Technological Collaboration_ Bridging the Innovation Gap Between Small and Large Firms_ - Nieto -...

8/10/2019 Technological Collaboration_ Bridging the Innovation Gap Between Small and Large Firms_ - Nieto - 2009 - Journal

1/26

Technological Collaboration:Bridging the Innovation Gap

between Small and Large Firms*jsbm_286 44..69by Mara Jess Nieto and Llus Santamara

This paper analyzes how technological collaboration acts as an input to theinnovation process and allows small and medium-sized enterprises to bridge the

innovation gap with their bigger counterparts. Based on a large longitudinal sampleof Spanish manufacturing firms, the results show that though technological collabo-ration is a useful mechanism for firms of all sizes to improve innovativeness, it is acritical factor for the smallest firms. The impact of this collaboration varies dependingon innovation output and type of partner. Specifically, the impact of collaboration in

small and medium-sized firms is more significant for product than process innova-tions. Regarding type of partner, vertical collaborationwith suppliers andclientshas the greatest impact on firm innovativeness, though this effect is clearer

for medium-sized enterprises than for the smallest firms.

Introduction

Most current economies are largelycomposed of small and medium-

sized enterprises (SMEs). In the Euro-pean Union, for instance, SMEsmake up 99 percent of industry andaccount for more than 70 percent of

*The authors thank the editor, Professor Massimo Colombo, and the two anonymous reviewers

for their helpful comments and suggestions. This study has been financially supported byprojects SECJ2007-67582-C02-02/ECON and CCG08-UC3M/HUM-4152. The usual disclaimerapplies.

Mara Jess Nieto is an Associate Professor at the Management and Strategic Division,University Carlos III of Madrid, Spain. Her current research interests include innovationmanagement, strategic alliances, and internationalization strategy, specially focused on SMEs.She has also made recent contributions on information technology.

Llus Santamara is an Assistant Professor at the Department of Business Administration,University Carlos III of Madrid, Spain. His current research interests include innovationmanagement, technological cooperation, and comparative institutional analysis. He also studies

the management of intangibles and accounting information systems.Address correspondence to: Mara Jess Nieto, Universidad Carlos III de Madrid, C/Madrid,

126. 28903-Getafe, Spain. Tel: +34-91-6245826; Fax: +34-91-6245707. E-mail: [email protected].

Journal of Small Business Management 2010 48(1), pp. 4469

JOURNAL OF SMALL BUSINESS MANAGEMENT44


2/26

employment.1 Their innovative capabil-ity is a crucial driver of sustainablecompetitive advantage in todays rapidlychanging markets, where the continu-ous development of new products andprocesses is the key to survival,growth, and profitability (Wolff and Pett

2006; Verhees and Meulenberg 2004;Forrest 1990). This situation has fuelledgrowing concern among managersand policymakers and has led to astrong commitment to use policyinitiatives to support innovation withinSMEs (Jones and Tilley 2003; Bougrainand Haudeville 2002; Hoffman et al.1998).

This interest is also apparent in aca-

demic circles. In the economics of inno-vation and technological changeliterature, the relationship between firmsize and innovation activity has receiveda good deal of attention (for an overviewsee Cohen 1995). The Schumpeteriandebate over which firmslarge orsmallare more able and more likely toinnovate is one of the oldest in politicaleconomics, and it continues to arousecontroversy today (Tsai and Wang 2005).The contradictory nature of both the con-ceptual and empirical findings, however,does not provide clear guidance on whatto expect in general (Stock, Greis, andFischer 2002, p. 541). Although it has notbeen possible to establish a strong rela-tionship between firm size and innova-tion per se, some empirical research

suggests that small and large firms havedifferent determinants of innovationefforts (Rogers 2004; Van Dijk et al.1997) and do not pursue the same typesof innovation (Fritsch and Meschede2001; Cohen and Klepper 1996; Noote-boom 1994).

These differences in innovationefforts and results can be explained by

the advantages traditionally ascribed tolarge and small firms. The main relativestrengths of SMEs lie in behavioraladvantages, whereas those of largefirms reside in their resource advan-tages (Rothwell and Dodgson 1994).Smaller firms generally enjoy internal

conditions that encourage innovative-ness, such as entrepreneurship, flexibil-ity, and rapid response (Lewin andMassini 2003; Schumpeter 1942). Onthe other hand, the relative weaknessesof small firms compared to large oneslie in the constraints they face ongaining access to critical resources andcapabilities for innovation (Hewitt-Dundas 2006). The advantages of scale

and scope provided by the size of largefirms make them better equipped forinnovations that require large and spe-cialized teams or sophisticated equip-ment (Cohen and Klepper 1992). SMEsare also usually at a disadvantagewhen it comes to intangible resources,as they have access to a smallerrange of knowledge and humancapital skills than large firms (Rogers2004).

