Download - Carayol & Matt 2004 - Does Research Organization Influence Academic Production

Research Policy 33 (2004) 10811102

Does research organization influence academic production?Laboratory level evidence from a large European university

Nicolas Carayol, Mireille MattBETA, UMR CNRS 7522, Universit Louis Pasteur, 61 Avenue de la Foret-Noire, 67085 Strasbourg, France

Received 31 July 2003; accepted 31 March 2004Available online 30 July 2004

Abstract

The paper analyses scientific research production at the laboratory level. The evidence on which the study is based describesprecisely the research activity over the period 19932000 of more than eighty labs belonging to Louis Pasteur University,a large and well-ranked European research university. The research organization of the labs is analysed by focusing on thecharacteristics of the research personnel in relation with the scores in two outcomes that are publications and patents. Thepaper proposes a five-classes typology of laboratories that highlights different styles of research organization and productivityat the laboratory level. It also studies the determinants of the publication performances of labs. We show how appropriatecombinations of inputs in academic labs may be strongly associated to high publication performances. We find that combiningfull-time researchers and university professors in labs tend to preserve incentives. Highly publishing labs also patent. The sizeof the labs, the individual promotions, and the role of non-permanent researchers and of non-researchers are also underlined. 2004 Elsevier B.V. All rights reserved.

JEL classification: L31; 031; 032; 034; 038

Keywords: Economics of science; Academic research laboratory; Organization; Typology

1. Introduction

The individual researcher is usually the standardlevel for analysing academic research production.Most studies concentrate on the effects of individ-ual determinants of academic productivity such asage (Zuckerman and Merton, 1972; Diamond, 1986;Stephan and Levin, 1997), cohort (Weiss and Lillard,1982), training (Garcia-Romero and Modrego, 2001)and gender (Stephan, 1998). Nevertheless, Stephan(1996) recently highlighted that such studies have a Corresponding author. Tel.: +33 390242103;

fax: +33 3902420710.E-mail address: [email protected] (N. Carayol).

weak ability to explain research productivity, due tothe collective nature of research. Thus, some schol-ars suggest that further investigation of academicresearch production should take into account somecollective level of organization, such as (specificallyin the institutional European context) the laboratorylevel (Dasgupta and David, 1994; Stephan and Levin,1997). Such a consideration is congruent with evenmore recent analysis, evidencing that the quality ofother researchers belonging to the laboratory is a cru-cial variable for explaining individual productivity1

1 These results are grounded in initial evidencing of the Mattheweffect (accumulative advantages in science) by Merton (1968), firstdiscussed in economics by David (1994). Carayol (2003a) intro-

0048-7333/$ see front matter 2004 Elsevier B.V. All rights reserved.doi:10.1016/j.respol.2004.03.004

1082 N. Carayol, M. Matt / Research Policy 33 (2004) 10811102

(Mairesse and Turner, 2002). Nevertheless, it is onething to assess collective effects on individual pro-ductivity and it is another to change the level ofanalysis as the results of Adams and Griliches (1998)demonstrate2. The results at a collective level maybe quite different from the individual researcherespecially due to important externalities among re-searchers within labs: Critical knowledge spillovers,reputation, sharing of equipment instrumentation andfacilities, complementarities between different typesof researchers, or even between different researchagendas.

While most of the literature focuses on the determi-nants of individual productivity or of the productivityof universities or groups of universities, few contri-butions analyse the academic production at the labo-ratory or team level. Such a concern brings into thescope papers more interested in organization and man-agerial issues, which tend to show that the laboratoryis indeed a critical level of organization. Crow andBozeman (1987) show that laboratory funding struc-ture is strongly correlated with the nature of theresearch realized, Joly and Mangematin (1996) il-lustrated various types of contracting and laboratorystrategies and Laredo and Mustar (2000) developed atechnique for stressing laboratories activity profilesaccording to five dimensions (production of certifiedknowledge, education and training, research and in-novation, involvement in the construction of publicgoods, and participation in public debates). Aroraet al. (1998) studying academic research at the lab-oratory (or team) level showed that (even if relyingon cross-section data) there are decreasing returnsof funding on quality adjusted publication whilethe most reputed teams reveal an elasticity of scien-tific performance with respect to funding which ap-proaches unity. Bonaccorsi and Daraio (2002) showedthat the distributions of the average age of scientistsin the labs and the size of the labs are correlated. Inthree out of six domains, they found that the size ofthe labs (i.e. the number of researchers) is negativelycorrelated with productivity.

duced a model of such dynamically biased competition groundedon the difference among the inherent productivity of research po-sitions in different universities.

2 They showed that university departments exhibit decreasingreturns to scale at the university departments level of analysis whilethe whole system of universities exhibits constant returns to scale.

Even if these studies improve considerably our un-derstanding of academic knowledge production, thereare still many unknowns to reach a robust industrialorganization theory of academic research production.Such an aim certainly lies beyond the scope of thispaper, the purpose of which is to improve our under-standing of how laboratories differ in organizationalterms, in outcomes and how organization is related toproductivity. By organization we mean the composi-tion of research laboratories in terms of labour force(full-time researchers, university professors, PhD andpost-doc students, non-researcher personnel) and itsfeatures (size of the team, discipline, etc.). We use bothpublications and patents to characterize the outcomeof research activity.

A first originality of our contribution is to analysevia correlations the structure of laboratories in termsof personnel and outcomes. For instance, we ques-tion whether and how permanent and non-permanentresearchers are connected, how non-researchers areallocated to different labs and the link between dis-ciplines and size. We use simultaneously patents andpublications as outcomes of research (while the lit-erature often focuses on one or the other). We enrichthis analysis by studying the link between patents andthe nature of publications (co-authored with foreigninstitutions or with firms). We also examine how dif-ferent types of personnel are linked to various outputs.A second contribution of this paper is related to theconstruction of a typology of five coherent classes oflabs that underline different types of research orga-nization and scientific production. We show that thecombination of personnel in the lab is correlated to itsperformance. In a third stage, we propose an econo-metric estimation that analyses the determinants ofthe publication performances of labs in order to testand complement our first results.

If an empirical study such as ours is still lacking(to the best of our knowledge), it is probably dueto the unavailability of appropriate data. The consti-tution of an original and unique database allows usto begin to tackle such issues. The data concern theresearch activity of more than 80 labs belonging toLouis Pasteur University (ULP) of Strasbourg. ULPis a large and well-ranked European research univer-sity, counting nearly 2000 permanent researchers andnearly 100 laboratories belonging to various scientificdisciplines.

