proximity and scientific collaboration: evidence from the global wine industry

PROXIMITY AND SCIENTIFIC COLLABORATION:EVIDENCE FROM THE GLOBAL WINE INDUSTRY

LORENZO CASSI*, ANDREA MORRISON** & ROBERTA RABELLOTTI***

*OST-Paris and CES-University of Paris1, Paris, France. E-mail: [email protected]**URU-Utrecht University, Utrecht, the Netherlands; and Kites-Bocconi University, Milan, Italy.E-mail: [email protected]***Università degli Studi di Pavia, Pavia, Italy. E-mail: [email protected]

Received: September 2012; accepted July 2013

ABSTRACTInternational collaboration among researchers is a far from linear and straightforward process.Scientometric studies provide a good way of understanding why and how international researchcollaboration occurs and what are its costs and benefits. Our study investigates patterns ofinternational scientific collaboration in a specific field: wine related research. We test a gravitymodel that accounts for geographical, cultural, commercial, technological, structural and institu-tional differences among a group of old world (OW) and new world (NW) producers andconsumers. Our findings confirm the problems imposed by geographical and technologicaldistance on international research collaboration. Furthermore, they show that similarity in tradepatterns has a positive impact on international scientific collaboration. We also find that interna-tional research collaboration is more likely among peers; in other words, among wine producingcountries that belong to the same group, for example, OW producers or newcomers to the wineindustry,

Key words: Proximity, international scientific collaboration, wine industry, gravity model,scientometrics, emerging countries

INTRODUCTION

The worldwide scientific community is consid-ered to be more strongly interconnected nowthan ever before, because many impedimentsto international collaboration and knowledgetransfer have been reduced in the last thirtyyears or so. Information and knowledgeexchanges have been facilitated by the lowercosts of travel and long distance communica-tions, and by the diffusion of tools such asvideoconferencing. Nevertheless, knowledgetransfer remains an uncertain and complextask that can be hampered by a range offrictions, knowledge leakage risks, cultural dis-tances or failures in transmission and absorp-

tion (Cowan & Jonard 2004; Picci 2010). Thismakes international collaboration amongresearchers a far from linear and straightfor-ward process.

Scientometric studies provide a good way ofunderstanding why and how internationalresearch collaboration occurs and what are itscosts and benefits (Katz & Martin 1997). Thereis a stream of literature that focuses on acomprehensive theoretical framework to studythe determinants of research collaborationsbetween different countries, as a way of investi-gating the spatial diffusion of scientific knowl-edge (Frenken et al. 2009). This approach isbased on the concept of proximity, which hasbeen shown to be an insightful analytical device

Tijdschrift voor Economische en Sociale Geografie – 2015, DOI:10.1111/tesg.12137© 2015 Royal Dutch Geographical Society KNAG

mailto:[email protected]



to disentangle different forms of distance – geo-graphical, social, cognitive, institutional andorganisational – all of which play a role inthe spatial distribution of economic activities(Boschma 2005; Torre & Rallet 2005). As sug-gested in Frenken et al. (2009), in its differentdimensions the concept of proximity can beused to test the unevenness of knowledgediffusion on different geographical scales,identifying the main drivers of researchcollaboration.

In line with this literature, our study investi-gates patterns of international scientific col-laboration in a specific field: wine relatedresearch. In recent years, the wine industryhas experienced increased globalisation in itstrade flows, making it a particularly interestingcase for investigating whether internationalresearch collaboration has also increased andencompassed new exporting countries (Cassiet al. 2011). Recent evidence suggests thatemerging countries, such as Chile, Argentinaand South Africa are rapidly catching up interms of knowledge production, as shown bytheir increasing share in international scienti-fic publications in wine related disciplines(Glänzel & Veugelers 2006; Cassi et al. 2011).This process of technological and scientificmodernisation has been spurred on by consis-tent investments in research by newcomers andby the support provided to new specialisedresearch institutions (Giuliani et al. 2011).Cusmano et al. (2010) highlight the phenom-enon of new players in the global wine marketbecoming dynamic adapters and adopters ofthe new institutional models and promotingresearch-driven transformation of the wineindustry.

In this paper, we focus on a selected group ofcountries active both as producers and consum-ers in the global geographical wine business,and investigate their mutual research collabo-rations and the factors that might hinder/facilitate these interactions. The analysis isbased on bibliographical data covering aperiod of 13 years (from 1992 to 2004),extracted from the Web of Science (WoS)edition of the Science Citation IndexExpanded TM (SCIE) of the Institute for Scien-tific Information (ISI, Philadelphia, PA)according to some selection criteria proposedin Cassi et al. (2011). We test a gravity model

that accounts for geographical, cultural, com-mercial, technological, structural and institu-tional differences among a group of old world(OW) and new world (NW) wine producersand consumers.

In line with previous studies (Hoekman et al.2010; Picci 2010; Montobbio & Sterzi 2012),our findings confirm the problems imposedby geographical and technological distanceon international research collaboration. Ourempirical analysis provides a novel and interest-ing finding that similarity in trade patterns hasa positive impact on international scientific col-laboration. In the case of the wine industry, theglobalisations of trade and of science do notoccur as independent processes but they havea mutual influence. We also find that interna-tional research collaboration is more likelyamong peers, in other words, among wine pro-ducing countries that belong to the samegroup, for example, OW producers or newcom-ers in the wine industry.

The paper is organised as follows. The follow-ing section discusses the relevant literature,provides some background information on thewine industry and introduces the researchquestions. The third section describes the dataand methodology and the fourth section pres-ents the empirical results. The fifth section con-cludes by summarising the contributions of thispaper and suggesting some directions forfuture analysis.

