftist or centrist reputations.The choices made by subscribers undoubt-edly reflect a wish to keep informed aboutparticular political positions (and parties).However, the broader contents of paperssuggest also that the motivation may reflecta wish to view larger events from withinthese particularperspectives. In any event,the choice often involves an act of (at leastimplicit) solidarity or endorsement, eventhough this is not the only reason for read-ing a paper.Religious orientation represents anothermajor preference dimension for newspaperreaders in the Netherlands. Although theDutch are known for Calvinism, millions ofDutch citizens belong to the Roman Catho-lic Church. In fact, religious preference inthe Netherlandsfollows, by andlarge, a geo-graphic pattern. Protestantism dominates inthe northernprovinces (though with RomanCatholic enclaves), while Catholicism domi-nates in the southern provinces that borderBelgium (i.e., Brabant and Limburg). Reli-gion constitutes a main component of one'ssocial identity, thus, many papers endorse aparticular aith.Finally, newspapers often target reader-ship bases with different education levels.This is particularly clear in the case of"high-brow"versus "low-brow" papers. Inmany countries, low-quality tabloids targeta less educated readership with spicy "hu-man interest" ssues, which vary from celeb-rity gossip to sports stories and crime events.Although British-style tabloids are not ac-tive in the Dutch newspaper market, somedailies clearly are tailored to a less educatedreadership.Othernewspapers maintain high-quality profiles by emphasizing backgroundinformation about, for example, high-browculturalevents andpolitical arguments.In so

religion) or a political or religious newspaperwith a geographically broader base (i.e., na-tional).6 Although analysts widely accept scale econo-mies in newspaperpublishing as fact (e.g., Rosse1980), actual demonstrationof their existence is

rare because of the difficulty of obtaining suchcost data.7 Given the limited time period and much miss-ing data, sophisticated statistical analyses couldnot be conducted.8 Peli and Nooteboom (1999) have demon-strated that the more such dimensions are salient,the more space there is for specialists.


9/25

MARKET PARTITIONING 41 5doing, these dailies attract a highly educatedreadership. Hence, our analysis includes theprovince-level distribution of different edu-cation categories as a relevant resource di-mension.Envision the joint distributionof the fourtypes of environmentalresources (age, poli-tics, religion, and education) in a province.The surface of this distributioncan be con-ceptualized in five dimensions, with re-source "density"as the fifth dimension. Thedensest spot in the distributionhas the great-est number of potential readers-it is herethat the strongest scale advantage mightarise. Ourdata show that this distributionisroughly unimodal in a typical province.Given such a distribution, those newspa-pers that position themselves to cover thepeak are best situated with respect to scalepotential from the environment, implyingthat their products appeal to readers in thelargest combined category of the four re-source dimensions. Moreover, a newspaperoperatingin only one province can tailor itsproducts to this distribution regardless ofwhere the peak is located. These are special-ists as we have defined them here. It is easyto see why they might show evidence ofeconomies of scale in this context: Biggerpapersare likely to sit on, or to be closer to,higher peaks.9By our definition, a generalist daily news-paper operates in all the provinces simulta-neously. If these areas display radically dif-ferentresourcedistributions,then generalistswould likely need to produce a very differ-ent paper for each province. In this case,generalists would not have much potentialadvantage over specialists and might evensuffer a disadvantage. Rather, the potentialscale advantagefor a generalist arises whenthe resource distributionsin at least some ofthe provinces display peaks at roughly simi-lar points: The similarity across marketsal-lows for common appeals to a largerreader-ship base. Despite differences with respectto the local identities of the provinces, thisis indeed the case for Dutch dailies; our data

9 Of course, in many contexts, we would notexpect specialists to show economies of scale.The pattern here results from our focus on thedaily newspaper market and our restriction tospecialist dailies.

show that the peaks of province-level mar-kets do possess rough similarities, on aver-age. It occurs among lower-educated (edu-cation dimension), nonreligious (religion di-mension), politically centrist (political di-mension) persons between ages 18 and 64(age dimension).Dispersion around the resource peaks alsomatters greatly. Suppose all provinces havevery narrowly dispersed resource distribu-tions. Then a generalist positioned near thepeak need not cover a wide range of spacealong the various dimensions in order toachieve scale advantage. Instead,it can gain,and readily protect, a very large readershipby keeping its target within the narrowpeaked area.Comparethis with the situationin which provinces have widely dispersedresource distributions.Here generalists havemuch more resource space to cover in orderto gain the same scale. This is undoubtedlydifficult for a single organization to do, as itsmarketappealmustbe broaderandthe placeswhere competitors might take hold are morenumerous. In general, the less concentratedthe resource distributions in the provinces,the fewer readers who can be served effec-tively with such an approach and the lesslikely the marketwill concentratedown to asmall numberof largeplayers. Conversely, ahomogenous resource base sets the stage fora generalist to accumulateeven strongerad-vantages and thereby begets concentration.HYPOTHESESWe specify hypotheses in terms of newspa-per structureand behavior in a province (thelevel at which we conduct our empiricalanalyses). Exploring the province level al-lows us to examine the relationshipbetweennational(generalist)andregional (specialist)papersin environmentsthatvary in their re-source dispersal. Based on the reasoningabove, we expect that:Hypothesis 1: Increasedresourcehomogene-

ity along any dimension of age (la), re-ligious background (lb), political pref-erence (1c), or educational level (Id) ofthe province's population is associatedwith increased national (generalist)newspapers' concentration at the prov-ince level.