SMEs, however, have alternatives tointernal development that may enablethem to bridge the resource gap thatexists with large firms (Miles, Preece,and Baetz 1999). In this respect, the lit-erature on innovation stresses the roleof cooperative R&D in overcoming thelack of internal resources and in

improving innovativeness and competi-tiveness, particularly for SMEs (Hewitt-Dundas 2006; Rogers 2004; Bougrainand Haudeville 2002; Nooteboom 1994;among others). Indeed, SMEs engagedin technological innovation have usedcooperative R&D for informationexchange, resource acquisition, technol-ogy transfer, and risk management.

1Based on data from the Informe 2003/7 del Observatorio de la PYME Europea prepared byDireccin General de la Pyme (the Spanish governments Board of SMEs). Firms with fewerthan 200 employees are classified as SMEs.

NIETO AND SANTAMARA 45


3/26

The collaborations, though, are nothomogeneous and their impact is notuniformly positive. As Freel (2003, p.766) points out, the impact of techno-logical collaborations varies amongsectors and the type of innovationpursued.

Consequently, we need to delvemore deeply into the specific role thatinnovation networking plays as a pos-sible determining factor in developingthe innovation capacity of SMEs(Edwards, Delbridge, and Munday2005). This study, then, sets out to dis-cover if collaboration, specifically tech-nological collaboration, enables SMEs toovercome their lack of resources and

capabilities, and how this in turn booststheir innovativeness. The paper ana-lyzes the potential effect of collabora-tion on innovation outcomes, in termsof process and product, and goes on toreveal if innovation networking canclose the gap between SMEs and largefirms. The study also measures the dif-ferent impact of collaborating withclients and/or suppliers (vertical col-laboration) versus collaborating withresearch organizations such as universi-ties or technology institutes. We believethis paper makes a novel contributionto the literature on innovation as notenough empirical research has beendone on the impact of cooperative R&Don performance, especially for SMEs(Okamuro 2007).

The paper is organized as follows.The next section develops our hypoth-eses based on the existing literature ontechnological collaboration, innovation,and firm size. The section on methodol-ogy describes the data, measures andmodel specification. The results are thenanalyzed, and the final section contains adiscussion of these findings and our con-clusions.

Theoretical FrameworkImproving Innovation Capabilitiesthrough TechnologicalCollaboration

In technological activities, alliances,and networks are the main sources of

innovation (Von Hippel 1988). An inter-firm alliance can be defined as a closeand deliberate collaborative relationshipbetween independent firms and/or insti-tutions to perform business activities. Inthis context, the terms collaborativearrangements, cooperative agree-ments, strategic alliances, or coali-tions are found throughout theliterature and are often used interchange-

ably (Forrest 1990). These concepts donot refer to agreements involving fullownership forms, but to different con-figurations based on long or short-termarrangements. In the specific case oftechnological collaboration, these alli-ances include collaborative R&D agree-ments, university and/or researchinstitute agreements and technologylicensing.

According to Duysters, Kok, andVaandrager (1999), alliances are nolonger regarded as peripheral, but as acornerstone of the firms technologicalstrategy. Indeed, over the last twodecades the use of external networksby firms of all sizes has increased tre-mendously (Hagedoorn 2002, 1996). Agrowing number of studies suggest that

external networks specifically createunique benefits and challenges forsmaller firms (Zahra, Ireland, and Hitt2000; Powell, Koput, and Smith-Doerr1996). In fact, alliance formation isarguably one of the most utilized strat-egies for resource acquisition and lever-aging by small to medium-sizedenterprises (Baum, Calabrese, and Sil-verman 2000; Miles, Preece, and Baetz

1999). Most recently, Hewitt-Dundas(2006) finds that although a lack ofpartners for innovation has a negativeimpact on the ability of small firms to



4/26

undertake innovation, it does not havea significant effect on the probability ofinnovating in larger firms. Her resource-based interpretation of this finding isthat the external resources and capabili-ties that small firms can access throughexternal innovation partnerships may

provide them with the stimulus andcapability to innovate that they wouldnot otherwise have.

Rogers (2004) points out that SMEsmay rely more heavily on externalknowledge networks as an input toinnovation than do large firms. Net-works allow SMEs to receive anddecode flows of information. They rein-force SMEs competitiveness by

enabling them to access new knowl-edge, sources of technical assistance,expertise, sophisticated technology,and market requirements; they alsostrategically reduce the irreversi-bility costs of the innovation process(Freel 2005; Bougrain and Haudeville2002).

Given that small firms seem to havepotentially more to gain from innova-tive partnerships than larger firms, thevery success of SMEs vis--vis theirlarger competitors may be due to theirability to use external networks moreefficiently (Nooteboom 1994; Rothwelland Dodgson 1994). SMEs, then, haveto use alliances astutely to overcomebarriers to growth imposed by absolutelimits to resources (Van Dijk et al. 1997;

Ahern 1993). In this sense, strategic alli-ance formation is regarded as one ofthe most critical strategic actions thatSMEs must undertake for survival andsuccess (Dickson, Weaver, and Hoy2006).

Based on the preceding discussion,we conclude that technological collabo-ration may be a good way of strengthen-ing and complementing the resource

endowments and capabilities of SMEsand of improving their innovativeness.Our first two hypotheses make thisexplicit:

H1: Technological collaboration im-proves innovation performance inSMEs.