THE INTERNATIONALISATIONOF RESEARCH

The literature – Scientific collaboration amongresearchers is an increasing phenomenon asshown by a striking increase in the number ofco-publications over the total number ofscientific publications (Wagner-Doebler 2001).According to Wuchty et al. (2007), collaborativeresearch accounts for well over 50 per cent of allresearch activities in many countries. Katz andMartin (1997) identify four main reasons forthis: (i) in several different fields, increasinginterdisciplinarity requires combinations ofknowledge sources; (ii) the costs of researchfacilities are leading to a pooling of resources;(iii) the increasing need for specialisationrequires ever more complex instrumenta-tion; and (iv) funding, particularly at the

2 LORENZO CASSI, ANDREA MORRISON & ROBERTA RABELLOTTI

© 2015 Royal Dutch Geographical Society KNAG

EU level, encourages international researchcollaborations.

Among the key topics of internationalisationof research, there is spatial scientometrics, agrowing field that explores the globalisationof science and the location of scientific activi-ties in specific places (Frenken et al. 2009).In studies on the geography of innovation(Breschi & Lissoni 2001; Gertler 2003;Boschma 2005; Torre & Rallet 2005), there is asmall number of empirical contributions thatfocuses on the role played by different types ofproximity – geographical, cognitive, organisa-tional, social, institutional and technological –in research collaborations (Ponds et al. 2007;Hoekman et al. 2010; Picci 2010; Paier &Scherngell 2011; Scherngell and Hu 2011;Montobbio and Sterzi 2012).

With regard to geographical proximity, thereis a general consensus on its facilitating role inknowledge exchange (see among others Jaffeet al. 1993; Audretsch and Feldman 1996). Ithas been claimed that geographical proximitybenefits research collaboration because it facili-tates face-to-face contacts and thereby theexchange of tacit knowledge. In an empiricalanalysis of co-publication intensity among 313regions in 33 European countries for theperiod 2000–2007, Hoekman et al. (2010) showthat physical distance has a negative effect onco-publication activity, although this decreasesover time and a gradual convergence towards amore interconnected European science systemis taking place.

In a different context, Scherngell and Hu(2011) analyse knowledge flows amongChinese regions proxied by co-authorship,and find robust evidence suggesting that evenmore than physical distance, cognitive distancehampers bilateral research collaboration.

As far as institutional distance is concerned,Hoekman et al. (2010, p. 663) suggest that theexistence of institutional barriers to cross-border collaboration makes it: ‘more difficult toalign incentives among researchers due to dif-ference in for instance funding schemes, insti-tutional frameworks and norms and values’.They distinguish between institutional differ-ences at three different spatial levels: regionalscience systems in which policy initiatives com-monly support collaborative projects (Cookeet al. 1997); the national science system, since

researchers in different countries are subjectto different incentive schemes; and linguisticareas with a common language that facilitatescommunication among researchers. Theirempirical findings confirm that co-publicationactivities are more likely to occur within thesame region, the same country and the samelinguistic area: ‘suggesting a mixture of simulta-neous processes driven by a ‘distance logic’ anda ‘territorial logic’ (Cooke et al. 1997, p. 472).

Ponds et al. (2007) show that institutional dif-ferences can influence collaboration patterns.In their study on the Netherlands, they findthat geographical proximity is more relevant incollaborations between academic and non-academic organisations than pure academiccollaboration. This suggests that institutionalsimilarity helps to overcome the barriers tolong distance interactions.

In a recent study on international techno-logical collaboration among patent inven-tors in advanced and emerging countries,Montobbio and Sterzi (2012) introduce tech-nological proximity alongside geographical,cultural and institutional distances, as a factorimpacting on the likelihood of co-patentingactivity. They find that the probability of tech-nological collaboration between inventorsbased in two different countries is higher ifcompanies and institutions are active in similartechnological fields. The higher marginalbenefit derived from collaboration with a tech-nologically similar partner partially offsets thecosts of collaborating with a geographicallydistant partner. Similarly, Picci (2010) analysespatterns of inventive activity in Europe usingpatent data and concludes that inventiveproximity is a strong determinant of bilateralcollaboration. Technological and cognitiveproximity facilitates collaboration also in thecase of formal research cooperation, such asEuropean Framework Program projects (Paier& Scherngell 2011).

In applied scientific fields, such as the case ofresearch on wine related disciplines, commonindustry characteristics (i.e. foreign investors,trade flows, climatic and soil conditions) mayalso influence collaboration among research-ers in different countries. The following subsec-tion provides some background informationon the wine industry and the role of science inits recent development.

PROXIMITY AND SCIENTIFIC COLLABORATION 3


Scientific research in the wine industry – Globalpatterns of wine production and trade havebeen changing fundamentally since 1980. Upto the end of the 1980s, OW countries, andparticularly France and Italy, dominated theinternational wine market. From the beginningof the 1990s, their supremacy began to be chal-lenged by new international players that haverecorded some spectacular performances interms of both exported volumes and values.These NW producers include affluent coun-tries that are relatively new to the wine sector,such as the USA and Australia, and less devel-oped, but rapidly growing emerging econo-mies, such as Chile, Argentina and South Africa(Giuliani et al. 2011). Whereas OW producers(France, Italy, Spain and Portugal) still lead inthe production, export and consumption ofwine, NW producers are gaining market shareamong consumers around the world, from only2.5 per cent of world exports in the early 1980sto more than 28 per cent in 2010 (OIV 2012).Increasingly, NW producers are gaining recog-nition in the high-end segments of the marketthat once were dominated by an elite group ofOW producers. Also, NW producers have beenquick to adapt their wines to new expandingmarkets, playing a major role in establishingand strengthening the emergence of a newparadigm based on a market-driven scientificapproach to wine production. As documentedin Giuliani et al. (2011), in the NW, intermedi-ary organisations, universities and researchcentres have been restructured and strength-ened to adapt to intensified internationalcompetition.

In this changing global context, Cassi et al.(2012) investigated the joint evolution of tradeand scientific collaboration networks. Theyargue that investment in science is both a pre-cursor to changes in the production and tradeof wine and that developments in internationaltrade are changing the direction of investmentand international collaboration in science.Their results indicate that a strong interdepen-dence between science and trade is affectinghow the dynamics of globalisation unfold, anddevelopments in networks of trade and scien-tific collaboration are increasingly occurring inparallel. Based on these findings, in our analysisof the determinants of international collabora-tion we introduce a measure for commercial

proximity. This assumes that two countries tar-geting the same final market face similar tech-nical problems (e.g. regulations on sulphites)and need to satisfy similar consumer tastes, andhence are likely to collaborate more.