10/25

416 AMERICAN SOCIOLOGICAL REVIEWWe also expect further concentration bythe generalists when the multidimensionaljoint resource distribution is concentratedaswell. This implies:

Hypothesis 2: Increased resourcehomogene-ity in terms of the joint distribution ofage, religious background, politicalpreference, and educational level of theprovince's populationis associated withincreased national (generalist) newspa-pers' concentrationat the province level.The familiar resource-partitioning argu-ment of Proposition 2 also should hold here.It implies effects of market concentrationon

a variety of specialist outcome variables re-lated to viability. Because, in the period westudy, few daily papers start and fail, we ex-amine conventional business outcomes re-lated to short-termperformance.We predict:Hypothesis 3: At the province level, in-creased national (generalist) newspa-pers' concentration is associated withincreased regional market share(3a) andcirculation growth (3b) of the popula-

tion of regional (specialist) newspapers.Taken together, the hypotheses suggest aprocess whereby increased resource homo-geneity leads to generalist consolidation,which in turn begets specialist viability. Ofcourse, the process may be more complex,yet still operateaccordingto the hypotheses.In the limit case of a fully homogeneous en-vironment, generalist concentration cannotgo hand in hand with the success of special-

ist organizations because some minimal re-source heterogeneity is required to sustainthe specialists (Proposition 2 and ScopeCondition 1). So, somewhat paradoxically,although resource homogeneity drives gen-eralist concentration, at the same time itmight hamperthe proliferation of specialists.To investigate this issue, we distinguishbetween the "Randstad"and the rest of thecountry. The Randstad is the large horse-shoe-shaped densely urban city chain ofAmsterdam-the Hague-Rotterdam-Utrechtin the West, covering the provinces of North-Holland, South-Holland, and Utrecht. Therest of the country consists of two distinctregions: (1) the predominantly ProtestantEast andNorth, which include the provincesof Drenthe, Friesland, Gelderland,

Groningen, and Overijssel; and (2) the pre-dominantly Catholic South, which consistsof the provinces of Brabant, Limburg, andZeeland.The Randstad is the political and culturalcenter of the Netherlands; the headquartersof all national newspapers are also locatedthere. So, in terms of identities, there is lessof an operative distinction between "na-tional" and "regional"in the Randstadthanthere is in the rest of the country. In otherwords, what is national is in fact often alsolocal in the Randstad.The main political andcultural events of the nation often occur inthis region, and the papers that reportthese

events are locally staffed and published. Soa national paper in the Randstad also hasconsiderable local appeal, unlike papers inthe rest of the country. Of course, there arealso events of purely local interestwithin theRandstad, and there are provincial papersthat focus on them. But the blurreddistinc-tion in this locality between national and re-gional identity means that on this critical(defining) dimension, resourceheterogeneityis actually lower in the Randstad.This dif-ference is reflected in the national newspa-pers' contents, which devote more attentionto local Randstad news than to local newsfrom other provinces.There is also less difference within theRandstad than elsewhere in the readershipprofiles (in terms of the welfare of the read-ers) between national and regional newspa-pers. Table 1 uses data from the newspaper-level readershipprofile surveys of the indus-try statistical office (Central Bureau forNewspaper Publicity of the Dutch Newspa-per Industry[CEBUCO]); surveys have beenconducted on a regular basis since 1965.Table 1 shows the average percentage ofnewspaperreaders n each of the welfare cat-egories. It also shows the absolute differ-ences from the average profile of nationalsfor regional papersappearing nside and out-side the Randstad,respectively. Analyses ofvariance of these profile differences (con-trolling for the year to which the differencespertain) clearly reveal that the profile dis-tance is significantly smaller inside theRandstadthanin the rest of the country.These observations suggest strongly thatnational papers are more likely to competehead-on with the regional newspapersin the


11/25

MARKET PARTITIONING 417Table 1. Percentage Distribution, Absolute Differences, and Analysis of Variance ComparingRegional and National Newspapers by Reader Welfare: Netherlands, 1968 to 1994

ReaderWelfareRelatively Lower Middle Higher Relatively NumberMeasure Poor Class Middle Class Wealthy of Cases

Average Percentage of ReadersRegional newspapersappearing 12.51 43.21 34.95 9.32 223outside the Randstad (4.21) (5.45) (8.10) (5.47)Regional newspapersappearing 11.00 39.89 35.84 13.27 148inside the Randstad (4.48) (7.65) (8.25) (7.11)National newspaperreference 7.86 31.73 40.42 19.99 371

groupb (1.67) (3.47) (6.38) (6.64)Average Absolute Differencec

Regional newspapersappearing 4.86 11.33 6.57 10.13 223outside the Randstad (3.04) (4.93) (4.31) (3.19)Regional newspapersappearing 3.61 8.81 5.06 7.64 148inside the Randstad (3.27) (6.3) (5.25) (4.47)

Analysis-of-Variance n Difference Scoresfor Each WelfareCategoryF-value of factor for year 4.80* 5.60* 4.64* 17.19*F-value of factor outside versus 17.61* 14.94* 5.26* 61.70*inside the RandstadR2 .18* .21* .17* .44* -

Source: CentralBureau or NewspaperPublicityof the DutchNewspaperIndustry CEBUCO).Note: Numbers n parenthesesare standarddeviations.The resultspertain o 371 regional newspaper-yearobservations or which welfareprofiledata are available.a Rows total to 100 percent.b Average profileof nationalnewspaperswhenwelfareprofiledataforregional newspapersareavailable.c Difference is between the welfare profile of a regional newspaper n year t and the averagewelfareprofileof nationalnewspapers n yeart.*p< .05 (one-tailedtests)

Randstad, even as the market concentrates.Outside the Randstad such competition isless direct, as regional newspapers empha-size unique local (i.e., non-Randstad andnonnational)news. Thus, resource-partition-ing processes shouldoperatemoreforcefullyoutside the Randstadthan within it:Hypothesis 4: Hypothesis 3 is more likely tohold outside the Randstad than withinthe Randstad where the national-re-gional identity distinction is less sharp.

DATA AND METHODSOBSERVATION PLANOur study includes every independent dailynewspaper that operated during the period1968 to 1994 in (any subset of) the 11 Dutchprovinces. We classified a daily newspaper

as independent if an independent editorialboard produced it. Thus, a newspaper con-sisting of several chain newspapers but pro-duced by the same editorial board was re-garded as one organization. As describedabove, we also made a distinction betweennationalandregional newspapers. Statisticalanalyses were conducted at the provincelevel, using data for every year from 1968 to1994. Observations are thus structured aspooled cross-sections and time-series. Be-cause there are 11 provinces in the Nether-lands, the resulting databaseconsists of 297province-yearobservations (27 years and 11provinces).MODEL SPECIFICATION AND ESTIMATIONTo test the hypotheses, we used a simple re-cursive system of two equations:


12/25

418 AMERICAN SOCIOLOGICAL REVIEW

Pit = f(Cit, Zit, Uit), (2)where Cit s a variable measuring generalistconcentrationin province i at time t, Rit s avector of environmental resource variables,Zit is a vector of control variables, Pit s ameasure of specialist segment performance,and eit and uit are disturbance terms. Obvi-ously, the first equationmodels the behaviorof the national newspapers in terms of gen-eralist concentration; it tests Hypotheses 1and 2. The second equation models the com-bined specialist performance; it tests Hy-potheses 3 and 4.DEPENDENT VARIABLESGENERALIST CONCENTRATION. We mea-sure national newspaper concentration bythe four-firm concentration ratio (generallyindicated as C4), which is calculated as thesum of the circulation of the four largest na-tional newspapers in a province divided bythe total circulation size of the nationalnewspapers in the province.10 To constructthis and other market-size-based variables,we relied on annual newspaper circulationinformation obtained from the yearly publi-cations of the industry's statistical officeCEBUCO (various years). Unfortunately, orthe years 1969 and 1971 circulationsize wasmissing for most of the newspapers. So esti-mates of circulation in these two years wereobtained by interpolation.