H2: Technological collaboration has agreater impact on innovation perfor-mance in SMEs than in large firms.

Differences Depending onInnovation Outcomes

Small and large firms do tend toemphasize different innovation activities(Cohen and Klepper 1996). Indeed, smalland large firms are probably good atdifferent types of innovation in accor-dance with their relative strengths andweaknesses (Nooteboom 1994). Large

firms are likely to be better at innova-tions that make use of economies ofscale and scope, or require large teamsof specialists, such as fundamental,science-based innovations, and large-scale applications (Cohen and Klepper1992). Small firms are relatively strong ininnovations where effects of scale are notimportant and where they can make useof their flexibility and proximity tomarket demand, such as new products orimprovements in existing ones for nichemarkets, and small-scale applications(Vossen 1998; Nooteboom 1994). Cohenand Klepper (1996) show that large firmshave a greater incentive to pursue alltypes of innovation because they canapply their innovations to a larger levelof output. The same authors state that

this incentive is relatively greater forinnovations that depend on existingoutput for their exploitation, as is thecase with process innovations.

Product innovations, for their part, arebetter instruments for entering marketsthan process innovations. Their charac-teristics enable them to answer clientneeds more quickly and capture newmarkets before competitors. In contrast,

process innovations possess advantagesthat usually lead to productivity gainsand cost reductions that indirectly affectmarket position. In line with this, Wolff



5/26

and Pett (2006) find that productimprovement orientation is positivelyassociated with growth and profitabilityin SMEs, whereas no relationship isfound with process improvement orien-tation.

All this may explain why SMEs tend to

concentrate their efforts more on productthan process innovations. The impact ofsize on innovation is always greater forprocess innovations because large firmshave more facilities (internal resourcesand capabilities) and incentives for thistype of innovation than SMEs do(Martnez-Ros 2000). Fritsch andMeschede (2001) make this point by ana-lyzing the resources devoted to product

and process R&D. They find that smallenterprises on average spend a muchhigher proportion of their R&D budgeton new products than on new processes.Thus, if small firms focus less on processinnovations, they will also produce alower absorptive capacity for this typeof innovations.

Bearing in mind that the impact ofcollaboration is not homogeneous(Freel 2003) and that SMEs and theirlarge counterparts follow different inno-vation processes, the impact of techno-logical collaboration should varyaccording to firm size and the innova-tion outcome pursued. As already seen,previous research reveals that SMEsinvest less effort in and develop lowerabsorptive capacity for process inno-

vations. In addition, as Ornaghi (2006)affirms, product improvements have alarger technological diffusion and maybe simpler to learn than process inno-vations, which are often linked to theskills of managers or technicians. We,therefore, postulate the followinghypothesis:

H3: The impact of technological collabo-ration varies depending on the inno-vation pursued. Specifically, for SMEsthis impact is greater for product than

for process innovations.

Differences Depending on

Type of PartnerIf collaboration agreements are good

instruments for developing and sustain-ing technological capability, it followsthat the choice of technological partneris crucial (Nieto and Santamaria 2007). Infact, significant differences exist amongtypes of partners and these can deter-mine how the collaboration is managedand what kind of innovation can be

achieved (Whitley 2002). In particular,there are significant contrasts in howfirms cooperate with agencies in thepublic science system and with suppliersand clients (vertical collaboration).2 Ana-lyzing the relationship between typeof partner and expected benefits forSME innovativeness, then, would beinteresting.

Vertical collaboration with clientsand/or suppliers allows a firm to gainconsiderable knowledge about new tech-nologies, markets, and users needs(Whitley 2002). Therefore, this collabora-tion has a significant impact on bothproduct and process innovation and isparticularly important for firms that tendto focus on a smaller set of businesses(Miotti and Sachwald 2003). As this kind

of focus is typical in small firms, SMEscommonly use their suppliers and clientsas a valuable source of technologicalinformation. Many innovations by smallfirms, then, are based on off-the-shelftechnologies, concepts, and/or resourcesoffered by supplying industries (Verheesand Meulenberg 2004). Consequently,

2

Most of the theoretical literature also analyzes cooperative agreements between competitors.As occurs in other studies (Cassiman and Veugelers 2002), however, the number of firms in oursample that collaborate with competitors is very low. For this reason we do not include thistype of agreement.



6/26

suppliers may play a more active role instimulating innovation by trying toinfluence the small firms innovationdecision. Meanwhile, the relation-ship between innovativeness andmarket intelligence makes client infor-mation a key resource for innovation in

small firms (Verhees and Meulenberg2004).

Research organizations (that is univer-sities and technology institutes) have nottraditionally focused on filling out theinnovation processes of firms, but onproviding them with new scientific andtechnological knowledge (Drejer andJrgensen 2006). This has changed in thelast few years, however, and these orga-

nizations have been under considerablepressure to move closer to industry. Twomain reasons exist for this change. First,governments have encouraged them toundertake more research directed atboosting the competitiveness of industry(Tether 2002). Second, pressure onfunding has pushed universities intogreater collaboration with industry(Gibbons et al. 1994). In fact, severalstudies document the important role thatuniversities and other research institu-tions have on technological innovation(Vuola and Hameri 2006; Bozeman2000).