In line with the recent literature on proxim-ity in research collaboration, and on the basis ofsome industry peculiarities characterising thewine industry, in this paper we address the fol-lowing research questions:

• Do geographical, technological and culturaldistances matter for establishing scientificcollaborations?

• To what extent do bilateral patterns of winetrade affect the extent to which countriescollaborate in science in the wine field?

• Does being an OW or a NW producer ormainly a wine consuming country, contributeto explaining the patterns of internationalscientific collaborations?

In order to answer these questions, we con-struct a country-level dataset that combinestrade and scientific publication data for a15-year period (1990–2004). We test a gravitymodel that includes geographical, cultural,commercial and technological distances as wellas differences among groups of wine exportingand importing countries.

DATA AND METHODOLOGY

Data – The empirical analysis is based on threemain data sources: bilateral trade flows, scien-tific co-publications and geographical, linguis-tic and historical relations between countries.Trade data come from the NBER-UnitedNations Trade Data (NBER 1962–2000) andthe COMPENDIUM (Anderson & Norman2006) datasets. The first database providestrade-bilateral flows at the 4 digit SIC sectorlevel from 1962 to 1999. The second database isspecific to the wine sector and provides infor-mation on international trade-bilateral flows(values) between the main wine importing andexporting countries for 1994 to 2004. In orderto extend the period of analysis, we mergedthese two data sources. In our final dataset, datafrom 1970 to 1993 are from NBER1 and datafrom 1994 to 2004 are from the COMPEN-DIUM.2 The empirical analysis includes the 24countries in the COMPENDIUM dataset, which



register an annual share of at least 1.5 per centof world global wine trade flows during theperiod 1980–2004.3 All the countries consid-ered account for more than 95 per cent ofworldwide wine export flows and more than97 per cent of wine related internationalco-publications in 2004.

Figure 1 shows patterns of trade for theperiod 1993–2004. Overall, international tradebetween countries has grown, with both OWand NW countries experiencing positivetrends. However, NW countries have grownmuch faster than OW countries, with Australiaand Chile the top performers among wineexporters.

Data on scientific co-publications areextracted from the WoS edition of the ISI andcover a period of 18 years from 1989 to 2006.The number of publications is a measure ofresearch activity (for a critical appraisal see Katz& Martin 1997). In order to define the field ofwine research, we adopt the same criteria as inCassi et al. (2011), which builds on Glänzel andVeugelers (2006).

The dataset contains 12,373 publicationsselected on the basis of three search criteria:

1. lexical criterion including specific searchstrings applied to publication keywords,titles and abstracts;4

2. journal criterion including all articles inthe three top journals – American Journal ofEnology and Viticulture, Australian Journal ofGrape and Wine Research, and Vitis;5

3. exclusion criterion excluding all articleswhere at least one of the authors is affiliatedto a hospital or a medical school becausethese articles are related not to wine produc-tion but to research on the health benefits ofor damage caused by wine consumption.

Figure 2 depicts co-publication patterns bygroups of countries. We group 24 countriesfollowing Cassi et al. (2012): the OW groupincludes traditional producers, the NW groupincludes emerging exporters, the core consum-ers (CONS) group comprises the major inter-national importers, and peripheral exportersand consumers (PER) is a mixed group includ-ing peripheral producers and emergingimporters with little or no production.6 Overall,international bilateral collaboration increasesover the period. All groups show positivegrowth rates. However, unlike the case for tradeflows, we observe that OW countries havegrown faster than any other group. In par-ticular, the figures suggest that the OW coun-tries are driving the worldwide growth inco-publication and that the NW countriesare only marginal contributors, their share

0

2000

4000

6000

8000

10000

12000

14000

16000

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

NW

OW

WW

Notes : Old world (OW); new world (NW); worldwide (WW).Source : Authors’ elaboration of NBER trade and Compendium data.

Figure 1. Wine exports by group of countries (1992–2004).



remaining stable from 1996 to 2001.7 After2001, NW and the other groups show a steepgrowth path.

Other information, such as geographical, lin-guistic and historical relations between coun-tries is retrieved from the CEPII gravity dataset.8

Finally, GDP data were obtained from the USDepartment of Agriculture Economic ResearchService.9

Model and variables – To model scientific col-laboration between countries we exploit agravity model, widely used in the trade litera-ture to explain inter-country trade flows(Anderson 1979; Bergstrand 1985). The gravitymodel resembles the Newtonian law of gravita-tion, where masses are proxied by the country’seconomic size and distance is measured bytheir geographical position. The model impliesthat trade between two countries increaseswith their economic size (i.e. their mass) anddecreases with their geographical distance. Pre-vious examples of the gravity framework in thescientific (Ponds et al. 2007; Hoekman et al.2010; Scherngell & Hu 2011) and technologi-

cal collaboration (Peri 2005; Picci 2010; Paier &Scherngell 2011; Scherngell & Hu 2011;Montobbio & Sterzi 2012) literature wereaimed at explaining the intensity of knowledgeflows between countries or organisations.Overall, these empirical studies show that col-laboration is positively related to actors’ size(mass) (e.g. number of publications) and nega-tively related to the distance between them.In line with this literature, our gravity modelis based on the following equation:

ln lnIntCopub A A geodistL T

ijt it jt ijt

ijt ijt

( ) = + + ( )+ +β β β

β β0 1 2

3 4 +++ +

βδ ε

5CW

ijt

ij ijt (1)

where IntCopubijt is the natural log of thenumber of co-publications between authorsaffiliated to organisations in country i and theirco-authors affiliated to organisations located incountry j, at time t. As usual, we define aresearch collaboration involving at least twocountries as a co-authorship (see Katz & Martin1997) if their respective scholars co-author atleast one scientific article.