N. Carayol, M. Matt / Research Policy 33 (2004) 10811102 1083

The remaining part of the paper is organized asfollows. The next section presents a literature surveyon academic research production and gives the readersome examples of the expected added value of the lab-oratory level of analysis. The third section providesinformation on the data and some descriptive statis-tics. In the fourth section, we develop a correlationstudy (inputinput, inputoutput, outputoutput) thatstresses some first features of research organizationand scientific production at the lab level. The fifth sec-tion is dedicated to the typology which synthesisescorrelations in highlighting the several designs of labresearch organization and production. The sixth sec-tion presents a study on the determinants of collectivepublication performance. The last section discussesthe main results obtained and concludes.

2. Literature survey

The objective of this part is not to present a de-tailed review of the broad literature developed by soci-ologists and economists of science. We have selectedamong the various topics specific questions related toour paper. With our data we are able to tackle issuesrelative to the position of the researchers (full-time sci-entists versus teach-and-research positions), the age ofthe research personnel, the prestige and the size of thelab and, we also consider patent and publications astwo possible outputs. First, we underline results com-parable with our findings, even if the levels of analysisor the methodologies differ. Lastly we present someexamples illustrating how the laboratory level of anal-ysis may contribute to the present state of our knowl-edge on academic research production.

2.1. Teaching versus research positions and researchproductivity

The sociology of science questions whether re-search and teaching in academia are complementaryor competitive activities. Some authors consider themas joint activities in the sense that one reinforces theother. Others regard them as conflicting roles withdifferent expectations and obligations (Fox, 1992,p. 293). She used a survey based on a sample of socialscience faculty to analyse how research and teachingactivities influence the publication productivity of

social scientists. The study covered four social sci-ences (economics, political science, psychology andsociology) in BA-, MA-, and PhD-degree grantingdepartments in the US. She showed that faculty mem-bers with high publication productivity exhibit stronginterest in commitment of time to and orientationto research. They are not strongly involved in bothactivities, but favour research activities. Her resultssuggest that the more productive researchers spendfewer hours teaching, preparing for courses and con-sider teaching as less important than research. Herfindings tend to prove that research and teaching areconflicting actions.

2.2. The research context and publicationproductivity

Empirical studies in the USA found that researchersat prestigious university departments are more produc-tive and cited than their colleagues in lower-rankeduniversities (Cole and Cole, 1973). A debate about thecausality between productivity and department pres-tige appears. Two hypotheses emerge (Allison andLong, 1990). Is there an effect of the department on thescientists productivity (departmental effect) or doesproductivity influence the prestige of the position ob-tained (selection effect)? These hypotheses are not in-compatible and could be mutually reinforcing, but aneed to assess their relative importance exists.

Long and McGinnis (1981) based their study on apopulation of biochemists and defined different orga-nizational contexts3: academic versus non-academicsectors and contexts, which do or do not encouragepublication. The probability of being employed in aspecific context is not strongly influenced by the pub-lication productivity or the citations. Once employedin a specific environment, individual productivitysoon conforms to the characteristics of that context.Allison and Long (1990) analysed job changes by sci-entists within the academic system in four disciplines(physics, chemistry, mathematics and biology). Sci-entists moving to more prestigious places increasedtheir rates of publication and of citation; those mov-ing to less prestigious institutions showed substantialdecreases in productivity. All these results tend to

3 Long (1978) showed similar results in biochemistry within theacademic system only.


underline a preponderance of the departmental effectover the selection one.

Mairesse and Turner (2002) analysed the impact oflaboratories characteristics4 on the individual produc-tivity of CNRS physicists. If the stock of publicationsof the lab increased by 10%, the individual researcherof this lab would publish in journals receiving 0.26more citations on average and he would also publish0.6 papers more per year on average. Thus, the perfor-mance of colleagues influences positively individualproductivity.

2.3. Age and research productivity

Some economists focused on the publishing activ-ity of scientists in life-cycle models, trying to ex-plain the link between age and scientific production.Diamond (1986) showed that the publishing activityof mathematicians at Berkeley declines continuouslywith age. Weiss and Lillard (1982) found, for Israeliscientists, that the average annual number of publi-cations first tends to increase, and then to decrease.Levin and Stephan (1991) considered six disciplinesand except for particle physicists, publishing activityfirst increases, reaches a peak around mid-career anddeclines. Mairesse and Turner, 2002 underlined sim-ilar life-cycle effects for a sample of French CNRSphysicists. However, Stephan (1996) emphasizes thatthe explanatory power of these models is rather lowand suggests that if an agepublication relationshipexists, other important factors probably influence pub-lication behaviours.

Bonaccorsi and Daraio (2002) analysed the re-search activity of the Italian National Research Coun-cil (CNR). They obtained a negative relationshipbetween productivity indicators and the average ageof researchers. However, the average age of seniorscientists and research directors is not significantlyrelated to productivity. They refuted the hypothesis ofa life-cycle effect and assumed that the average ageof a laboratory reflects its attractiveness and scientificvitality. They implicitly had in mind the existenceof a virtuous circle: higher prestige induces greater

4 They take into account the size of the laboratory (numberof researchers), the productivity of the lab, the quality of thepublications of the lab (impact factors of the journals) and thedegree of international openness (share of papers written withforeign co-authors).

resource availability for young researcher positionsand thus increases the attractiveness.

2.4. Size and research productivity

At an aggregate level (university or groups of uni-versities), some authors analysed the nature of thereturns of scale for scientific production. One of themain problems of measuring the returns of scientificproduction lies in the possibility to assess all inputsused. In other words, the omission of one or moreinputs could unduly lead to revealing decreasing re-turns. For instance, Adams and Griliches (1996) useddata on 50 universities and five scientific fields. Theyfound diminishing returns to the individual universityR&D, for the number of papers and the total citations.They changed the nature of the input and replacedR&D by scientists and engineers. The coefficients onpapers and citations increased, thus indicating mea-surement problems linked to the choice of input. In alater work (2000), they considered the scientific pro-ductivity of US universities in eight scientific fields.They showed that at an aggregate level, the researchproduction follows constant returns to scale as atthe individual university level, diminishing returnsprevail. These differences are explained by more im-portant measurement errors at the individual leveland by the existence of spillovers, which may onlybe captured by an aggregate analysis.

At the laboratory level, it becomes less relevantto study returns of scale, nevertheless the size of thelab may be a crucial variable to take into account.Bonaccorsi and Daraio (2002) found that the size oflaboratory is never positively correlated with produc-tivity; on the contrary in three domains (chemistry,environment and engineering) size and productivityare negatively linked. In almost all fields the mostproductive labs are the smallest, and the least produc-tive ones may be large or small. Mairesse and Turner(2002) stressed that the influence of the size of thelaboratory on the individual production (in numberand quality) is significant but small. The relation isnegative for the average number of papers and slightlypositive for the citations.