Although large firms are more likelyto collaborate with universities andresearch centers (Bayona, Garcia-Marco,and Huerta 2002), several empirical

studies indicate that SME innovationsreceive significant spillovers from univer-sity research (Piergiovanni, Santarelli,and Vivarelli 1997; Acs, Audretsch, andFeldman 1994). Public researchinstitutesapart from conducting theirown studiescan provide small localfirms with solutions to technical prob-lems in product development. Indeed,the previous studies show that public

research institutes play a key role inlocalized knowledge spillovers receivedby small firms (Izushi 2003; Beise andStahl 1999).

And yet formidable barriers mayprevent SMEs from exploiting researchorganizations. These barriers existbecause SMEs are often unaware of theirreal requirements or have problemsexpressing them (Lambrecht and Pirnay2005). Firms may not know what types

of services they need to develop innova-tions (Izushi 2003) or which technologi-cal partners have the most relevantcapabilities (Geisler 1997). For thesereasons fluid and clear communicationbetween technological partners andfirms is important, particularly when thefirms are SMEs (Smallbone, North, andLeigh 1993). Another barrier, however,may exist because research organizations

are not overstocked with experts withthe specific technoeconomic capabili-ties to support SME innovation processesand resolve the communication prob-lems that can arise (Rolfo and Calabrese2003).

These arguments seem to suggest thatdifferent types of collaboration (verticalpartners versus research organizations)will have different impacts on the inno-vativeness of SMEs. Collaboration withresearch organizations (despite theirincreasing focus on industrial needs)may be problematical for SMEs, whereaspartnerships with clients and suppliersbecause of their characteristicsseemlikely to have the greatest impact on SMEinnovation processes. Our last hypoth-esis, then, posits:

H4: The impact of technological collabo-ration varies according to the partnerchosen. Specifically, for SMEs theimpact of vertical partnerships is

greater than the impact of collabora-tion with research organizations.

MethodologySample and Data

The source for our empirical analysisis theSpanish Business Strategies Survey(SBSS). This is an annual firm-level panelof data compiled by the Spanish Ministry



7/26

of Industry and the Public EnterpriseFoundation; it has been used by manyother researchers to study innovation(Cuervo-Cazurra and Un 2007; Beneito2006; Huergo 2006, among others). TheSBSS contains an interesting and wideset of variables on Spanish firms operat-

ing in all manufacturing industries of theclassification NACE-Rev.1.

The sample is representative of thepopulation of Spanish manufacturingfirms. Firms with between 10 and 200employees are selected through arandom stratified sample (according tofirm size and industry classification),and firms with more than 200 employ-ees are surveyed on a census base

(Huergo 2006). The empirical analysis isbased on the balanced sample of firmswith information available for the com-plete period from 1998 to 2002. Ourfinal sample contains 6,500 observationsfrom 1,300 firms that have remained inthe survey during the whole five-yearperiod.

MeasuresDependent Variables: Innovation Out-puts. The dependent variables are rela-tive to firm innovation performance in aspecific period t. In order to capture thedifferent innovation outputs, three sepa-rate measures were used: innovation,product innovation, and process innova-tion. Innovation is a dichotomous vari-able that takes value 1 when the firm

declares at least one product innovationand/or at least one process innovation inthe survey year; otherwise its value is 0.

Product Innovation is a dichotomousvariable that takes value 1 when the firmdeclares it has introduced completelynew products or products with importantmodifications, products with new func-

tions resulting from innovation, or hasmade changes to the design, presenta-tion, materials, or composition of theproduct. Otherwise its value is 0. Lastly,

Process Innovation is assumed to havehappened when the firm indicates it hasintroduced some significant modification

in the production process. This modifi-cation may involve the introduction ofnew machines or new methods of orga-nization, or the introduction of both. It isalso a dichotomous variable.

Independent Variables. Given theobjectives of the research, the studyattempts to measure the potential inno-vative behavior of SMEs, especially how

this behavior is affected by technologicalcollaboration. To take account of techno-logical collaboration, a dichotomousvariable (Collaboration) that indicateswhether firms have collaborated techno-logically with other firms or researchorganizations was constructed. This tech-nological collaboration includes differentmodes of technology partnering thatHagedoorn (1993) defines as contractualarrangements: joint R&D agreements,technology exchange agreements, andother technology flows such as licensing.Collaboration takes value 1 when thefirm is involved in at least one of thesecontractual arrangements; the variable islagged one period in the models.3

In order to gauge the impact of col-laboration depending on firm size, firms

were sorted using three dichotomousvariables: (1) Small takes value 1 whenthe firm has fewer than 50 employees;(2) Medium takes value 1 when the firmhas between 50 and 200 employees; and(3)Largetakes value 1 when the firm hasmore than 200 employees. This makes itpossible to measure the effect of collabo-

3

The models in this study include independent variables lagged one period. Other estimationswith variables lagged two periods were also performed and highly robust results wereobtained. In order to retain the largest number of years and observations possible, theone-period option was selected.