0

50

100

150

200

250

300

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

NWOWWorld WideCONSPER

Notes : Old world (OW); new world (NW); consumer countries (CONS); peripheral exporters and consumers(PER).Source : Authors’ elaboration of ISI-Web of Knowledge data.

Figure 2. Wine co-publications by groups of country (1992–2004).



Ait and Ajt respectively capture the mass ofcountries i and j at time t. Since the co-publication relationship is undirected (unliketrade where there is a source and a destina-tion), the coefficients of the two countries areassumed to be equal and the effect of their twomasses is captured by β1. In our analysis, mass ismeasured as the economic size of the countryproxied by the natural log of the annual GDP(lngdp) and the share of wine exports (lnexport),and scientific relevance, proxied by the log ofthe country’s annual publications (lnpub).Overall, a country’s economic and scientificrelevance indicates the level of absorptivecapacity and the scientific infrastructure.Hence, in line with the literature, we can expecta country’s international scientific collabora-tion to be positively influenced by its economicand scientific size. However, it should be notedthat size could have a negative effect on inter-national collaboration. Indeed, if a country’snational scientific community is well devel-oped, there might be fewer incentives for itsscholars to search for foreign partners. On theother hand, scholars that belong to smallnational scientific communities are likely to behighly motivated to establish collaborationsabroad in order to avoid isolation. Althoughthe above arguments hold in general, it is worthhighlighting that in the wine sector a sizablegroup of small countries has a large domesticwine industry (e.g. Austria, Portugal, Greece,Moldova, Armenia). In these cases, despitetheir relatively small economic size, suchstrong specialisation might entail intensiveinternational scientific collaborations.

In our model, the distance terms refer todifferent types of proximity. The trade litera-ture focuses mostly on geographical distanceand on its negative effect for trade relations.The underlying hypotheses are that transporta-tion and communication costs increase withdistance. Similarly, gravity studies on knowl-edge flows confirm the negative effect ofgeographical distance on technological col-laboration (Hoekman et al. 2010; Picci 2010). Alarge part of scientific and technological knowl-edge is tacit; hence it is more effectively trans-ferred and absorbed if the actors are physicallyclose. These latter hypotheses matter evenmore in the wine case, which is an appliedscientific field where researchers usually need

to conduct experiments on the field in orderto gather data and investigate local-specificplant or vineyard conditions. In our model, thegeographical distance between two countries(geodist) is measured as the log of the distance inkilometres between their capital cities (lndist inthe econometric model).10

However, as discussed above, other types ofdistance can hinder scientific interactions(Katz & Martin 1997; Hoekman et al. 2010)and general knowledge exchanges (Boschma2005). In our gravity model, we account for thetechnological, commercial and cultural dis-tance between two countries, defined asexplained below.

Technological Distance (T) is measured as thecomplement of Jaffe’s (1988) index, using the25 ISI-Web of Science sub-categories:

Dist tech P P P P P Pit jt it t jt jt_ ijt = − ′ ′( ) ′( )[ ]( )1 1 2

where Pi is the distribution vector of country I’sstock of publications at time t for the 25sub-categories identified as wine-related.11 Weexpect two countries to collaborate more if theyare technologically close, that is, if they areactive in similar scientific fields. The Jaffe indexranges between 0 and 1; it is equal to 1 if thepairs of countries have the same distribution ofscientific activities, and 0 if they do not shareany of these activities.

For Commercial Distance (C) we introduce twomeasures. First, and similar to TechnologicalDistance, the complement of Jaffe’s index(Dist_commijt) is computed using export quotasfor year t for the 24 potential partner countries.This variable is equal to zero if two countries arecompeting in the same foreign market, that is,exporting the same proportions to the samecountries, and tends to its maximum value (i.e.1) the more these two countries export to dif-ferent countries. Following Cassi et al. (2012),we hypothesise that the Commercial Distancebetween two countries is negatively correlatedwith their scientific collaboration. Indeed,winemakers who target the same marketsusually sell to customers with similar consump-tion habits and tastes and this implies thatthey often adopt similar oenological practices(i.e. oxygenation) and face the same technicalbarriers (e.g. regulation about sulphites).



Therefore, we might expect a more intensecollaboration than among winemakers target-ing very different markets.

Another indicator introduced into themodel is the Salton Index, which captures therelative importance of market j for country i:

Salton X X Mijt ijt it jt= ×( )( )1 2 .

where Xi and Mj respectively represent the totalvalue of exports of country i and imports ofcountry j and Xij measures the value of theexports of country i to country j (in 2000 USDollars). By construction, it is:

0 1≤ ≤ ≠Salton Salton Saltonij ij jiand .

Therefore we know that Saltonij = 0 if country idoes not export to country j; Saltonij = 1 ifcountry i exports only to country j and thislatter does not import wine from any othercountry.

Different from Commercial Distance, whichconsiders each country relation to any othercountry, the Salton Index captures the impor-tance of country i’s export to country j’s market(i.e. imports). Since we only take account ofnon-directed relations, we compute this indica-tor as the average value of Saltonij and Saltonji.

The obtained measure captures the (average)reciprocal importance of the two countries’markets. As for the indicator of CommercialDistance, even in this case we expect that thescientific collaboration between two countriesincreases together with their commercialdependence, since exporters tend to adapttheir wines to local tastes and scientificresearch might facilitate this adjustmentprocess (Giuliani et al. 2011).12

Cultural and historical differences, such as ashared language or being a former colony, alsoinfluence the likelihood of scientific collabora-tion among researchers in different countries.These latter factors certainly have a role also inthe wine sector, where there are examples ofcolonial linkages (e.g. Spain and Chile orArgentina) and language similarities (e.g. theUK, Australia, New Zealand, the Netherlandsand South Africa). Hence, in line with the exist-ing gravity literature and in order to capturesuch time invariant features (L), in the econo-metric model we include:

• a dummy equal to 1 if the pair of countrieshas an official language in common(Comlang_off) and 0 otherwise;

• a dummy equal to 1 if the pair of countrieshad colonial ties before 1945 (Colony) and 0otherwise.