Concerning the production of patents, Wallmark(1998) and Henderson et al. (1998) showed that thelargest research organizations apply for more patents.Payne and Siow (2003) and Foltz et al. (2000) found a


positive and significant effect of federal research fund-ing on patents. Coup (2003) used patent counts andpatent citations as outputs, different R&D lags, typesof universities, technology classes, and time effectsand concluded to the presence of constant or dimin-ishing returns to scale.

2.5. Patent and publication

Stephan et al. (2002) used data from the 1995 Sur-vey of Doctorate Recipients and analysed the patentactivity of a sample of doctoral scientists and engi-neers, focusing on the relationship between patentingand publishing at the individual level both in academiaand in industry. Their main question was whether pub-lications and patents were related activities or not.They first underlined that for the whole sample, lessthan 20% of scientists applied for at least one patentwhile 70% published at least one paper in the periodcovered by the analysis. For academia, these percent-ages are 10% for patenting and 83.3% for publica-tions. The probability for one scientist to apply for atleast one patent is significantly related to whether ornot this scientist has published at least one paper. Thenumber of patent applications is positively and sig-nificantly linked to the number of papers published.Patenting and publishing activities seem thus to bestrongly related at the individual level.

2.6. Expected effects at the laboratory level ofanalysis

Since we focus on the laboratory level of analy-sis, our expectation is to shed new lights on the is-sues mentioned above. Let us briefly present someexamples, which may illustrate some of the expectedadded values the lab level analysis may have in thisrespect.

The question whether teaching and research arecomplements becomes really crucial at the laboratorylevel. In France, a scientist may either be full-time re-searcher or may have to spend a part of their time toteach. To analyse the complementarity of both posi-tions at the lab level, we will have to test if a rightproportion of both positions has a positive impact onproductivity. Full-time researchers may enhance theconnectivity of university professors to frontline re-search. On the other hand, university professors may

improve the productivity of full-time researchers justby driving PhD students to them. This raises new ques-tions in terms of access to non-permanent researchhuman resource (PhDs and post-docs) and its impacton productivity, which are ignored by the specializedliterature.

Researchers age is also an interesting example ofa variable that may induce collective effects. In a lab-oratory, seniors or experienced researchers may in-crease the productivity of juniors thanks to collectivework or simply due to informal contacts. Conversely,junior researchers may stimulate the productivity ofolder ones, known to have fewer incentives in theirlate careers. Thus the question does not any more con-cern the productivity trajectory over the life cycle butconcerns gathering researchers of different ages whomay be in some respects complementary.

Another issue concerns the relation between the out-puts of the research production process. Do patentsand publications correlate or exclude each other atthe laboratory level? Here again the laboratory levelhas an added value. The question is no more relatedto the allocation of the time of one scholar betweenresearch projects leading to patents and more funda-mental ones. It becomes a question of allocation of re-sources within the labs: scholars may be specialized ondifferent but complementary research agendas, whichmay or may not have a potential for patenting; scholarsmay also benefit from the presence of non-permanentresearchers especially allocated to patenting activities.It may also be an issue of managing research agendashaving different time lines.5

3. The data

The objective of this part is to present the generalcontext of our statistical analysis. We will briefly de-scribe the different research activities of Louis PasteurUniversity and its scientific reputation. Then, we willexplain the way we collected the data, present theircharacteristics and richness and provide some resultsof the distribution of labs.

5 For some empirical evidence on the management of researchagendas in laboratories and how this influences their strategiesfor collaboration with firms one may see Carayol (2003b) andreferences therein.


3.1. Louis Pasteur University

The data concern the research activity of one sin-gle university, namely Louis Pasteur University (ULP)of Strasbourg. This university is quite large and di-versified. Seventeen separate institutional components(i.e. engineering schools, teaching and research units,and various institutes) are located in six campuses inthe Strasbourg area in which around 18,000 studentsare enrolled. Research and teaching activities cover awide range of subjects in the fields of medical sciencesas well as mathematics, computer science, physics,chemistry, life sciences, geology, geophysics, astron-omy, engineering sciences. Human and social sciencesare also present in the specific fields of economics,management, geography, psychology and educationalsciences.

ULP has an old tradition of fundamental researchand a long-term standing of scientific excellence. Itsresearchers have received numerous national and in-ternational scientific prizes, including the Nobel Prizefor Chemistry awarded to Jean-Marie Lehn in 1987for his contribution to the field of Supra-molecularchemistry.6 Overall, ULP is one of the largest Frenchuniversities in terms of research. The Third EuropeanReport on Science and Technology Indicators 2003ranks it first among the French universities in terms ofImpact and 11th among European universities. Suchresearch capacity is enhanced by a close-knit with themajor national research bodies such as the NationalCentre for Scientific Research (CNRS) and the Na-tional Institute for Health and Medical Research (IN-SERM) present in the Strasbourg area.

3.2. The variables and some descriptive statistics

The presentation of the variables depends mainlyon the method used to collect them. We collected datarelated to the personnel and the laboratories from in-ternal administrative sources. Information related topublications came from the ISI Web of Science, and

6 Also Ferdinand Braun was awarded the Nobel prize in 1909in physics for his radio telephone (together with Marconi). OtherNobel price or Field Medal winners spent some time in their careersat Strasbourg University: Wilhem Roentgen, Alphonse Laveran,Hermann Staudinger, Andre Lwoff, Louis Neel, Louis Pasteur,Rene Thom.

the French Patent Office provided patent data. We willpresent the characteristics of our variables and the waylabs are distributed according to these variables.

3.2.1. Laboratories and personnelWe collected the variables from administrative re-

ports completed for the 1996 contractual affiliationround. Such a round occurs every four years. Alllaboratories (and also faculties and institutes) have toproduce a standardized document, which is usuallydivided into two distinct parts: (i) a prcis of thepast four years and, (ii) a project for the next fouryears. Thus the data concern the period from 1993to 2000, which may be separated into two four-yearsub-periods: 19931996 and 19972000. These doc-uments are evaluated through standard peer reviewprocedures conducted by both the Ministry of Re-search and Education and funding agencies such as theCNRS and INSERM whose support is expected. Werecorded their decisions concerning the affiliation. Theaffiliation to CNRS and INSERM means increasedfunding for research facilities and positions. The affil-iation operates through a peer review process mostlytaking into account scientific production argumentsand constitutes clearly a signal of the labs scientificexcellence.

We collected the standard information mentionedin those documents. We gathered many typical infor-mation about the personnel of the labs, specifying thenumber of individuals in each detailed category of per-sonnel and individual information on permanent re-searchers including the name, the sex, the age and thestatus.