SPECIALIST PERFORMANCE. We use twospecialist performancemeasures: (1) marketshareof the regionals (calculated as the sumof the circulationof the regional newspapersin the province divided by the total circula-tion of all newspapers in the province) and(2) growth of regional circulation (regional

circulation/lagged regional circulation, de-noted by growth). We do not treat regionals'circulation growth and regionals' marketshare as separateoutcome measures of spe-cialist viability. Specifically, we attempt tomodel their interrelationship,as we assumethat there is an upper bound to circulationgrowth at high market share.INDEPENDENT VARIABLESRESOURCE DISTRIBUTION VARIABLES.Coveringthe four basic dimensions of news-papers' resource bases, the environmentaldistribution variables describe a province'sage structure,political preferences, religiousbackgrounds,and education profile.We summarize the age distribution by thepercentage of persons in each of three cat-egories: those younger than 19, between 20and 64, and older than 64.11 These data are

10 It is important to stress that the concentra-tion variable for the national papers (Ci,) usesonly the circulation data of national papers; itmeasures the extent of concentrationamong onlythose papers classified as national. The regionalmarket share and circulation growth variablesmeasure the extent to which regional papershavepenetrated the full market. The two types of mea-sures are not inherently interdependent: What-ever marketshare the national papers hold mightbe evenly distributedacross many papersor con-centratedin one or a few.

1 Note that for each resource dimension wefocus on the relative distribution of the popula-tion over the different categories (i.e., the per-centage of people in each of the resource catego-ries) rather than on the magnitude of the re-sources in each category. We do so because thetheory is about the consequences for the fate ofgeneralists and specialists of different environ-mental resource distributions, conditional uponenvironmental (center and periphery) munifi-cence. That is, we ask, provided that the differ-ent resource spaces are large enough, how do thedistributive features of the key environmentalre-sources influence (1) the industry's structure (interms of concentration and density) and (2) thegeneralists' and specialists' performance? Themagnitude argument adds this threshold condi-tion explicitly. That is, a market niche must besufficiently large to offer opportunitiesfor viablenewspaper operation, both in the industry's cen-ter and in the market's periphery. In the Dutchcontext, all provinces are beyond this thresholdlevel of carrying capacity, implying that viableregional niches are in place throughout the dataset. So the analyses can safely focus on resource-partitioningtheory's key hypotheses as to the dy-namic generalist-specialist interaction in suffi-ciently munificent resource spaces. To summa-rize, the theory is about the effect of different re-source distributions,given a certainenvironmen-tal carrying capacity. We can show empiricallythat the concentration models do not work wellwhen the magnitude of these differentiated re-sources are used as the independent variables(i.e., the number of people in each category in-


13/25

MARKET PARTITIONING 419taken from the Regionaal StatistischZakboek(CentraalBureau voor de Statistiek[CentralBureau of Statistics, CBS], variousyears). The market"center"of this distribu-tion falls-not very surprisingly-in the agecategory from 20 to 64.12The second and third resource dimen-sions-political and religious preferences-arerepresentedby variablesbased on votingbehavior. We obtained the distribution ofvotes for different parties at national elec-tions from Statistiekder Verkiezingen CBS,various years). These elections normally oc-cur every four years (i.e., when the govern-ment is not forced to resign earlier). As vot-ing behavior typically changes incremen-tally, we deemed it appropriate o interpolatethe voting behavior for the intermediate pe-riods without elections. We use two sets ofcategorical distinctions for the political andreligious resource distributions. The firstvariable describes the distribution of votesacross the political orientation of the partiesin the election, measuring the percentage ofvotes for partiesin each of the following fivecategories: extreme left, left wing, center,right wing, and extreme right. The secondvariable describes the distribution of votesacross four categories of religious orienta-tion for the parties participatingin the elec-tion: Roman Catholic, Protestant, Christian(i.e., targeting Catholics and Protestantsalike),13 and nonreligious. Then, we calcu-late the percentageof people voting for par-ties in each of these four groups. Note thatthe center of the marketon these two distri-butions lies within the nonreligious and cen-ter-partycategories, respectively.14

The fourth resource dimension-educa-tion-summarizes the distribution of thehighest educational level achieved by theemployed population in the province: low(primary school), low-medium (lower sec-ondary school), medium (higher secondaryschool), and high (higher education-e.g.,university). These data were collected formseveral sources: Volkstelling (CBS 1960,1971), Arbeidskrachtentelling (CBS 1977,1979, 1981, 1983, 1985), and EnqueteBeroepsbevolking (CBS 1990, 1991, 1992,1993, 1994). Estimates of the percentages inthe four education categories were obtainedby linearinterpolation or years for which nodata are available. The center of the marketon this distribution lies within the low edu-cation category.We summarize the four environmentalre-source distributionswith Herfindahl indices.These indices are calculated by summing thesquareof the proportionof cases falling intoeach category of the focal distribution. Ahigh score indicates that the resource distri-bution is concentrated into one category(maximum score = 1.0). When the score islow (minimum = 1/n, where n = the numberof categories), concentration is low, imply-ing that the diversity of resourcesin theprov-ince is high. The dataon the resource dimen-sions cannot be broken down into cross-clas-sified categories. Regrettably,we cannot findsource documentsto providenumbersof per-sons by age by religion by politics by educa-tion (or even by any two of the dimensions).So, although we cannot construct a defini-tive measure of the joint distributionof re-sources, we do analyze the effects of overallresource concentration, using an index cal-culated by averaging the four Herfindahlin-dices (labeled resource-concentrationindexin the remainderof the paper).15' 16

stead of the percentage). Note thatwe control forthe impact of carrying capacity by including aprovince-level measureof carrying capacity (i.e.,number of inhabitants).12 For each resource dimension, we define thecenter as the category with the highest observedmaximum duringthe period under study.13 A Christian party is more general than its

Catholic or Protestantcounterpartsbecause it tar-gets Christian voters regardless of whether theyare Catholic or Protestant (cf. the Dutch Chris-tian Democratic Appel [CDA], which has a pro-file similar to the German ChristianDemocraticUnion [CDU]).14 The classification of partiesby political ide-ology likely relates to their identities in terms of

religion. For instance, relatively fundamentalistreligious parties are likely to belong to the rightwing. Conversely, leftist parties probably do notstress any religion at all. Indeed, the correlationsbetween the categories of the political and reli-gious orientation of the parties reveal these ten-dencies. For instance, Protestant and extremeright (r = .88, p < .05, and N = 297) are stronglyrelated as are nonreligious andleft wing (r = .77,p < .05, and N = 297).