To take into account the differences discussedabove among groups of wine countries, a set ofcontrol dummies (W) is included in the model.We expect that producing countries withsimilar histories in wine production (either NWor OW wine producers) have closer researchcollaborations. In particular, over time OWcountries have set up well-established institu-tions supporting the development of their wineindustries. Hence, we expect deep-rootedlinkages among them, which are likely tostrengthen the scientific collaboration amongresearchers, specialised in the wine field. ForNW countries, we can expect intense researchcollaboration based on common problems andsimilar institutional frameworks.13

At the same time, since NW countries arelatecomers in the sector, their knowledge basein wine might be less extensive than that of theOW countries. Also, some NW countries, suchas South Africa, Chile and Argentina, areemerging economies, hence their scientificinstitutions might be less advanced comparedto those in OW countries (Kunc & Tiffin2011). These arguments suggest that scientificorganisations in NW countries may interactmore intensively with established researchcentres located in OW countries rather thanwith other countries within their group. Basedon what has been said so far, our predictionabout the NW country (NW) dummy is open.

For consumer countries (CONS), ceterisparibus we might expect that they would col-laborate more intensively with either NW orOW producer countries than among them-selves since they are likely to lack specific exper-tise in the sector. The same expectation holdsfor the PER group.

The gravity model in equation (1) can beestimated using different econometric tech-niques. Studies by Santos Silva and Tenreyro(2006) and Burger et al. (2009) show that stan-dard techniques, such as ordinary least squares(OLS), generate biased estimates due to anumber of different econometric problems



(e.g. heteroscedasticity, excess zeros). Also,unlike the gravity models that investigate thedeterminants of bilateral trade, analysis of sci-entific collaborations implies the use of countdata, which makes OLS an even less suitablemethod of analysis.

A ‘natural’ solution to these shortcomings isto use the Poisson pseudo-maximum likelihood(PPML) estimator or the related negative bino-mial or Zero inflated negative binomial (SantosSilva and Tenreyro 2006; Burger et al. 2009).We chose the PPML estimator as particularlysuitable since in our case the dependent vari-able (i.e. co-publications) is a count variableand its distribution is skewed. All econometricspecifications use robust standard errors thatare clustered to control for error correlation inthe panel (Cameron & Golotvina 2005). Wealso include country and time dummies tocontrol for unobservable characteristics. Adetailed description of the variables introducedin the econometric analysis is included inTable 1.

EMPIRICAL RESULTS

We test the gravity equation (1) using paneldata on bilateral scientific co-publications for aset of 24 countries. Table 1 presents our resultsunder different econometric specifications; all

of them include time and country dummies.Country dummies control for unobservablefactors (e.g. macroeconomic or political insta-bility, cultural differences not accounted for byother variables, institutional factors, govern-ment policies or differences in stocks of humancapital) in both countries i and j. Timedummies capture unobservable time-varyingfactors that may affect bilateral relationsbetween countries (e.g. changes in politicalconditions and in public policies).

The first column in Table 2 shows the esti-mates of the basic gravity model, which includemass and geographical distance. We find thatthe Geographical Distance is strongly significantand with the expected sign, showing that geog-raphy hinders international collaborations.This is in line with other studies that useco-authorship as a proxy for scientific collabo-ration and find that geography matters in dis-ciplines as different as physics, biotechnologyand humanities, and in areas as diverse as Euro-pean and Chinese regions (Ponds et al. 2007;Hoekman et al. 2010; Scherngell and Hu 2011).We also investigate whether the impact of Geo-graphical Distance on bilateral collaborationschanges over time. The coefficient of theparameter does not follow a clear trend and itdoes not decrease,14 on the contrary it slightlyincreases over time. This finding confirms find-

Table 1. List of Variables.

IntCopubijt Log of the number of co–publications between one author affiliated in an organisationlocalised in country i and another author affiliated to an organisation in country j attime t

Lngdp Log of GDP of countries i and jLnexp Log of exports of countries i and jLnpub Log of publications of countries i and jLndist Log of the Geographical Distance between each pair of countries in kmDist_tech Technological DistanceDist_comm Commercial DistanceSalton Salton IndexComlang_off Dummy variable which is 1 if the pair of countries has an official language in common, 0

otherwiseColony Dummy variable which is 1 if the pair of countries shares a colonial past, 0 otherwiseOW Dummy variable which is 1 if the pair of countries is OW (OW), 0 otherwiseNW Dummy variable which is 1 if the pair of countries is NW (NW), 0 otherwiseCONS Dummy variable which is 1 if the pair of countries is Consumer (CONS), 0 otherwisePER Dummy variable which is 1 if the pair of countries is Peripheral Consumer or Producer

(PER), 0 otherwiseNOW Dummy variable which is 1 if the pair of countries is OW (OW) and NW (NW), 0

otherwise



ings in similar studies on bilateral knowledgeflows (Hoekman et al. 2010; Montobbio &Sterzi 2012) and is generally in line with evi-dence in the trade literature on the ‘missingglobalisation puzzle’ (Coe et al. 2002; Disdier &Head 2008).

As expected, the mass indicator has a positiveeffect on bilateral collaboration between coun-tries, but not on exports, which is not signifi-cant although it has a positive sign. Both GDPand number of publications are positive deter-minants of international co-authorship, whichconfirms that the logic of the gravity equation

works also in the context of science-based rela-tions between countries. In line with the litera-ture, we find that the economic and scientificsizes of countries are strong attractors.

In Table 2, columns 2 and 3 present the aug-mented versions of the gravity model. Column2 includes the measures for different distancesand the results confirm most of our expecta-tions. The Salton Index is positive andsignificant suggesting that tight trade interde-pendence among countries is an importantdeterminant of scientific collaboration. Asexpected, the Commercial and Technological

Table 2. PPML estimation findings.