Concerning the status, in France permanent re-searchers may occupy two types of positions: eithera university professor type position (with teachingand research) or a full-time research position. Theformer belong to the university while the latter areemployed directly by the large national public re-search organizations such as the CNRS or INSERM.Nevertheless both categories work together in theuniversity labs.7 In addition, for both categories,there is a clear promotion (from Assistant Professorto Full Professor; or from Researchers to Director

7 These institutions also have some labs independent of theuniversities, but these are not to be found in our sample, and sowe do not consider them.


of Research) at mid-career on the basis of scientificproductivity. Such a promotion does not imply tenuresince, in France, Assistant Professors and Researchersare tenured from the very beginning of their careers.Nevertheless it implies a significant increase in wagesand social status within the academic sphere.

We recorded 83 distinct laboratories in 1996. Wehave reliable and complete information for all but twoof them about for which we miss the complete char-acterization of their permanent researchers. Thus forsuch variables as grade, age, sex of permanent re-searchers, a sub-population of 81 labs will be consid-ered. Among the 83 labs, 43 are funded by the CNRS,8nine by INSERM, while 31 are supported by the Min-istry of Research and by the resources of ULP. Amongthe 1460 permanent researchers, 760 are full-time re-searchers directly paid by the CNRS and INSERMand 700 are university scholars. On average, the per-manent researchers are 51.5 years old. Among them,360 are females (24.6%). Fifty-seven percent of per-manent researchers occupy junior-like positions. Wealso find some 1940 non-permanent researchers: 1230PhD students and 710 post-docs. Lastly some 1120non-researchers (administrative staff and technicians)and 410 visitors are recorded.

ULP laboratories are rather small in terms of per-manent researchers: 61 labs, i.e. 74%, have less than20 permanent researchers and eight declare more than41 scientists. The distribution for non-permanent re-searchers exhibits the same characteristics: 70% oflabs have less than 20 non-permanent researchers and21% more than 41. For a large majority of labs (53,i.e. 65%), more than 50% of permanent researchersoccupy junior-like positions. The average age of per-manent scientists is above 50 in 70% of the labs, fivelabs have an average age between 40 and 45. Womenrepresent less than 25% of the permanent researchersin 53 laboratories (65%).

We also got the main scientific disciplines of thelabs according to the specific categories defined by theUniversity. These are the following: astronomy (1 lab),biotechnology (12 labs), chemistry (9 labs), genetics

8 Two types of association with the CNRS exist: UMR (UniteMixte de Recherche) and UPR (Unite Propre de Recherche). Thelatter is more closely supported by the CNRS. Nevertheless, theselabs may be supported by the university and/or may host universityresearchers.

and cellular and molecular biology (9 labs), geogra-phy (2 labs), mathematics (1 lab), mechanics (1 lab),medicine (22 labs), odontology (1 lab), neurosciences(7 labs), condensed matter physics and chemistry (4labs), subatomic physics (2 labs), earth sciences (2labs), information sciences and technology (3 labs),humanities and social sciences (8 labs).

To complement such information, we used datawe had on permanent researchers in order to col-lect information on their disciplinary affiliation atthe most detailed level possible as indicated by theinstitutions to which they are affiliated (UniversityNational Council, CNRS, INSERM). Such classifi-cations do not perfectly match at the sub-disciplinelevel we are interested in. Nevertheless, thanks to anormalization grid produced by the OST (Observa-toire des Sciences et Techniques) specifically for theFrench system, we were able to allocate nearly allpermanent researchers to 50 different sub-disciplinesaccording to a unique nomenclature selected as thereference (namely the one of the National UniversityCouncil) (sub-discipline level normalized through theOST nomenclature; cf. OST, 2003). We found that11 laboratories are mono-disciplinary, 15 have twodisciplines, and 21 have three disciplines. The modeof the distribution is four disciplines in 22 labs. Themaximum number of disciplines is eight and only twolaboratories are concerned.

3.2.2. The outcomesOur database also integrated information about

the various outcomes of research production, namelypublications and patents. For each permanent re-searcher we collected his/her published articles (usingSCI, SSCI and Arts and Humanities ISI databases).We found more than 26,000 occurrences over the19932000 period. This amount includes some dou-ble counting as quite a few ULP researchers mayhave co-authored papers. By dividing each occurrenceby the number of co-authors we obtain the effective(normalized) scientific contribution of each author.9The total scientific performance is 6040. The mediannumber of co-authors is five. We differentiate be-tween two types of co-authorship. A co-publicationis international if at least one co-author belongs to

9 Since publication data were collected from our list of perma-nents, an author is necessarily a permanent researcher.


a non-French institution. Some 10,400 occurrences(i.e. 40%) exhibit international co-authorship. Wealso screened co-publications in which at least one ofthe co-authors belongs to a firm. One thousand andtwo hundred publication occurrences with industrialpartners exist and represent 4.6% of all publicationoccurrences.

Concerning the total scientific performance, 67 lab-oratories have published between 1 and 100, the dis-tribution between these labs is heterogeneous and themode is 2040 publications in 18 laboratories. Threelabs have published between 300 and 500 papers. Thedistribution of the average performance by permanentresearchers is uni-modal. The mode of the distribu-tion is an average of 24 articles per permanent re-searchers in 27 laboratories. Twelve laboratories showan average of more than eight papers. These resultsshow that a small number of labs publishes a largeamount of papers and that a large number of labspublish a small number of articles in a quite similarfashion as at the individual level as first evidenced byLotka (1926).

The behaviour in terms of co-publication is rathercontrasted: in five laboratories the average number ofco-authors is between 1 and 2 and in five other lab-oratories the average is above 8. The mode is 56co-authors on average for 30 laboratories. The dis-tribution of international co-authorship is heteroge-neous. In nine laboratories, more than 50% of thepublications are international. In 62 laboratories, lessthan 37.5% of the papers signal a foreign co-author.The mode is between 25 and 37.5% of internationalpublications in 28 laboratories. Ten laboratories neverpublished with industry and at the other extreme onelaboratory published between 25 and 30% of its pa-pers with industrial partners. The mode is between0.1 and 5% of publications with private partners in 38laboratories.

We also looked for the French and European patents,which had been invented by at least one of the ULPpermanent researchers. To do so we matched our per-manent researchers table with the patent data pro-vided by the French Patent Office (INPI) covering thesame 19932000 period. We found 850 occurrencesof French or European patent applications. After acleaning process, which eliminated the extensions ofFrench to European patents, we ended up with 463patents invented by researchers from ULP over the 8

years considered. Some 189 patent applications wererecorded in the first sub-period (19931996) and 274in the second one (19972000) giving an increase rateof 45% between the two sub-periods.