15 Although the lack of cross-classified datameans that we cannot observe potential cleav-


14/25

420 AMERICAN SOCIOLOGICAL REVIEWCONTROL vARIABLES. The first pair ofcontrol variables in the models describesgeneral social and economic conditions ofthe province. We controlled for the pro-

vince's newspaper carrying capacity by mea-suring the province-level population in mil-lions. We expect that larger carrying capac-ity facilitates the resource-partitioning pro-cess because the sustainability of specializa-tion between regional and national papers (interms of Adam Smith's division of "labor")is likely to be limited by the size of the mar-ket. So a large carrying capacity might fa-cilitate both concentration of the nationals aswell as performance of the regionals. Theeconomic condition of the provincewas con-trolled by including the number of province-level unemployed inhabitants (in 100,OOOs).This variable is an important indicator ofwelfare. When times are hard, one might ex-pect the circulation of regional papers to fallas people cut back on expenses. Similarly,generalist newspapers, less sensitive to thecutting back of newspaper consumption,might become dominant in hard times, in-creasing generalist concentration.These data

come from a variety of sources. The data oneconomic conditions at the province level(unemployment) were obtained fromRegionale Gegevens overArbeid (CBS, vari-ous years). Unemployment information wasnot available for the years 1968, 1969, and1970. These gaps were closed by linear ex-trapolation. The Regionaal StatistischZakboek (CBS, various years) was used toobtain the number of inhabitantsper prov-ince.The second pair of control variables per-tains to province-level densities of nationaland regional newspapers. It is importanttohave a C4 measurethat is independentof thenumber of generalists because resource par-titioning requires assessing the distributionof market sharesover generalist newspapersgiven a constant number of nationals. Obvi-ously, we expect the effect of the province-level density of national newspapers on C4to be negative. The province-level density ofregional newspapers might hamperthe con-centration of national papers because itprobably increases the competitive pressureof the regionals on the nationals (over andabove the intense competition among the na-ages, this may not be problematic for three rea-sons. First, the resource-concentration index ei-ther over- or underestimates the actual concen-tration of the joint or multidimensional resourcedistribution. As we explain below, however, ineach case, the over- or underestimation of re-source concentration works against our hypoth-eses, thus making our tests conservative. Notethat by simply averaging the four Herfindahl in-dices, we implicitly assume that the distributionof cases over categories in one resource dimen-sion is constant over the different categories inanotherdimension. This is, of course, not realis-tic. When minority groups on one dimension areoverrepresentedon anotherdimension, the aver-age index will, as a result, overestimate resourceconcentration. In that case, the "true"multidi-mensional resource distribution will have asmaller center (and might even be polymodal),which hampers concentration of the generalists(at relatively high overestimated levels of re-source concentration). Conversely, we underesti-

mate resource concentration when the center onone dimension is overrepresented on anotherdi-mension. In that case, the "true" multidimen-sional resource distribution will have a largercenter, facilitating concentration of generalists(at relatively underestimated levels of resourceconcentration). Taken together, in both cases ourestimates of the effect of resource concentration

on concentrationof the generalists are in all like-lihood conservative. Second, if the dimensionsline up (which we cannot observe), the types ofcleavage are likely to be similar over differentprovinces and time periods (e.g., it is likely thatpeople over age 64 are overrepresented in the"right-wing" voting category and that this is thecase for each province throughoutthe sample pe-riod). As a result, if the concentration index isless reliable because of the absence of cross-clas-sified data, it tends to be so in a systematic way,making comparisons between provinces and overtime possible. Third,we show estimates based onthe resource-concentration ndices for illustrative(not definitive) purposes, as well as to check therobustness of our findings.

16 Because the theoretical value of theHerfindahl indices depends not only on the dis-tribution of the proportions but also on the num-ber of categories, we computed a weighted aver-age of the four Herfindahl indices to obtain anoverall resource-concentration index weightedfor differences in the number of categories. Thecorrelationbetween the weighted andunweightedaverage is very high (r = .99, N = 297). As it doesnot matter for the findings which average we use,we report results only on the unweighted re-source-concentrationindex.


15/25

MARKET PARTITIONING 421tionals). High between-group competitionmight make it more difficult for a few na-tional newspapers to become dominant.In the specialist performancemodels, highnational newspaper density in the provinceis likely to increase competitive pressure onthe regional papers and as a result might re-duce the growth of circulation and marketshare of the regionals. Similarly, the pres-ence of many national papers in the provincemight reduce the number of untappednichesavailable for regional newspapers, therebyreducing the performance of regional news-papers. However, high density of nationalpapers might also increase within-groupcompetition among the nationals and facili-tate the performanceof regional newspapers.When assessing specialist performance, it isagain important to predict regional marketshare and circulation growth given a con-stant number of regional papers appearing nthe province. The presence of many regionalpapers suggests a fragmented market, mak-ing it difficult for the regionals to becomedominant as a group. Conversely, high re-gional newspaper density might also reducethe relative competitive pressure from thenationals and as a result facilitate the perfor-mance of regional papers. To summarize, theultimateeffect (positive or negative) of bothnational and regional newspaper density onspecialist performance s difficult to predict.This is because high relative density of a fo-cal organizationalform might indirectly re-duce the competitive pressureof other forms(i.e., between-group competition) while atthe same time causing more competition be-tween members of the same focal form (i.e.,within-group competition).The thirdcluster of control variables cap-tures three specific firm-level events. Thefirst two events have to do with the dateswhen two relatively large national newspa-pers formally announced their decision tostop nationwide distribution after a long up-hill battle associated with an incrementalprocess of step-by-step change. These eventshappened in 1972 and in 1982. To modelthese shifts, we used two from-national-to-regional dummy variables. In 1972 a na-tional newspaper (Het Vrije Volk)decided towithdraw from nationwide circulationto be-come a regional newspaper in the Randstad(South Holland, in the area of Rotterdam).