8/26

ration in each size category with the fol-lowing crossed effects (also included aslagged variables):

Collaboration Small CollaborationMedium Collaboration

L

=

+

+ argge Collaboration

To test the final hypothesis, firms thathave collaborated technologically withclients and/or suppliers (Vertical) andothers that have done the same withuniversities and/or technology institutes(Research Organizations) were identi-fied. These dichotomous variables werealso included in the models as lagged

variables and were subdivided in eachsize category. Thus:

Vertical Small Vertical MediumVertical L e Vertical

= +

+ arg

Rearg

sarch Organizations Small ROMedium RO L e RO

=

+ +

To enable us to compare the impact oftechnological collaboration dependingon the size of the firm, each modelincludedtogether with the interactiontermsthe three previously defined cat-egories of firm size: Small, Medium,and

Large.

Control Variables. To accuratelymeasure the effect that technological col-

laboration has on innovativeness, otherpotential inputs to firms innovation pro-cesses need to be considered: in particu-lar, formal R&D activities (internaland/or external), design and the use oftechnological consultants. To avoid prob-lems of simultaneity with the innovationgeneration process, these innovationactivities are lagged one period. R&Dinvestment is now generally seen as a

crucial determinant of innovation,because it helps firms create, absorb,exploit, and transform new knowledgeinto new products and/or processes

(Becheikh, Landry, and Amara 2006).Formal R&D captures the decision toperform R&D activities (in-house and/or contracted) and is measured as theratio of R&D expenditures to totalsales. Design covers a wide rangeof activitiesincluding architectural,

fashion, interior, graphic, industrial, andengineeringthat can be implementedin a variety of contexts. Design allowsthe firm to incorporate what customerswant, what can be more efficiently pro-duced, and what fits better with thefirms other products, strategy, andimage (Walsh 1996). The dichotomousvariable Design measures the perfor-mance and/or contracting of these activi-

ties. Lastly, technological consultantsmay represent another source of innova-tion. Consultants often interact withnumerous firms across a variety of indus-tries and may transfer tacit knowledgedeveloped through this ongoing experi-ence (Bierly and Daly 2007). Thedichotomous variable Consultant identi-fies if a firm uses a technologicalconsultant.

Additional controls for firm-specificcharacteristics and sector of activitywere included. The age of the firm(Age)calculated as the number of yearssince a firms foundationmeasures firmexperience and learning, and is also acommonly used variable in empiricalstudies of innovation (Kumar and Saqib1996). Numerous studies recognize the

effect of ownership structure on innova-tion and track its influence by focusingon foreign ownership, although theempirical evidence is not conclusive (seeBecheikh, Landry, and Amara 2006). Tocontrol for its potential impact, the per-centage of foreign equity in a firmscapital was included (Foreign). Exportand internationalization have a positiveand significant effect on innovation

(Galende and De la Fuente 2003). Toremain competitive in foreign markets,firms need to innovate constantly(Veugelers and Cassiman 1999). Export



9/26

intensity (Export), then, was alsoincludedcalculated as the ratio of thefirms sales in foreign markets to totalsales.

Eighteen industry dummies were usedto capture the effect of sector character-istics related to life cycles and techno-

logical regimes on innovation. The firmsactivity classification is an aggregation ofthe two-digit manufacturing industriesclassification in the NACE-Rev.1 (forsimilar classifications see Huergo 2006).

Table 1 contains the descriptive statis-tics and correlations of the independentand control variables used in this study(with the exception of the sectordummies).

Model SpecificationTo test the first two hypotheses, two

sets of models were specified with thedependent variable Innovation (with nodistinction between product and processinnovations). The first model (Model A1)analyzes the impact of Collaborationoninnovativeness, whereas the second(Model A2) attempts to capture thesubtleties of technological collaborationfor each size category. Both modelsinclude the rest of the innovation activi-ties, the firm-specific controls, and thesector dummies.

Positive and significant coeffi-cients of the interaction terms SmallCollaboration and Medium Collabo-ration are required for H1 to be sup-

ported. A greater and significant impact ofcollaboration for SMEs in comparison tolarge firms must be observed for H2 to besupported. Two Wald tests were per-formed to test the significance of thedifference of coefficient estimates.

Given the binary character of thedependent variable, probit models werespecified. To address concerns of unob-served heterogeneity, a random-effects

panel probit model was employed. Ourdecision to use a random-effects modelinstead of a fixed-effects model wasbased on the following: (1) our sample

was drawn from a large population; inthis setting, it might be more appropriateto view individual specific constant termsas randomly distributed across cross-sectional units (Greene 2000, p. 567); (2)estimates computed using fixed-effectsmodel can be biased for panels over

short periods. This is not a problem withrandom-effects models (Hsiao 1986;Heckman 1981). Given that all the firm-year observations in our sample werepresent for only five years, random-effects was the preferred approach; and(3) fixed-effects models cannot includetime-independent covariates. Our analy-sis could be limited without thesevariables.