(1) (2) (3)

Lndist −0.305*** −0.233*** −0.224***[0.0596] [0.0578] [0.0593]

Lnpub 1.131*** 0.910*** 0.940***[0.102] [0.117] [0.115]

Lngdp 1.324* 1.394** 1.344*[0.702] [0.705] [0.708]

Salton 1.740*** 2.014***[0.560] [0.550]

Dist_comm −0.355* −0.251[0.188] [0.198]

Comlang_off −0.0966 −0.249[0.244] [0.241]

Colony −0.0268 0.0754[0.152] [0.122]

Dist_tech −1.439*** −1.297***[0.287] [0.298]

OW 0.685***[0.207]

NW 0.500*[0.271]

NOW 0.127[0.384]

CONS −0.627***[0.218]

PER 0.558**[0.276]

Lnexp 0.0331[0.0448]

Constant −17.02** −16.80** −16.60**[7.193] [7.136] [7.039]

Observations 3,446 3,446 3,446Year dummy YES YES YESCountry dummy i YES YES YESCountry dummy j YES YES YESPseudo log-likelihood −1819.2938 −1794.4781 −1781.2097R-squared 0.581 0.593 0.598

Note : ***p < 0.01, **p < 0.05, *p < 0.1.



Distances both have a negative impact on bilat-eral collaboration between countries. The elas-ticity is particularly high for the TechnologicalDistance indicating that collaboration in sciencerequires both actors to be active in similarfields. Indeed, actors with a common scientificbackground can more easily communicate,understand each other and eventually fullymaster the knowledge diffused.15 This result issupported by empirical work on the determi-nants of international knowledge exchange(see Picci 2010; Montobbio and Sterzi 2012).

In contrast to most of the existing empiricalwork, the indicators of cultural similarity(Comlang_off and Colony) are insignificant andthis might signal some peculiarities of thesector under investigation. If we examine thenetwork of collaborations between countries(Figure 3), we observe that former colonies orcountries with a common language are oftenweakly connected or not connected at all. Forexample, Chile has links with France but notwith Spain, despite the common language andthe colonising history. Argentina collaboratesonly with Spain. South Africa is linked with theUSA and Australia and also collaborates withthe Netherlands, which colonised the country,but it does not have links with the UK, whichruled the country for centuries. The sameapplies to former British territories, such asAustralia and New Zealand. These peculiaritiesmay be explained by the inclusion of dummiesto control for the type of wine country (i.e. OW,NW, CONS, PER) as in column 3 of Table 2.

Our findings indicate that countries belong-ing to the same group, that is, OW, NW, PER,show higher intensity of bilateral collabora-tions. The dummies also capture an additionaldimension of similarity: wine industry incum-bents (OW) tend to collaborate more with theirpeers, latecomers (NW) have stronger interac-tions among the NW group and inter-groupcollaborations between OW and NW (see NOWvariable in Table 2) are not significant. Thesefindings confirm that countries within the samegroup face similar problems (e.g. climate con-ditions) and are characterised by similar insti-tutional settings (e.g. trade agreements, healthand environmental standards). Therefore,intra-group collaborations might offer moreopportunities to solve problems or promotenew ideas for innovation, than relations with

researchers located in countries that belong toa different group. In non-producing countriessimilarity reduces collaboration.16

CONCLUSIONS

The barriers to international research collabo-ration represented by geographical and tech-nological distances are the focus of severalempirical studies in different scientific disci-plines and diverse geographic contexts(Hoekman et al. 2010; Picci 2010; Scherngell &Hu 2011; Montobbio & Sterzi 2012). Our studyconfirms that geography and a common scien-tific background are important for interna-tional collaboration. Our evidence also showsthat the importance of physical distance doesnot decline over time. This finding suggeststhat globalisation generally is constrained bygeography, despite the increase in interna-tional scientific collaborations and sustainedgrowth in international trade.

Empirical analysis provides a novel and inter-esting result showing that similarity in tradepatterns has a positive impact on internationalscientific collaborations. Economic globalisa-tion (through trade) and knowledge globalisa-tion (proxied by scientific co-publication) arenot independent processes but they ratherinfluence each other. This means that interna-tional scientific collaboration could facilitatethe adaptation of wines to local tastes andtherefore might increase when countries areconnected through trade relations.

A peculiar and original result is that intra-group collaborations are favoured by theoccurrence of similar problems (e.g. climateconditions) and analogous institutional settings(e.g. trade agreements, health and environmen-tal standards). This confirms the existence oftwo different patterns in wine production: a NWparadigm based on a market-driven scientificapproach and an OW model characterised by anemphasis on terroir and regional specificities,and by a strict regulatory framework imposingadditional constraints and reinforcing regionaldifferences (Giuliani et al. 2011).

The importance of terroir or special charac-teristics of each different agricultural local soil,weather conditions and farming techniques,has a strong influence on the characteristics ofthe wines produced and on the oenological and



The International Network of Research Collaborations (1992–1996)



Figure 3. The International Network of Research Collaborations (1992–2004).



viticulture competencies required to growvines. In future analyses, the introduction of avariable to test the importance of terroir wouldcontribute to improving our understanding ofthe specificity of this industry.17 Territorialspecificities would become more evident inanalyses at a regional level. This might be par-ticularly interesting in the case of EU countriesfor capturing the richness and extreme varietyof old producing countries such as France, Italyand Spain.

Acknowledgements

Funding from Fondazione Cassa di Risparmio diTorino (CRT) is gratefully acknowledged.

Notes

1. In the SIC classification we select Code 1121.2. The COMPENDIUM dataset is chosen for the

overlapping period, which allows us to check fordiscrepancies.

3. The selected countries are: Argentina, Australia,Austria, Belgium, Bulgaria, Canada, Chile,Denmark, France, Germany, Great Britain,Greece, Hungary, Ireland, Italy, Japan, the Neth-erlands, New Zealand, Portugal, South Africa,Spain, Sweden, Switzerland, and the UnitedStates. We excluded the former Communistcountries (e.g. USSR/Russia, FM Yugoslavia, theRepublic of Moldova, and the Czech Republic)because of lack of territorial consistency alongthe period of analysis, and also some Asian coun-tries (e.g. Singapore and Taiwan) because theCOMPENDIUM dataset provides only regionalaggregated data for them.