As a result, this dummy variable (1 = after1972; 0 otherwise) is likely to have a posi-tive effect on both generalist concentrationand specialist performance (the latter espe-cially in the Randstad). In 1982, another na-tional newspaper (Courant Nieuws van deDag) decided to withdraw from nationwidecirculation and to focus on North Holland(region Amsterdam). Again, we expect apositive effect of this dummy variable (1 =after 1982; 0 otherwise) on both generalistconcentration and specialist performance(the latter again especially in the Randstad).The third event is a from-national-to-re-gional transformation.An independent na-tional newspaper (Het Algemeen Dagblad)also produced two regional editions in theprovince of Drenthe (from 1968 to 1973).These editions were chain papers and werenot independent newspapers. These chainpapers became an independent regionalnewspaperin Drenthein 1974 (Drentse Cou-rant). This change caused a huge jump in themarket share of the regional newspapersanda decrease in the C4 of the nationals in thatprovince. So we expect this dummyvariable(1 = for 1968 through 1973; 0 otherwise inDrenthe) to be positive in the generalistcon-centration models and negative in the spe-cialist performancemodels.'7'18

17 A methodological issue and a theoretical is-sue have been raised about the use of these pe-riod dummy variables. On methodologicalgrounds,one might claim thatthese dummy vari-ables are not necessary because the effects ofthese changes act directly via measured vari-ables. Nevertheless, detailed analyses of the er-rors of the estimated equations reveal that inclu-sion of these dummy variables is essential foreconometric reasons because dropping themcauses huge outliers in the errors. This is espe-cially the case when the Drenthe dummy is ex-cluded in the specialist performance equations:The average error s three times as high when thedummy variables are dropped. Analyses of theerrors clearly show that these shifts are not cap-tured by the other independentvariables, makingit necessary to include a dummy variable to con-trol for this "measurementerror."The other twoevents also lead to smaller sudden shifts in thedata that we controlled for in order to increaseprecision of estimation.The theoretical complaint is that the threeevents indicate adaptive behavior of individualnewspapers and therefore contradict an assump-


16/25

422 AMERICAN SOCIOLOGICAL REVIEWTable 2 gives descriptive statistics for thevariables. It shows considerable variationacross provinces in the level of national pa-per concentration as well as in specialist vi-

ability.ESTIMATIONFor estimation, we used the Feasible Gener-alized Least Squares (FGLS) estimator n thestatistical package STATAwith the assump-tions of first-order autocorrelation (commonto all panels) and heteroscedasticity amongpanels. We assumed no cross-sectional cor-relation.These assumptionswere justified inmodel-fit comparisons with estimates ofboth smaller (less complex) and largermod-els. Because with FGLS coefficients are es-timated by applying ordinary least squares(OLS) regression on data corrected for

Table 2. Descriptive Statistics for Province-Level Variables: Netherlands, 1968to 1994

Variable Mean (S.D.)Concentration of national news- .84 (.07)papers (C4) in the provinceAge

0 to 19 30.96 (4.47)20 to 64 57.20 (3.50)>64 11.83 (1.80)

Religious PreferenceaRoman Catholic 5.27 (11.36)Protestant 9.24 (9.20)Christian 23.35 (14.14)Nonreligious 60.77 (8.67)

Political PreferenceaExtreme left party 5.28 (2.62)Left party 37.18 (6.79)Center party 32.16 (8.76)Right party 15.76 (4.13)Extreme rightparty 8.26 (5.54)

EducationLow 25.48 (13.39)Low-medium 32.22 (5.67)Medium 28.39 (12.97)High 13.91 (5.57)

Age concentration index .44 (.02)Religion concentration index .48 (.06)Political concentration index .29 (.04)Education concentration index .31 (.03)Resource-concentration index .38 (.02)Density of regional newspapers 7.36 (3.38)in the provinceDensity of national newspapers 8.92 (1.08)

in the provincePopulation (in millions) 1.27 (.86)Unemployment (in l00,OOOs) .26 (.25)Dummy variable for Drenthe .02 (.14)Dummy variable for "after 1972" .85 (.36)Dummy variable for "after 1982" .48 (.50)Market share of regional news- .64 (.16)papers (t) in the provinceMarket share of regional news- .64 (.16)papers (t- 1) in the provinceCirculation growth of regional 1.02 (.10)newspapers in the province

Sources: Data sources described in the text onpages 418 to 422.Note: Numbers in parentheses are standard devia-tions; N = 297.a Measured by voting behavior.

tion of inertia. However, organizational ecologydoes not argue that organizations never try toadapt and change. The argument is that adapta-tion of core features is extremely difficult be-cause of uncertainty and inertia, and therefore islikely to decrease subsequent performance(Carroll and Hannan 2000). The difficulties as-sociated with core changes seem to apply to thetransformations from national to regional of HetVrijeVolk and CourantNieuws van de Dag. Bothcases can be typified as permanently failing or-ganizations, and the changes did not revitalizethem. Before Het Vrije Volk decided to becomeregional, its circulation droppedfrom 283,276 in1968 to 219,108 in 1971. The year of the corechange, circulation dropped further to 163,999and never recovered. In fact, it steadily decreasedto a low of 76,410 when the paper finally disap-peared in 1991. The same story goes for CourantNieuws van de Dag. In 1972, its circulationreached a peak (183,765), declining to 138,139the year before the core change. In 1982, its cir-culation was only 97,750. Again the core changewas not successful, as circulation graduallydropped to 49,757 in 1994. Apart from notingthat these core changes did not ameliorate thefate of the papers, such events were the excep-tion and not the rule, which is also consistentwith the inertia argument.18 Another control variable that would be anobvious candidate to include is the price of thenewspapers. However, in the Dutch newspaperindustry, price competition is impotent as a re-sult of (legal) cartel-like price agreements thatbasically fix the price differences between all thedailies, locally and nationally.