In order to test H3 and H4, we differ-entiate between product and processinnovations. As these two types of inno-vations may be related to each other(Fritsch and Meschede 2001; Martnez-Ros 2000), the error terms of the twomodels are likely to be correlated. Thus,an extension of probitknown as bivari-ate probit (Greene 2000) is usuallya more appropriate estimator. Thebivariate probitmodel has the followingspecification:

Z x y z yz

i i i i i i

i

1 1 1 1 1 1 1

1

1 0 00

= + = > =

; ,ifif

Z x y z yz

i i i i i i

i

2 2 2 2 2 2 2

2

1 0 00

= + = > =

; ,ifif

i i N1 2 0 0 1 1, , , , ,( ) ( )

This model produces estimates of thecoefficient vectors b1 and b2 for the twoequations, ofr(the correlation betweenthe error terms eijof the equations), andof the standard errors for these param-eters. Our tests revealed that the corre-lation between the equations wasstatistically significant, which is an

indicator that the bivariate modelwas more effective than the separateprobit models (Greene 2000, pp. 853854).



10/26

Table1

Descr

iptiveStatistics,C

orrelations,andC

ollinearityDiagnosticsoftheIndep

endentand

Contr

olVariables

Mean

S.D

.

1

2

3

4

5

6

7

8

9

10

VIFa

VIFb

1

Collaboration

0.3

37

0.4

72

1.6

5

2

Vertical

0.2

58

0.4

38

0.8

3

1.6

2

3

RO

0.2

24

0.4

17

0.7

5

0.4

9

1.5

3

4

Sma

ll

0.5

15

0.4

99

-0.4

6

-0.4

2

-0.40

2.6

5

2.7

5

5

Med

ium

0.1

99

0.3

99

0.0

3

0.0

1

-0.01

-0.5

1

1.5

5

1.6

0

6

FormalR&D

0.0

07

0.0

23

0.3

4

0.3

3

0.29

-0.1

8

0.0

2

1.1

6

1.1

8

7

Des

ign

0.2

89

0.4

53

0.3

2

0.3

3

0.23

-0.2

7

0.0

4

0.1

9

1.1

7

1.1

8

8

Con

sultant

0.2

26

0.4

18

0.4

2

0.3

9

0.37

-0.2

8

-0.0

1

0.1

8

0.2

1

1.2

1

1.2

1

9

Age

25.2

0

20.9

3

0.2

1

0.1

8

0.18

-0.3

4

0.0

6

0.1

0

0.1

6

0.1

0

1.1

7

1.1

7

10

Exp

ort

0.1

98

0.2

64

0.3

6

0.3

3

0.30

-0.4

5

0.1

2

0.1

7

0.2

0

0.1

8

0.1

1

1.3

8

1.3

8

11

Foreign

18.7

2

37.7

1

0.2

8

0.2

7

0.21

-0.4

3

0.0

7

0.0

5

0.1

0

0.1

1

0.1

9

0.3

3

1.3

3

1.3

3

Mean

VIF

1.4

7

1.5

0

S.D.,standarddeviation;VIF,varianceinflationfactor;RO,researchorganization.

aModels

A1,

A2,andB.

bModelC

.



11/26

Two bivariate probit models were esti-mated to test H3 (Model B) and H4(Model C). Both models include the inde-pendent variables, the firm-specific con-trols, and sector and year dummies. TheWald tests comparing the difference ofcoefficient estimatesproduct innova-

tion versus process innovationmust besignificant in the SME categories tosupport H3. The set of Wald tests for H4compare the difference of coeffi-cients estimatesvertical versus RO(research organization) collaborationineach size category and type of innova-tion. For H4 to be supported, a greaterand significant effect of vertical partner-ships must be observed for SMEs.

Our models were also analyzed forpotential endogeneity and multicollinear-ity problems. We test for endogeneity byusing an instrumental variable approachfor probit regressions (Smith and Blun-dell 1986). The test specifies that theexogeneity of one or more of theexplanatory variables is under suspicion.Our interest was to check that no endo-geneity problems between collaborationvariables and innovation results existed.Under the null hypothesis, the model isappropriately specified with all explana-tory variables as exogenous. Under thealternative hypothesis, the suspectedendogenous variables are expressed aslinear projections of a set of instru-ments,4 and the residuals from thosefirst-stage regressions are added to the

model. Under the null hypothesis, theseresiduals should have no explanatorypower. Estimation of the model gives riseto a test for the joint hypothesis that each

of the coefficients on the residual seriesis zero. The Smith-Blundell test statisticis evaluated with respect to a c2 distribu-tion in the number of potentially endog-enous variables, and the associated

p-value either rejects or not the nullhypothesis. The test was repeated for the

different combinations of collaborationvariables and innovation outputs for allof the models. The null hypothesis ofexogeneity was not rejected in any case,5

thus indicating that the collaborationvariables should be employed.