4. We used the search terms: GRAPEVIN* ORWINES OR WINE GRAP* OR WINE PRO* ORRED WINE* OR WHITE WINE* ORWINEMAKING OR ENOLOG* OR VITICULT*OR OENOLOG* OR WINE CELL* OR WINEYEAST* OR WINERY OR WINERIES OR VITIS.In line with Glänzel and Veugelers (2006), wedefined and tested the set of search terms. Westarted our search with the term wine, whichultimately was excluded because it produced sig-nificant noise in the results. Also, in most of theextracted publications, the word wine appears inthe title or abstract or as a keyword.

5. In line with Cassi et al. (2011) and different fromGlänzel and Veugelers (2006), we include thejournal Vitis.

6. We classify Bulgaria, France, Greece, Hungary,Italy, Portugal and Spain as OW producers;Argentina, Australia, Chile, New Zealand andSouth Africa as NW producers; Belgium,Canada, Great Britain, Ireland, Japan, the Neth-erlands, Switzerland and the United States asCONS; Austria, Denmark, Germany, andSweden as PER.

7. Note that the positive trend in CONS is due tothe inclusion of the USA.

8. Data retrieved from http://www.cepii.fr/anglaisgraph/bdd/gravity.htmn (last accessed24 September 2012). See Appendix for adetailed description.

9. Available at http://www.ers.usda.gov/data/ (lastaccessed 24 September 2012).

10. For a detailed description of the Geodist datasetfrom which the gravity variables are taken, seethe Appendix.

11. Different classifications might lead to discordantresults due to the granularity of the measure‘technological distance’, as pointed out inSavorelli and Picci, 2013.Unfortunately we areunable to test different measures of technologi-cal distance. In the wine scientific field there isno other classification available in the literature,the one derived from ISI being the most reliable.

12. Note that, by construction, the expected sign ofthe Salton Index is the opposite of the CommercialDistance indicator.

13. For instance, Cusmano et al. (2010) stress thatSouth Africa first and Chile more recently havebuilt systems of supporting institutions inspiredby the successful Australian wine system.

14. In the basic gravity model (see column 1) weinteracted the time dummies with the geographi-cal distance indicator to estimate the effect ofdistance in different years. Results do not show aclear trend. Since we have a rather small timeseries, we also tested the distance effect on twosub-periods. We find that the coefficient (–0.21)is not significant in the first period (1992–1998),while it is higher and significant (–0.33) in thesecond period (1999–2004).

15. We have also tested the non-linearity relationbetween the probability of collaboration and thetechnological distance, considering this latteralso in square terms. Doing so, we check for theexistence of an inverted U-shape curve that canbe interpreted as a research for diversity (i.e.complementarity) constrained by the costs ofcommunication and collaboration. However, the



http://www.cepii.fr/anglaisgraph/bdd/gravity.htmn

http://www.cepii.fr/anglaisgraph/bdd/gravity.htmn

http://www.ers.usda.gov/data/

square term is not significant. For the sake ofreadability, this specification has not beenreported in the paper.

16. Note that the inclusion of dummies does notaffect the sign or significance level of distance(other than commercial distance).

17. In a previous version of this analysis, we made anattempt in this direction including the distanceof the capital city from the equator, to capturethe climate conditions of each country assumingthat wine producers facing similar weather con-ditions share common problems (e.g. weatherconditions, diseases). However, due to theroughness of the measure used, the variable isnot significant in the econometric test and there-fore is not included in the current econometricmodel.

REFERENCES

Anderson, J.E. (1979), A Theoretical Foundationfor the Gravity Equation, The American EconomicReview 69, pp. 106–116.

Anderson, K. & D. Norman (2006), Compendium ofWine Statistics. Adelaide: Centre for InternationalEconomic Studies, University of Adelaide.

Audretsch, D.B. & M.P. Feldman (1996), R&DSpillovers and the Geography of Innovation andProduction. American Economic Review 86, pp. 630–640.

Bergstrand, J.H. (1985), The Gravity Equation inInternational Trade: Some Microeconomic Foun-dations and Empirical Evidence. Review of Econom-ics and Statistics 67, pp. 474–481.

Boschma, R.A. (2005), Proximity and Innovation: ACritical Assessment. Regional Studies 39, pp. 1–14.

Breschi, S. & F. Lissoni (2001), KnowledgeSpillovers and Local Innovation Systems: A CriticalSurvey. Industrial and Corporate Change 10, pp. 975–1005.

Burger, M.J., F.G. van Oort & G.-J. M. Linders(2009), ‘On the Specification of the GravityModel of Trade: Zeros, Excess Zeros and Zero-Inflated Estimation’, ERIM report series researchin management ERS–2009–003–ORG, RSMErasmus University/Erasmus School ofEconomics.

Cameron, A.C. & N. Golotvina (2005), Estimationof Country-pair Data Models Controlling for ClusteredErrors: With International Trade Applications.Working Paper 06–13, Department of Economics,University of California at Davis.

Cassi, L., A. Morrison & R. Rabellotti (2011), TheChanging Geography of Science in Wine: Evidencefrom Emerging Countries. In: E. Giuliani, A.Morrison & R. Rabellotti, eds., Innovation andTechnological Catch Up: The Changing Geography ofWine Production, pp. 43–69. Cheltenham: EdwardElgar.

Cassi, L., A. Morrison & A. ter Wal (2012), TheEvolution of Trade and Scientific CollaborationNetworks in the Global Wine Sector: A Longitudi-nal Study using Network Analysis. Economic Geogra-phy 88, pp. 311–334.

Cowan, R. & N. Jonard (2004), Network Structureand the Diffusion of Knowledge. Journal of EconomicDynamics and Control 28, pp. 1557–1575.

Cusmano, L., A. Morrison & R. Rabellotti (2010),Catching–up Trajectories in the Wine Sector: AComparative Study of Chile, Italy and South Africa.World Development 38, pp. 1588–1602.