17/25

MARKET PARTITIONING 423autocorrelation and heteroscedasticity, R-squares cannot be reliably interpreted(Kmenta 1986). Instead, we used and reportWald chi-square statistics, which comparethe goodness of fit of the specific modelwith a model containing only a constant. Be-cause we expected the control variables tobe important in explaining both generalistconcentration as well as specialist perfor-mance, we used the same control variablesin both equations.FINDINGSTable 3 presents the FGLS estimates of theconcentration of the national newspapers asa function of resources in the province.Model 1 uses the noncenter categories ofeach resource distribution as the focal inde-pendent variables (thus omitting the centercategories); Model 2 highlights the marketcenter categories (omitting the categories onthe tails).The estimates agreewith expectations: Thesign for every noncentercategoryis negative,whereas every centercategory has a positivesign. In addition,most estimated coefficientsfor both Models 1 and 2 are statistically sig-nificant. In the case of the political prefer-ence and education variables, this is true forboth center and noncenter categories, sug-gesting that these variables are very impor-tant in driving concentration. In the othercases, the nonsignificanteffects are confinedto only two categories, implyingthat thejointeffect of the full set of distinctions of eachvariablesuggests that inclusion is warranted.So the age, the religious background,the po-litical preferences, and the educational leveldistributionsof a province's populationseemto influence the concentrationof the nationalnewspapers at the province level.19

To analyze the robustness of the findings,we reestimated the equation using singleconcentrationmeasuresof each resource dis-tribution rather than the categorical vari-ables. That is, we used Herfindahl indices ofage demography,voting behavior, religiousbackground, and educational level distribu-tions to measure resource concentrationwithin a province. The estimates are pre-sented in Model 3 in Table 3. Consistentwith the results of Models 1 and 2, the coef-ficients of the concentrationmeasures of theage, political, and educational distributionsare positive, as expected (although not sig-nificant for the education variable). The ef-fect of religion concentration is, however,negative (and not significant), which contra-dicts the pattern of results from Models 1and 2. Apparently, the Herfindahl indices inModel 3 are only positive when the centercategories of the resource distributions arepositive and significant. Thus, focusing onrobust findings only, we conclude that diver-sity within a province in terms of age demo-graphics, political orientation, and educa-tional level reduces the extent of marketconcentration among the national newspa-pers within a province. Note that the fit ofthe models using Herfindahl ndices is lowerthan for Models 1 and 2 (as shown by Waldchi-square goodness-of-fit measures). Thisis, however, not surprisingbecause the cat-egory variables contain more relevant re-source distribution information than do thecompressed indices. In addition,this dropingoodness of fit is in our view not theoreti-

19As mentioned above, the political and reli-gious orientations of the parties are correlated.To explore the consequences of this relationship,we reranthe concentration models with the reli-gious categories while droppingthe political cat-egories, and vice versa. It turnsout that the coef-ficients for thereligious categories change appre-ciably when the political categories are dropped.Thus, the religious resource variables only con-tribute to explaining concentration in the ex-pected direction net of the effects of the politicalresource variable. If we drop the religious cat-

egories, however, the findings related to the po-litical categories remain virtually the same. Wealso reran a model to test for the effect of the av-erage of the four center categories of Model 2.The coefficient of this average is close to zeroand not significant. Generally, these additionalanalyses show that the models are most convinc-ing when the categories of each of the four dis-tributionsare included simultaneously. Based onthese findings, we conclude that: (1) The moredetailed and rich the description of the resourcedistribution,the better the prediction of the gen-eralist concentration level; (2) it is the uniquecontribution of each of the resource dimensionsthat makes the difference; and (3) the politicalresourcedimension seems to be primary n its ef-fect, suggesting it is the strongerorganizing prin-ciple.


18/25

424 AMERICAN SOCIOLOGICAL REVIEWTable 3. FGLS Coefficients Predicting Concentration of National Newspapers in the Province:Netherlands, 1968 to 1994

Model 1 Model 2 Model 3 Model 4Independent Variable Coef. (S.E.). Coef. (S.E.) Coef. (S.E.) Coef. (S.E.)Age

0 to 19 -.008*** (.002)20 to 64 .007***(.002)> 64 -.004 (.003)

ReligiousPreferenceaRoman Catholic -.009***(.003)Protestant -.007***(.001)Christian -.007***(.001)Nonreligious .001 (.001)

Political PreferenceaExtreme left party -.013***(.002)Left party -.011***(.001)Center party .003***(.001)Right party -.008***(.001)Extreme right party -.007***(.001)

EducationLow .001* (.000)Low-medium -.003***(.001)Medium -.001***(.000)High -.005*** (.001)

Age concentration index .646** (.231)Religion concentration index -.053 (.042)Political concentration index .325***(.057)Education concentration index .090 (.065)Resource concentration index .368** (.124)Density of regional newspapers -.002** (.001) -.001* (.000) -.001 (.000) -.001 (.001)in the provinceDensity of national newspapers -.018*** (.003) -.020*** (.003) -.020*** (.003) -.022*** (.003)in the provincePopulation (in millions) -.008 (.006) .008 (.006) .017** (.006) .018** (.006)Unemployment (in 100,000s) .017 (.011) .029** (.011) .026** (.011) .021* (.011)Dummy variable for Drenthe .127***(.012) .132***(.010) .129***(.010) .123***(.010)Dummy variable for "after 1972" .046***(.006) .024***(.005) .028***(.005) .016***(.005)Dummy variable for "after 1982" .018** (.007) .015** (.005) .020***(.005) .016***(.005)Constant 2.393***(.161) .369** (.133) .579*** .110) .847***(.063)Wald chi-square 889.0*** 658.1*** 649.5*** 585.8***

Sources: Data sources described in the text on pages 418 to 422.Note: Numbers in parentheses are standard errors; N = 297.a Measured by voting behavior.*


19/25

MARKET PARTITIONING 425Table 4. FGLS Coefficients Predicting Market Share of Regional Newspapers in the Province:Netherlands, 1968 to 1994

Model 1 Model 2 Model 3IndependentVariable TotalCountry RandstadOnly Outside RandstadOnlyConcentration f nationalnewspapers -.078 -.764*** .243***(C4) in the province (.064) (.235) (.068)Density of regionalnewspapers .001 .001 -.003**in the province (.001) (.002) (.001)Density of nationalnewspapers -.002 -.019 .002in the province (.004) (.013) (.004)Population in millions) -.079*** -.008 .026**(.010) (.016) (.010)Unemployment in 100,000s) .003 -.003 -.036*(.018) (.037) (.017)Dummy variablefor Drenthe -.410*** NAa -.459***(.013 (.014)Dummyvariablefor "after1972" .019** .007 -.000(.006) (.018) (.006)Dummy variable or "after 1982" .003 .036* -.008(.006) (.018) (.006)Constant .817*** 1.229*** .519***(.075) (.269) (.072)Waldchi-square 1,602.02*** 23.29*** 1,454.14***Numberof cases 297 81 216