To test for multicollinearity, an analy-sis of the variance inflation factor (VIF)was conducted. Individual VIF valuesgreater than 10 indicate a multicollinear-

ity problem (Neter, Wasserman, andKutner 1989), along with average VIFvalues greater than six. The values pre-sented in Table 1 show that problems ofmulticollinearity do not exist in any ofthe models.

Analysis and ResultsTable 2 displays some descriptive

figures on different size firms (small,medium, and large), their innovation out-comes and their degree of technologicalcollaboration. These results show that forthe full sample (6,500 observations),small firms are much less likely to inno-vate than large ones, whichever the inno-vation outcome analyzed. Thus, amarked innovation gap between SMEsand their large counterparts exists. When

the focus is limited to the subsample ofcollaborative firms (2,181 observations),however, this gap appreciably narrows.Moreover, the percentages of firms

4Specifically, several variables that the literature identifies as potential determinants were usedto instrumentalize collaboration (Izushi 2003; Miotti and Sachwald 2003; Cassiman and Veugel-ers 2002, among others): search for public finance, technological forecasting activities, partici-pation in international innovation programs, and market expansion goals. These instruments

were lagged one period with respect to the collaboration.5Smith-Blundell test statistics give different Chi-squared distributions between 0.1106 and2.0559. Their p-values were between .35 and .84. These test results are available from theauthors on request.



12/26

achieving product innovations aresimilar in all three categories of firm size.

Despite the substantial improvementin the innovation outcomes of collaborat-ing SMEs, small firms tend to collaboratefar less than large ones do. Fewer thanone out of eight small firms collaborates,a figure that rises to slightly more thanone out of three for medium-sized firmsand almost three out of four for largefirms. This tendency of SMEs to collabo-rate less than large firms is generalizable

to a large number of European countriesfor the period under study (EuropeanCommission 2004).

Figure 1 illustrates these points byshowing the historical trends of theproduct and process innovation gaps forthe years in the sample (19982002); itclearly shows the narrowing of the inno-vation gap in the subsample of collabo-rating firms, especially for product

innovations.A series of econometric analyses wereperformed to test the hypotheses putforward in the theoretical section and to

corroborate the picture suggested by thedescription of the sample. Table 3 showsthe estimation results of two series ofprobit models (Models A1 and A2) withrandom effects; these results test H1 andH2 by analyzing the impact of techno-logical collaboration on innovation out-comes (with no distinction betweenproduct and process innovation). Esti-mated coefficients and marginal effectsfrom our probit models are presented foreach model. Marginal effects were com-

puted at the means of the independentvariables and show how much the prob-ability of a firms innovating grows withan increase in that independent variable,while holding the other independentvariables constant.

Model A1 reveals the positive impactof collaboration on the likelihood offirms in the sampleregardless ofsizeinnovating. Model A2 captures

the different marginal effect of collabo-ration depending on size category.These results confirm the relationshippostulated in H1. Thus, technological

Table 2Technological Collaboration and Innovation Results

Percentageof Firms

Total Sample(6,500 observations)

Collaborative Firms(2,181 Observations)

Total SmallFirms

MediumSizedFirms

LargeFirms

Total SmallFirms

MediumSizedFirms

LargeFirms

TechnologicalCollaboration

33.5 12.2 36.3 70.1

TechnologicalInnovationa

44.6 32.5 46.6 65.2 72.2 69.5 68.3 74.5

ProductInnovation

24.6 14.6 26.9 41.3 48.2 47.8 47.2 48.7

ProcessInnovation

34.9 24.1 35.1 54.4 57 45.4 50.2 63.2

aProduct and/or process innovations.



13/26

collaboration increases the likelihoodthat SMEs achieve innovations. The Waldtests used to test H2 are shown at thebottom of Table 3. H2s postulation thatthe impact of collaboration will begreater for the subsample of SMEs is

corroborated for small firms.6 This effect,however, is not greater in medium-sizedfirms. Collaboration, then, may behelping the small firms to bridge theinnovation gap with large firms.

As expected, the marginal effect ofother innovation activities (formal R&D,design or consultants) is positive and

significant for innovation output. Thenegative and significant effect exerted byfirm age suggests that older firmsdevelop routines, procedures, and inertiathat may be barriers to innovation.

Bivariate probit models were used to

test H3 and H4. These specificationsallow us to explore the differential effectof collaboration on two measures ofinnovation outputs for each category offirm size. As explained in the section onmethodology, the r parameter is highlysignificant, signaling that the error struc-tures of the equations are correlated.

6

The robustness of these findings was tested with additional measures related to firm size.Specifically, the introduction of alternative firm size categories and the inclusion of the crossterm of sales and collaboration dummy were tested. In all cases the results were highly robust.These test results are available from the authors on request.