The distribution of patent applications over the labsis skew. Thirty-eight laboratories have no patent, 31laboratories have between 1 and 10 patents, five lab-oratories have between 11 and 20 patents and onelab obtained between 51 and 60 patents. The distri-bution of labs according to the number of patents perpermanent researcher also shows a decreasing shape.In 26 laboratories, the number of patents per per-manent researcher is between 0.01 and 0.5, and ineight laboratories it is between 0.5 and 1. Finally,two labs have between 2 and 4 patents per permanentresearcher.

4. A first analysis of the structure of the labs

The objective of this section is to provide a firstanalysis of the structure of the laboratories in order toinfer some intuitions to build our typology. To reachthis aim we conduct correlation studies.10 The first setof correlations describes the common and potentialdistinctive features of the organization of laboratoriesin terms of types, number and disciplines of personnel.The second group of correlations provides informationabout how publications and patents are linked to eachtype of personnel. Finally, we also study outputoutputcorrelations in order to have a first investigation aboutwhether the labs, which publish more, are also the onesthat patent more and if the nature of the publicationscounts. Distinguishing between sub-periods may helpus track causal relations.

4.1. The organization of labs in terms of personnel

We mainly use the various characteristics of the per-sonnel to analyse the organization of research labo-ratories. We provide results about the way permanentand non-permanent researchers are linked, about theallocation of non-researchers and about the link be-tween size and disciplines.

10 All correlation coefficients reported in the text have a signifi-cance level exceeding 95%. A coefficient higher (lower) or equalto 0.3 (0.3) reaches the 99% significance level.


4.1.1. The specific attraction power of full-timescientists and university professors

University professors are usually associated withfull-time researchers (0.35) within labs. Full-time re-searchers are personnel belonging to the public re-search organizations such as CNRS or INSERM. Thisresult is not surprising as university labs affiliated tothese organizations usually involve both types of per-sonnel, but in various proportions.

We also find that the number of permanent re-searchers and the total non-permanent researchers aresignificantly correlated (0.66). At a more disaggre-gated level, we notice that PhD students are primarilycorrelated with professors (0.67). Nevertheless, theyare also strongly correlated with CNRS researchers(0.60). This result does not mean that full-time re-searchers often supervise PhDs. It could be due tothe fact that CNRS researchers are mainly located inthe most recognized labs, which may provide moregrants, thus attracting more PhDs. That observationseems to be supported by the fact that PhDs are neg-atively correlated with EA (Associated Teams) labo-ratories (0.32). The latter are only supported by theMinistry of Research and involve only a few full-timepermanent researchers. Thus, PhD students seem tobe primarily allocated to laboratories in which uni-versity professors are present. This may be explainedby the importance of personal contacts during thelate stage of their graduate studies. Excellence alsomatters for the matching process of students to labs,but appears to be secondary.

On the contrary, post-docs seem to value only fameand excellence when choosing labs, especially foreignpost-docs who are correlated with CNRS researchers(0.46) and high institutional recognition (0.23), es-pecially in UPR-type of CNRS association (0.32).They are not correlated with university professorswhile at least a small but significant correlation wasexpected.

4.1.2. The non-research personnel are allocatedaccording to disciplinary practical needs

The non-researcher personnel are strongly cor-related with CNRS scientists (0.70) and withnon-permanent researchers (0.72). Their allocationseems to be also very dependent on the disciplinaryenvironment. Non-researchers are correlated withsubatomic physics (0.47) and genetics and cellular

and molecular biology (0.23), which require (espe-cially the former) heavy instrumentation and manytechnicians and engineers. One may observe thatthe presence of non-researchers is not significantlycorrelated with any variable characterizing the insti-tutional recognition of the labs. These observationstend to support the hypothesis that non-researchersare mainly allocated for practical needs (determinedby disciplinary environments) and through a pure lin-ear scale fashion not being driven by excellence orreputation considerations.

4.1.3. Labs of different disciplines have different sizesWhen looking at the disciplinary differences, we

found that the laboratories in medicine tend to be of asmaller size. These labs seem also to have a rather lowinstitutional recognition. This may reflect a low qual-ity of research but also explain their ability to get sub-stitute funding from their simultaneous commitmentwithin the University Hospital. In that respect, thelaboratories specialized in the promising and rapidlygrowing field of neurosciences are exactly the oppo-site: funding agencies support them.

The labs in genetics and cellular and molecular biol-ogy, which are also of a big size especially consideringthe number of post-docs, appear to be strongly sup-ported by funding agencies. Subatomic physics labsare also of a large size, but only in terms of CNRS per-manent scientists and non-researchers (technicians),but not in terms of non-permanent researchers (PhDsor post-docs).

4.2. Different positions lead to various scientificperformances

The aim of this part is to study the connectionsbetween the various outcomes of research and thecharacteristics of the personnel. We examine re-spectively the contribution of university professors,full-time scientists and non-permanent researchers.By contribution, we mean publications in general, butalso those co-authored with researchers belonging toforeign institutions or to private firms and patents.

4.2.1. The unequal contributions of universityprofessors

One main and probably not-so-surprising observa-tion concerns the productivity of university professors.


The correlation between publication performance perpermanent researcher and university professor is nega-tive (0.30). Moreover, university professors are neg-atively correlated with the share of authors amongpermanent researchers (0.29) while there is no sig-nificant correlation between university professors andboth the variance and the kurtosis of performanceamong authors. Thus apparently, professors who pub-lish may be quite as productive as CNRS researcherswhile other professors totally gave up publishing. Thisdifference may be explained by the fact that some fac-ulties favour research activities and others dedicatemore efforts toward administration or teaching activi-ties. This result could thus be justified by the idea de-veloped by Fox (1992) that teaching and research maybe considered as conflicting activities (cf. Section 2.1).

4.2.2. The weak performances of junior positionsPublication performance is also strongly negatively

correlated with junior positions (0.42). This resultcould be explained by a lack of experience of youngerresearchers. Nevertheless, scientific performance ap-pears not to be significantly correlated with age. Sev-eral complementary explanations may be provided.Some junior scientists may probably be un-promotedresearchers with lower research abilities or publicationincentives. Or, this result may be partially influencedby the fact that a high share of junior researchers oc-cupies Assistant Professor positions, combining teach-ing, research and administrative activities.

The share of publications written with at least oneindustrial partner is negatively correlated with bothuniversity professors (0.30) and junior-like scientists(0.26). They both seem to be more oriented towardthe scientific community. As both categories are alsothe least productive ones, they may not be attractive inthe eyes of potential industrial partners. This observa-tion is congruent with the fact that the kurtosis of thedistribution of industrial collaborations among authorsis correlated with university professors (0.78). The lat-ter is also true concerning PhD candidates (0.57), whomay fuel such collaborations.