Sources: Data sources described n the text on pages418 to 422.Note: Numbers n parenthesesare standard rrors.a "NA" ndicatesnot applicable(Drenthe ies outside the Randstad).p


20/25

426 AMERICAN SOCIOLOGICAL REVIEWTable 5. FGLS Coefficients Predicting Circulation Growth of Regional Newspapers in the Province:Netherlands, 1968 to 1994

Model 1 Model2 Model3IndependentVariable TotalCountry RandstadOnly Outside RandstadOnlyConcentration f nationalnewspapers .102 -.127 1.274***(C4) in theprovince (.092) (.278) (.187)Marketshareof regionalnewspapers -.087 -.176* -2.319***(t- 1) in the province (.055) (.095) (.136)Density of regionalnewspapers .001 -.001 -.019***in the province (.001) (.004) (.004)Densityof nationalnewspapers .008 -.020 .046**in the province (.011) (.020 (.017)Population in millions) -.011 .005 .022

(.013) (.016) (.033)Unemployment in 100,000s) -.025 -.035 .035(.034) (.050) (.064)Dummy variablefor Drenthe -.271 NAa -1.466***(.167) (.082)Dummyvariable or "after 1972" -.000 -.035 -.004(.018) (.035) (.027)Dummyvariablefor "after1982" -.019 -.025 -.084***(.015) (.031) (.026)Constant .929*** 1.427*** 1.386***(.144) (.363) (.272)Waldchi-square 25.57** 11.16 420.06***Numberof cases 286 78 208

Sources:Data sources described n the text on pages 418 to 422.Note: Numbers n parenthesesare standard rrors.a "NA" ndicatesnot applicable(Drenthe ies outsidethe Randstad).*


21/25

MARKET PARTITIONING 427We did additional analyses to checkwhethergeneralist concentrationindeed me-diates the relationshipbetween resourcecon-centration and specialist performance out-

side the Randstad. First, we find that re-source concentration (measured with theoverall resource-concentration ndex) is sig-nificantly relatedto concentrationamong thenationals, both inside and outside theRandstad. Next, we estimated the effect ofthis resource-concentration index on themarket share and circulation growth of theregionals, with and without the concentra-tion level among the nationals included inthe regression (and using the same controlvariables as in Tables 4 and 5; results avail-able on request). As expected, outside theRandstad we find evidence that resourceconcentration has a positive impact on mar-ket share and circulation growth of theregionals via the concentrationlevel amongthe nationals. Inside the Randstad,however,we find the opposite effect. That is, resourceconcentration has a strongnegative effect onregional market share, partly because it in-creases concentration among the nationals.The findings are not significant for regionalcirculation growth inside the Randstad.Taken together, the supplementary resultsprovide evidence, at least in part, for a me-diation model outside the Randstad. Insidethe Randstad,the findings supportour con-clusion that these markets tend to behave aspredictedby industrialorganizationeconom-dummyvariable or "after1982,"which has theexpected sign). Perhaps his is because in theRandstadheregionalnewspapersannotbecon-sideredto be a different"organizationalorm"comparedwith the nationals,making t difficultto modelperformanceeparately. or nstance,fan increasing arrying apacity s beneficial orthe nationalsn theRandstad,t indirectlyhas anegativeimpacton the shareof the regionals.Thismightexplainwhytheeffect of populationswitches rompositive (outside heRandstad)onegative (inside the Randstad).The results inTable5, in whichregionals'circulation rowthis modeled,reveala similarpattern.Somevari-ables, however, ose significance e.g., popula-tion andunemployment utsidethe Randstad).This is probably ue to the inclusionof themar-ket shareof the regionalsat t - 1, which has alargenegativeimpacton growth.As expected,thereis an upperboundon regionalnewspapergrowthathighmarket hare.

ics: In homogeneous markets (with high re-source concentration)with scale economies,a few generalists will eventually dominate,pushing specialists out of the market.DISCUSSIONAlthoughmany studies of resourcepartition-ing have appeared, these have usually notaddresseda core partof the theory-the partdealing directly with competition amonggeneralists. Under certainconditions, an ex-pected outcome of this competition is highmarket concentration,an importantphenom-enon of wide general interestto sociologists.So neglect of this part of the theory has lim-ited its potential value.Our study was designed to investigateconcentrationin a market that is also under-going segmentation. We developed a modelof how the distributionof environmentalre-sources drives concentrationamonggeneral-ists. Specifically, we hypothesized that thecompetitive struggle among generalists forthe center of the market will be most pro-nounced in markets in which the environ-mental resources are concentrated. As isusual in studies of resource partitioning,wealso hypothesized that specialist organiza-tional forms will flourish when marketcon-centration among generalists increases andthe level of untapped environmental re-sources is high enough. We examined thesetheoretical claims by analyzing the Dutchdaily newspaperindustryfrom 1968 to 1994at the province level. For each of the Neth-erlands' 11 provinces, we constructeddistri-butionalmeasuresof the potentialreadershipbase along four dimensions that have beenimportantfor newspapers: age demography,religious background,political preferences,and educational level. The statistical analy-ses involved estimating equations that pre-dict the behavior of Dutch daily papers byorganizationalform (generalist or specialist)relative to variationsin the four majorenvi-ronmentalresource distributions, as well asto the populationprocesses set in motion bythem.The findings support the theoreticalclaims, at least for the dimensions of age,politics, and education. That is, greater ho-mogeneity of persons within a province interms of age demography,political orienta-