Coe, D.T., A. Subramanian, N.T. Tamirisa & R.Bhavnani (2002), The Missing Globalization Puzzle.IMF Working Paper 02/171, Washington, DC.

Cooke, P., M. Gomez–Uranga & G. Etxebarria(1997), Regional Innovation Systems: Institutionaland Organisational Dimensions. Research Policy 26,pp. 475–491.

Disdier, A.-C. & K. Head (2008), The Puzzling Per-sistence of the Distance Effect on Bilateral Trade.Review of Economics and Statistics 90, pp. 37–48.

Frenken, K., S. Hardeman & J. Hoekman (2009),Spatial Scientometrics: Towards a Cumulative Re-searchProgram. Journal of Infometrics3,pp.222–232.

Gertler, M.S. (2003), Tacit Knowledge and theEconomic Geography of Context, or the Indefin-able Tacitness of Being (There). Journal of EconomicGeography 3, pp. 75–99.

Giuliani, E., A. Morrison & R. Rabellotti, eds.(2011), Innovation and Technological Catch Up: TheChanging Geography of Wine Production. Chelten-ham: Edward Elgar.

Glänzel, W. & R. Veugelers (2006), Science forWine: A Bibliometric Assessment of Wine andGrape Research for Wine Producing and Consum-ing Countries. American Journal of Enology and Viti-culture 57, pp. 23–32.

Hoekman, J., K. Frenken & R.J.W. Tijssen (2010),Research Collaboration at a Distance: ChangingSpatial Patterns of Scientific Collaboration withinEurope. Research Policy 39, pp. 662–673.

Jaffe, A.B. (1988), Demand and Supply Influences inR & D Intensity and Productivity Growth. Review ofEconomics and Statistics 70, pp. 431–437.



Jaffe, A B., M. Trajtenberg & R. Henderson(1993), Geographic localization of knowledgespillovers as evidenced by patient citations”, TheQuarterly Journal of Economics 108, pp. 577–598.

Katz, J.S. & B.R. Martin (1997), ‘What is ResearchCollaboration?’, Research Policy 26, pp. 1–18.

Kunc, M. & S. Tiffin (2011), University Involvementin Wine Region Development: A ComparativeCase Study between Universidad de Talca (Chile)and Universidad de Cuyo (Argentina). In: E.Giuliani, A. Morrison & R. Rabellotti, eds.,Innovation and Technological Catch Up: The ChangingGeography of Wine Production, pp. 95–117. Chelten-ham: Edward Elgar.

Mayer, T. & S. Zignago (2011), Notes on CEPII’sDistances Measures (GeoDist). CEPII Working Paper2011–25.

Montobbio, F. & V. Sterzi (2012), The Globalizationof Technology in Emerging Markets: A Gravity Model onthe Determinants of International Patent Collaborations.Working Paper 08/2012, Università di Torino.

NBER (1962–2000), NBER United Nations WorldTrade Data. Available at <http://cid.econ.ucdavis.edu/>. Accessed on 26 September 2012.

OIV (2012), Statistical Report on World Viticulture.Paris: International Organization of Wine andVine.

Paier, M. & T. Scherngell (2011), Determinants ofCollaboration in European R&D Networks:Empirical Evidence from a Discrete Choice Model,Industry and Innovation 18, pp. 89–104.

Peri, G. (2005), Determinants of Knowledge Flowsand Their Effect on Innovation, Review of Economicsand Statistics 87, pp. 308–322.

Picci, L. (2010), The Internationalization of Inven-tive Activity: A Gravity Model Using Patent Data.Research Policy 39, pp. 1070–1081.

Ponds, R.H.F., F.G. van Oort & K. Frenken (2007),The Geographical and Institutional Proximity ofResearch Collaboration. Papers in Regional Science86, pp. 423–443.

Santos Silva, J.M.C. & S. Tenreyro (2006), TheLog of Gravity. Review of Economics and Statistics 88,pp. 641–658.

Savorelli, L. & L. Picci (2013), The TechnologicalSpecialization of Countries: An Analysis of Patent Data.

Discussion Paper 201301, University of St.Andrews.

Scherngell, T. & Y. Hu (2011), CollaborativeKnowledge Production in China. Regional Evi-dence from a Gravity Model Approach. RegionalStudies 45, pp. 755–772

Torre, A. & A. Rallet (2005), Proximity and Local-ization. Regional Studies 39, pp. 47–59.

Wagner-Doebler, R. (2001), Continuity and Dis-continuity of Collaboration Behaviour since 1800from a Bibliometric Point of View. Scientometrics 52,pp. 503–517.

Wuchty, S., B.F. Jones & B. Uzzi (2007), TheIncreasing Dominance of Teams in the Productionof Knowledge. Science 316, pp. 1036–1039.

APPENDIX

Gravity variables – GeoDist made availablean exhaustive set of gravity variables pro-vided by Cepii (see Mayer & Zignago (2011) formore details). The dataset is available at thefollowing web address: http://www.cepii.fr/anglaisgraph/bdd/distances.htm

The GeoDist provides two distinct files:1. a country–specific one: geo_cepii2. a dyadic one: dist_cepii including a set of

different distance and common dummyvariables used in gravity equations to identifyparticular links between countries suchas colonial past, common languages,contiguity.

In our exercise, we used the following vari-ables taken from the dyadic dataset• Geodist measures the distance in Km between

two countries. Geodesic distances are calcu-lated following the great circle formula,which uses latitudes and longitudes of themost important cities/agglomerations (interms of population);

• Comlang_off indicates if two countries share acommon official language;

• Colony indicates if two countries have everhad a colonial link.



http://cid.econ.ucdavis.edu/

http://cid.econ.ucdavis.edu/

http://www.cepii.fr/anglaisgraph/bdd/distances.htm

http://www.cepii.fr/anglaisgraph/bdd/distances.htm

proximity and scientific collaboration: evidence from the global wine industry

Documents