4.2.3. Full-time scientists publish and patentFull-time researchers are strongly correlated with

the number of publication occurrences and with thepublication performance. CNRS researchers havestrong connections with the international scientific

community: their correlation with the share of inter-national collaborations among all publication occur-rences is 0.34. A scale issue seems to appear sincethis share is correlated with the number of permanentresearchers (0.31) and the number of non-researchers(0.31). Nevertheless, the latter result may be due to anindirect effect since there are more non-researchersin labs belonging to disciplines such as subatomicphysics in which publications count impressive num-bers of co-authors. The variance of international col-laborations among authors is strongly correlated withboth CNRS researchers and non-researchers. The kur-tosis of the distribution of international collaborationsamong authors is equally correlated with all size vari-ables: this seems to indicate that when size increases,the long-range collaborations are more concentratedon a lower share of permanent researchers.

CNRS researchers are the most important contrib-utors (0.55) to patent production, while universityprofessors are not significantly correlated with patentproduction. This may be due to historically groundeddifferences, which may be progressively reduced:university professors are positively correlated with anincrease in patenting (0.24) and with a move fromnon-patenting to patenting (0.25) between the twosub-periods. A possible scale effect may exist, sincewe observe that CNRS researchers and PhDs are alsocorrelated with the increase rate in patenting activity(0.49 and 0.40). Nevertheless this explanation may becontroversial and the results may primarily express aprogressive popularity of the patenting activity amongthe scientific community.

4.2.4. Non-permanent researchers are importantactors in the patenting process

Finally, the distribution of publication performanceamong authors within labs may be affected by thepresence of non-permanent researchers: the varianceof publication performance is correlated with for-eign post-docs (0.33). Moreover PhDs and foreignpost-docs are strongly correlated with the kurtosis ofscientific performance distributions among authors(0.37 and 0.47). These results seem to indicate thatthe hosting of non-permanent researchers benefitsunequally to the permanent researchers of the lab.Their efforts are probably allocated to the benefitof the most famous scientists or, in other words,post-docs are strongly attracted by labs with a few


highly productive researchers. These explanationscould in some ways be supported by observing thatthere is a high correlation between the kurtosis ofthe number of co-authors and the PhDs and foreignpost-docs (0.41 and 0.44), indicating that the morenumerous the non-permanent researchers the skewerthe distribution of co-authors among publications.

The non-permanent researchers seem to be evenmore important for patenting than the permanentresearchers (high correlation coefficient especiallyfor post-docs 0.69). The French post-docs are cor-related with the patenting productivity of permanentresearchers (0.24). Permanent scientists may probablyassign more applied (even if potentially less aca-demically rewarding) problems to French post-docs.The latter may also be willing to select these moreapplied activities in order to acquire a first practicalexperience valuable on the private job market. Fi-nally non-researchers are also correlated with patents(0.51).

4.3. Correlations between outcomes

We first look at the potential correlations be-tween the publication performance and interna-tional co-publication on the one side, and industrialco-publication, on the other side. Next we turn towardthe potential reinforcement or exclusion betweenpublishing and patenting.

4.3.1. Industrial and international publications arecorrelated

The share of international co-publications amongall publications is not correlated with the publicationperformance per permanent researcher apparently in-dicating that the labs which collaborate more with in-ternational partners are not the ones that publish moreon a national basis. Nevertheless the share of inter-national publications in the first sub-period is corre-lated with the publication performance (0.26) indicat-ing probable scale or threshold effects in publicationperformance of the laboratory for getting access to in-ternational publication networks.

Industrial collaborations per permanent researcherare strongly correlated with publication performanceper permanent researcher (0.52) indicating that thelabs, which publish more, have also more research col-laborations with firms. Moreover, it appears that the

labs in which permanent researchers are the most con-nected to the international scientific community arealso the ones that collaborate more with firms. Thatmay express the fact that internationally visible scien-tific labs tend to attract more firms.

4.3.2. Publications and patents are related activitiesThe hypothesis that pure scientific productivity is

related to invention appears to be strongly supportedby the data against the exclusion hypothesis: thepublication performance per permanent researcher isstrongly correlated with the patents per permanentresearcher (0.33). The result is also true when lookingat the correlation with patents per researcher, even ifless strongly (0.24). Since the increase in patentingis strongly correlated with the whole publication per-formance (0.37) there may be a scale effect here. Ourresults seem thus to confirm the findings of Stephanet al. (2002) (cf. Section 2.5).

Patents per permanent researcher and patents perresearcher are even more strongly correlated with col-laborations with industry (0.44 and 0.37). The causal-ity seems to go this way since the rate of increase inpatenting between the two sub-periods is strongly cor-related with the collaborations with industry (0.49).The intensity of publication collaborations with indus-try seems to be important for the intensity in patent-ing since the share of industrial collaborations amongall publication occurrences of the lab are positivelycorrelated with patents per researcher and permanentresearcher (0.27 and 0.29).

Patents are also related to international co-publi-cation: the rates of increase in patenting and interna-tional publication occurrences are correlated (0.29).Nevertheless the size issue may generate such re-sults. The important observation is that the numberof international collaboration occurrences is the onlyvariable significantly correlated with the zero topositive patenting dummy (0.29). Such a result tendsto indicate, in the second sub-period considered, thatthe most internationally connected labs may haveshifted from a non-patenting and exclusive publishingbehaviour to a patenting behaviour.

To support that observation and more generally,the recent patents seem to be more strongly corre-lated with publication performance than the ones ofthe first sub-period: patents in period 19931996 arecorrelated with publication performance of the two


Table 1The qualitative variables used for building the typology

Variable Description Modalities

Univ-prof Number of university professors High, medium, lowFull-time-res Number of full-time researchers High, medium, lowPhD Number of PhD students High, medium, lowPost-doc-f Number of foreign post-docs High, medium, lowPost-doc-n Number of national post-docs High, medium, lowJunior-pos Share of junior positions (non-promoted) among permanent researchers High, medium, lowAge-perm Average age of permanent researchers High, medium, lowNon-res Number of non-researchers High, medium, lowDisc-entropy Sub-disciplinary entropy within the lab High, medium, lowAuthors-perm Share of authors among permanents High, medium, lowPerf-perm Average publication performance normalized of permanent researchers High, medium, lowVar-perf Variance in publication performance (norm) among permanent researchers High, medium, lowPatent-perm Patent invention performance per permanent researchers High, medium, lowInstit-recogn Institutional support by national agencies as CNRS or INSERM High, lowPub-internat Share of international collaboration among publication occurrences High, lowPub-indus Share of international collaboration among publication occurrence High, low

sub-periods with coefficient 0.55 and 0.54; whilepatents of sub-period 19972000 have the followingcorrelation coefficients with the publication perfor-mances: 0.65 and 0.64. This may indicate that patentstend to be increasingly grounded in publication per-formance without any feedback decrease publication.