22/25

428 AMERICAN SOCIOLOGICAL REVIEWtion, and educational level amplifies the ex-tent of market concentrationamong the gen-eralist (national) newspapers.Likewise, generalist market concentrationleads to higher specialist viability, but onlywhen specialists have ample opportunitytodifferentiate themselves from generalists(i.e., when specialists and generalists arenotdirect competitors). Our explanation is thatwithout sufficient resource-space heteroge-neity on the key dimension of identity,scale-driven competition produces a limited num-ber of large generalist papers at the expenseof the performance of specialist dailies. Thefindings underscore a fundamentalclaim inecological theory: that "the number of dis-tinct resources sets an upperboundon diver-sity in the system" (Hannan and Freeman1977:944).The theory suggests a unidirectional pro-cess whereby specialists come to occupythe uninhabitedresource space left over asa result of generalist competition. That is,the specialist followers occupy the residualspace that the generalist leaders are not ableor willing to fill. Less thought has beengiven to mechanisms flowing the otherway.Although our recursive modeling approachis consistent with the spirit of the theory,future work might do well to explorewhether some specialist outcomes influencegeneralist behavior and concentration lev-els.22 Additionally, future research may fo-cus on yet another feedback loop-thatfrom the organization to the environment.For example, generalists may, in theirsearch for scale, homogenize consumertastes, thus transforming the resource spaceitself (Boone and van Witteloostuijn 1995).In this context, advertising campaigns andlobbying activities are examples of general-ist behavior that attempt to do preciselythis. And vice versa, specialists may influ-ence the taste space to their advantage, mo-bilizing latent demand for niche products,

22 Wetriedsimultaneous-equationodeling otest for feedbackeffects fromspecialistperfor-mance o generalist oncentration. he estimatesof themodelswe specifiedwerebroadly onsis-tent with our findings here. Unfortunately,wewerenotableto findtheoretically efensibleex-ogenous instrument ariables hatworkedwell,whichcreatedproblems f identification.

which "heterogenizes" the resource envi-ronment. Unraveling these possible recipro-cal causalities presents an interesting chal-lenge for future research.What are the general implications of theideas examined here and associated find-ings? We consider this question on two lev-els: first, concerning the theory,and second,concerning the general research approach.For resource-partitioning theory, it is in-structive to note that some broadly similarmodels are actively used in other social sci-ence contexts. In economics, followingHotelling (1929), a literaturehas developedon the profit-maximizing location of firms

in product space and the associated equilib-rium marketstructure.Here, both the firms'profit-maximizinglocation and the market'sequilibrium structureare determinedby thedistribution of demand, which in general isassumed to be distributed along a line or acircle (Thisse and Norman 1994). InHotelling's benchmarkcase, with a flat de-mand distribution firms locate close to themarket center. This result is known as theprinciple of minimum differentiation.Sutton's (1991) dual market structuretheory provides an interesting comparisonto resource partitioning. In this theory, themarket center is populatedby large product-differentiation generalists, leaving the mar-ket peripheryfor small low-price specialistsas a result of a peaked distribution of de-mand (i.e., a unimodal distribution imply-ing concentratedresources).In political science, research on politicalparties also contains some broadly similarideas. Following the seminal contributionsof Downs (1957a, 1957b), a rapidly ex-panding literature is devoted to explainingpolitical party location issues on the basisof the characteristics of the voter space(Enelow and Hinich 1984). The argumenthere is that where parties eventually locatedepends on the shape of the distribution ofvoter preferences. In a two-party systemwith a unimodal-low-variance voter distri-bution, Hotelling's (1929) principle ofminimumdifferentiation holds true. That is,parties locate in the center. However, in amultiparty system with a polymodal-high-variance voter distribution, each party tendsto differentiate away from its rivals, whichis the principle of maximum differentiation.


23/25

MARKET PARTITIONING 429What distinguishes the Hotellian economicsof product differentiation from theDownsian literatureabout party differentia-tion is that the former assumes firm-levelflexibility while the latter argues thatpoliti-cal parties are ideologically immobile. Thelatter's assumption resembles organiza-tional ecology's central notion of relativeinertia (Carroll and Hannan2000). Interest-ingly, although these theories originate indifferent disciplines, they all share a com-mon theme-the shape of the resource dis-tribution determines the rules of the com-petitive game and ultimately the structureof populations of social systems-and acommon prediction-the concentration ofresources triggers the escalation of compe-tition for the center.The different outcomes predictedby thesemodels and theories matterbecause the so-cial consequences of organizational varietymay be far-reaching. In many organiza-tional populations, a diversity of organiza-tional forms enhances societal welfare byincreasing the range of options open to citi-zens (in their roles as consumers, voters,volunteers, etc.). This is also true in ourcontext of daily newspapers, as newspaperdiversity has long been associated in critics'minds with a politically healthy democracy(Bogart 1981). Generally speaking, entre-preneurialbehavior by specialists may be amajor vehicle for product innovation andtechnological progress (Schumpeter1943).Concerningthe general researchapproachused, we think it is instructive that our find-ings suggest that much of what happens atthe level of competitive processes is trig-gered initially by the underlyingfeatures ofthe resource space. Different resourcespaces produce very different market struc-tures. Although organizationsare invariablyviewed as open systems, researchersdo notcommonly analyze the relationshipbetweenthe features of the resource environmentand their consequences on organizationalbehavior. Moreover, starting conceptuallyfrom the underlyingresource-space featuresmay be a way to start reconciling findingsin many different literatures, from the soci-ology of social movements andthe econom-ics of product differentiationto the politicsof party positioning and the partitioningoforganizationalpopulations.

The study also suggests that organiza-tional diversity is at least in part an endog-enous process driven by concentrationamong the large generalist players in themarket. This goes against the widely ac-cepted intuitionthat marketconcentrationissome kind of an end state characterizedby aconvergence in products, services, organiza-tional structures, and strategies, ultimatelyreducing competition and consumer choice(e.g., the anti-globalization movement).However, the theory and findings examinedhere suggest that this view sometimes needsqualification. Although escalating competi-tion among dominant companies may in-crease standardizationand reduce organiza-tional variety in the short term, it may alsosow the seeds for future diversity. In otherwords, consolidation may set in motion aprocess of specialization at the populationlevel of analysis, making the coexistence ofdifferentforms (i.e., generalists and special-ists) a sustainableand even likely outcome.Christophe Boone is Professor of OrganizationTheory and Behavior at the Faculty of AppliedEconomics of the University of Antwerpenin Bel-gium. His current research spans different levelsof analysis with afocus on individual differencesand top management team composition and mo-bility, organizational performance and industrydynamics. He has published in journals such asJournal of Management Studies, OrganizationStudies, Journal of Economic Psychology andStrategic ManagementJournal.Glenn R. Carroll is the Laurence Lane Profes-sor of Organizations in the Graduate School ofBusiness and (by courtesy) Professor of Sociol-ogy at Stanford University.His most r

Boon Witteloostuijn Caroll (2002)

Documents