Figure 1Evolution of Innovation Gap and the Effect of

Technological Collaboration

Full sample

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

1998 1999 2000 2001 2002

Small firms

Medium firms

Large firms

Collaborative firms

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

60.00%

1998 1999 2000 2001 2002

Small firms

Medium firms

Large firms

Percentage of innovative firms (Process innovation)

Percentage of innovative firms (Product innovation)

Full sample

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

60.00%

70.00%

1998 1999 2000 2001 2002

Small firms

Medium firms

Large firms

Collaborative firms

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

60.00%

70.00%

80.00%

1998 1999 2000 2001 2002

Small firms

Medium firms

Large firms



14/26

Table3

InnovationPerformance,Techn

ologicalCollabora

tion,andFirmSize

ModelA1

ModelA2

Probit

Estimates

Marginal

Effects

(dy/dx)

Probit

Estimates

Marginal

Effects

(dy/dx)

Collabor

ation

0.3

91***(4.3

1)

0.1

37***(4.3

7)

SmallC

ollaboration

0.9

89***(7.0

9)

0.3

75***(8.2

3)

Medium

Collaboration

0.4

42***(3.0

7)

0.1

74***(3.0

7)

Large

Collaboration

0.6

51***(5.0

7)

0.2

54***(5.1

7)

Small

0.4

23***(2.7

6)

0.1

45***(2.9

2)

-0.8

88***(-7.1

2)

-0.3

32***(-7.8

0)

Medium

0.3

80**(2.1

1)

0.1

26**(2.3

0)

-0.4

31***(-2.7

6)

-0.1

57***(-2.9

8)

Large

0.5

59***(2.8

1)

0.1

95***(3.0

2)

-0.2

12(-1.1

6)

-0.0

80(-1.1

8)

FormalR

&D

3.6

23***(2.7

8)

1.2

82***(2.7

5)

5.1

30***(3.3

2)

1.9

69***(3.3

2)

Design

0.2

13***(2.6

1)

0.0

75***(2.5

9)

0.4

79***(6.3

1)

0.1

86***(6.3

1)

Consulta

nt

0.2

02***(2.5

8)

0.0

69**(2.4

5)

0.3

33***(4.2

9)

0.1

30***(4.2

6)

Age

-0.0

05**(-2.2

7)

-0.0

02**(-2.2

2)

-0.0

03*(-1.7

6)

-0.0

01*(-1.7

5)

Foreign

-0.0

02*(-1.8

7)

-0.0

01*(-1.8

2)

-0.0

01(-1.3

5)

-0.0

01(-1.3

5)

Export

-0.1

49(-0.8

6)

-0.0

53(-0.8

6)

0.1

49(0.9

6)

0.0

57(0.9

6)

WaldTestofFullModel:c2

173.3

4***

437.8

5***

LogPseudo-Likelihood

-1,6

60.1

3

-2,

230.4

2

Compari

sonTestsa

b

Small

Collab>

b

Large

Collab:

c2=

4.1

2**

b

Med

Collab>

b

Large

Collab:

c2=

1.2

2(n.s.)

t-Values

areinparentheses.Secto

raldummiesareincluded

inthemodels.

Marginal

effectsarecomputedats

amplemeans.For

dummyvariables,themarginaleffectcorrespondstothed

iscretechangefrom

0to

1.n.s.,non-significant.

aWaldtestcomparingthedifferen

ceofcoefficientsestimat

es.

*p

b

Large

Collab

(process):c2=

0.0

1(n.s.)

t-Values

areinparentheses.Secto

ralandyearlydummiesa

reincludedinthemodels.

Marginaleffectsareco

mputedatsample

means.Fordummyvariables,the

marginaleffectcorrespo

ndstothediscretechang

efrom

0to1.n.s.,non-significant.

aWaldte

stcomparingthediffere

nceofcoefficientsestim

ates(productinnovation

versusprocessinnovation)ineachsize

category

.

*p

b

Med

RO

(product):c2=

5.1

5**

b

Med

Vert(process)>

b

Med

RO

(process):c2=

0.0

1(n.s.)

b

Large

Vert(p

roduct)>

b

Large

RO

(product):c

2=

9.4

4***

b

LargeVert(process)>

b

Large

RO

(proce

ss):c

2=

2.4

8(n.s.)

t-Valuesa

reinparentheses.Sectoralan

dyearlydummiesareinclud

edinthemodels.

Marginaleffectsarecomputedatsamplemeans.Fordummy

variables,

themarginaleffectcorrespon

dstothediscretechangefrom


aWaldtestcomparingthedifferenceof

coefficientsestimates(verticalversusRO)ineachsizecategoryandtypeofinnovation.

*p

b

Large

Collab(p

rocess):c

2=

0.1

8(n.s.)

t-Valuesa



edinthemodels.


variables,





coefficientsestimates(produc

tinnovationversusprocessin

novation)ineachsizecatego

ry.

*p

b

Med

RO

(product):c2=

4.6

5**

b

MedV

ert(process)

b

Large

RO

(product):c2=

5.6

7**

b

Large

Vert(process)>

b

Large

RO

(proc

ess):c2=

0.2

8(n.s.)

t-Valuesa



edinthemodels.


variables,





coefficientsestimates(verticalversusRO)ineachsizecate

goryandtypeofinnovation.

*p

Technological Collaboration_ Bridging the Innovation Gap Between Small and Large Firms_ - Nieto -...

Documents

Transcript of Technological Collaboration_ Bridging the Innovation Gap Between Small and Large Firms_ - Nieto -...