5. The typology of laboratories

As we have seen earlier, the labs differ according toboth their input and their output structures. Moreoverwe have obtained many insights on how the variablescorrelate. We are now turning toward a deeper analy-sis that should allow us to identify various styles ofresearch production at the lab level. There are ob-viously different solutions to associate inputs in labsthat generate different amounts of the two types ofoutcomes. Since our typology is intended to mix bothconsiderations, the variables retained for the MCA arerelated to both. The variables used for building thetypology are presented in Table 1. All variables havebeen transformed into qualitative variables having twoor three modalities. Information on the variables se-lected are presented in Table 1. The typology is builtfor the 81 labs about which we have complete infor-mation on the above-mentioned variables of interest.

To build the typology of laboratories we follow astandard methodology composed of a multi-corres-

pondence analysis (MCA) followed by an ascendanthierarchical classification (AHC).11 Following theMCA, we selected four axes, which collectively rep-resented 39.13% of the inertia. Then, the AHC wasrealized based on the co-ordinates of the labs onthese four axes. Five coherent classes of labs wereselected because, in retaining these five types, thewithin-classes variance of the total variance wasnearly 44.7%, which is usually admitted to be a goodratio. Then the centres of the classes obtained may berepresented on the four axes space (cf. Figs. 1 and 2).

As Fig. 1 shows, the first axis (contributing for14.1% of the inertia) accounts for the size of labsand other variables, which are structurally associatedwith scale issues such as institutional recognition andshare of international collaborations among publica-tions which have been highlighted to be subject toscale effects. On this axis, the fifth class is mainly op-posed mainly to class 2 and to the other classes to alesser extent. The second axis (10.3% of the inertia)opposes strongly research-intensive labs with the onesalso concerned with teaching and PhD supervision.Along this axis class 4 (and to a lesser extent class 1)is opposed to class 2 (and 5 to a lesser extent). Fig. 2gives the projections of the variables modalities andthe classes on the last two axes. The third one (8% of

11 The reader might refer to Greenacre (1993) or Benzecri (1992)for a precise description of the methodology.


Fig. 1. The centers of the classes and the variables contributing to the fist and second axes of the MCA.

the inertia) opposes highly publishing labs to the ones,which are more median in that respect. According tothis axis, classes 4 and 3 are especially distant. Thelast axis retained (6.5% of the inertia) illustrates thedistinction between younger, more international moreattractive toward post-docs and more open toward in-dustry to older, less attractive and less connected withfirms. This is why we synthetically labeled this axisopenness intensity.

5.1. The classes of laboratories

In this part, we discuss the characteristics of the fiveclasses. Some of our arguments are based on descrip-tive statistics, which are not presented due to spaceconstraints but can be produced upon request. In a sec-ond subsection, we discuss and test what we can leanfrom comparing classes.

5.1.1. Class 1: the standard research intensive labsClass 1 involves 22 labs. The laboratories of this

class are of a rather small size and research intensive

since they count on average a high share of full-timeresearchers as compared to university professors.They host only few PhDs and few post-docs, which isquite surprising considering their research-intensivenature. The ratio of permanent researchers over all re-searchers is high (50% on average). The simultaneouspresence of many non-researchers reflects a proba-ble substitution between non-permanent researchersand non-researchers. While the labs have on averagefewer sub-disciplines than other labs (probably dueto their small size), their entropy index is the highest.The average age of permanent researchers is higherthan average. Moreover, the share of junior-positionslies below average, which may indicate on averagelower abilities of these permanent researchers.

The scientific performance of these labs is higherthan that of the average labs, with 5.96 papers written(in individual contribution) per permanent researcher.The papers are written on a less international basiseven if permanent researchers have written as manyinternationally co-authored papers as the average.The collaboration with industry is below the average.


Fig. 2. The centers of the classes and the variables contributing to the third and fourth axes of the MCA.

This may help explain that permanent researchersproduce only 0.12 patent, (one-third of the average).Sixty-three percent of the labs never patented overthe period. The labs belong mainly to the medicinefield, most of the labs are involved in neurosciences,one in chemistry and one in bio-pharmacy. The labsof this class are nearly equally shared between highand low recognition which implies that this classcounts a third of the less recognized labs and half ofthe labs supported by INSERM. This observation ledus to compare the characteristics of the highly recog-nized ones with the other ones. The two noticeabledifferences were that the recognized ones count morefull-time researchers, attract more non-permanentresearchers and tend to publish substantially more.Nevertheless the two sub-groups are quite similar inall other respects, especially concerning the ratherlow inventing activity.

5.1.2. Class 2: the teaching oriented labs in thefields of social and human sciences

This class gathers 10 labs, among which we findeight belonging to the social sciences, humanities and

geography (the remaining two belong to bio-pharmacyand biology). Their size is below average. Permanentresearchers are mainly university professors (the shareof full-time researchers is close to zero). They countmany PhD students and nearly no post-docs. The shareof females among permanent researchers is higher thanthe average (28% against 23%).

Eight of the labs are not supported by the fundingagencies. The share of authors among permanent re-searchers is extremely low: only 45% of permanentresearchers publish. Considering these authors, theiraverage performance is only one-third of the aver-age performance of authors in labs. Only few of thepublications are written through international cooper-ation and the collaborations with co-authors outsideacademe represent only 1.6% of the publication oc-currences.

5.1.3. Class 3: the non-research intensive andindustry oriented labs

Class 3 counts 15 labs with a size below aver-age. They are non-research intensive as we findmainly professors and only few full-time researchers.


This explains the presence of many PhDs and fewpost-docs. Moreover, the share of permanent re-searchers is higher than average (54% against 47%).Permanent researchers are much younger than theaverage, and the junior-like positions represent 70%of the permanent positions.

The ratio of authors among permanent researchersand the publication performance of authors are be-low average. They publish 1.63 papers less than in theaverage labs. Nevertheless, the proportion of interna-tional collaborations is higher than average as well asthe share of collaborations with industry (8.7% of thepublications of the labs). But this did not favour verymuch patenting activity since their average inventiveperformance is still below the general average (only0.3 patent per permanent researcher). These specifici-ties are probably not related to their disciplinary pe-culiarities. The labs belong to various disciplines: fivelabs are active in bio-pharmacy, two each in biology,physics-related fields, engineering and medicine, andone each in chemistry and neurobiology. Nearly halfof the labs are supported by the CNRS.

5.1.4. Class 4: the elite research intensive labsThe 12 labs of class 4 are of a small size even if

slightly bigger than the ones of class 1 and smaller thanthe ones of class 3. The share of full-time researchersamong permanent researchers is slightly below aver-age. The share of juniors and the average age are bothbelow the average of all labs. This tends to indicatea quite good individual recognition of the permanentresearchers. The ratio of permanent researchers is thelowest of all classes (only 38%). This comes from nu-merous PhD students (nearly as many as in class 3but with much fewer professors), twice as many na-tional post-docs and more than three times as manyforeign post-docs than in class 1. With the same num-ber of sub-disciplines as average, the sub-disciplinaryentropy is the highest (0.9) but never very high (max-imum at 1.3).

These labs exhibit a strong cohesion. Only 2% ofthe permanent researchers do not publish at all. Perma-nent researchers publish on average 2.8 papers morethan the average, and authors publish 1.6 papers morethan the authors of class 1 (the other research inten-sive class). While the share of international collabora-tions among publication occurrences is below average,the number of international collaborations is clearly

the highest compared to all other classes. With 8.7%of the publication occurrences being co-authored withindustrial partners, the number of industrial collabo-rations per permanent researchers is more than twicethe average. This probably contributes to explainingthat most labs of this class patent (85%) and that theypatent nearly twice as much as the average (10 patentsinvented) and nearly three times as much per perma-nent researcher (more than one patent per permanentresearcher). These labs belong to different disciplines:five belong to medicine, four to bio-pharmacy, two tochemistry, and one to biology.

5.1.5. Class 5: the large laboratoriesThis class covers 22 labs, which may be distin-

guished by their large size. They have 37 permanentresearchers on average. Most of them benefit from animportant support from CNRS, which allows them toaccommodate more than 20 full-time researchers onaverage. This is the highest ratio of full-time scientistsover permanent researchers among all classes. The ra-tio of permanent researchers is quite similar to theaverage, but these labs have fewer PhD students andessentially more foreign post-docs. They also countmany non-researchers. Age, share of junior-like posi-tions, share of females, entropy are similar to average.

One should notice that neither the size nor the im-portant ratio of full-time scientists allowed these labsto be more productive: their patenting and publicationperformances per permanent researcher are similar tothe average (even if the patenting activity is increas-ing, 20% of the labs do not patent at all). The shareof international collaborations is higher than averagebut that may be due to a higher number of co-authors.The share of collaborations with industry amongpublication occurrences is below average. These labsbelong to a very large range of disciplines: five inchemistry, five in biology, four in physics-relatedfields, two in engineering, two in earth sciences, oneeach in bio-pharmacy, astronomy, mathematics andneurosciences.

5.2. Do organizational aspects really explainscientific performances of labs?

The typology helps understanding the relations ex-isting between a group of input variables and a set ofoutcomes. One important conclusion of the typology


exercise may consist in proving that the way person-nel are associated within labs has an impact on thecollective productivity. We highlight the existence ofa virtuous combination of factors leading to both highpublishing and high patenting performances.

Indeed, some labs (class 4) are by far the most pro-ductive in terms of publication performance and (evenmore) in terms of patents. These labs exhibit an inter-esting combination of inputs: permanent researchersare specialized in different sub-disciplines (highest in-terdisciplinary entropy), are nearly equally split be-tween professors and full-time researchers (with anslightly higher share of teach-and-research positions)and, attract many non-permanent researchers. If wecompare these labs with those of the classes 1 and3,12 we shall see that although they have quite compa-rable sizes, they exhibit different input structures in-ducing different output performances. Class 3 is moreoriented toward teaching and PhD supervision whilehaving few post-docs and a below average publicationperformance. Class 1 is the more research orientedwith more full-time researchers than university pro-fessors among permanent researchers. These labs pub-lish significantly more than the average. Nevertheless,they publish even less than the labs of class 4 and theirpatent performance is only one-third of the average.

However, there may be main important potentialshortcomings that might alter these conclusions. Theseare the following.

(i) Classes and disciplinary differences: Even if theinformation on lab disciplines has not been usedas such to build the typology, there are strong ex-pectations that classes and disciplines are not in-dependent. As a matter of fact we have observedthat class 2 groups together all labs from socialsciences. The question is to what extend are dis-ciplines and classes related. At this stage we stilldo not know whether belonging to a given classdoes or does not only translate the specific orga-nization and performances of a given discipline.

(ii) Internal coherence of classes according to out-comes scores: The (average) scores in outcomesof labs that we have used to compare classes

12 Since classes 2 and 5 tend to be very specific (class 2 bringstogether social sciences and class 5 assembles big labs) we leavethem aside to concentrate our comparisons between classes 1, 3and 4.

(especially class 1 versus class 4) might hidean important heterogeneity with respect to suchvariables. The within-classes coherence that isensured by the procedure used to build the typol-ogy might be essentially effective on the othervariables used (listed in Table 1) so that the labsof a given class cannot be clearly associated tothe mean values of outcomes.

(iii) Disciplines and average outcome scores ofclasses: Even if the risks contained in (i) and (ii)are at least partially avoided, it remains possiblethat the differences in average outcomes of labsbetween classes may only be due to the relativerepresentation of some disciplines. If so, thereis no valid statement on the relation between re-search organization and production performancethat can be derived from comparing averageperformances of labs in different classes.

The evidence presented in Tables 26 allows us toaccount for these three potential biases and thus toconfirm our statement on the relation between researchorganization and production performance.

Table 2 shows that as expected, the disciplines arenot uniformly present in classes. Eighty percentage oflabs in class 2 belong to social and human sciences,labs in class 5 are predominantly in the fields of bi-ology, chemistry and physics, the ones of class 3 arequite evenly distributed among disciplines (excludingsocial and human sciences and mathematics). Classes1 and 4 are exclusively concerned with biology, phar-macy, medicine and chemistry. Nevertheless, 68% ofthe labs in class 1 are in the field of medicine whileonly 42% of class 4 are in this field. From these obser-vations, we can conclude that if classes and disciplinesare far from being independent, one cannot state thatclasses only reflect disciplinary difference in knowl-edge production. If we consider specifically classes 1and 4 that have been most compared: both have labsof the same set of disciplines but with different sharesof each discipline.

Tables 3 and 4 show how the labs in classes areallocated with respect to performances in publicationand patents per permanent researchers. We observethat performance scores within classes are not homo-geneous. Some classes such as classes 5 and 1 ex-hibit high internal heterogeneity both for publicationsand patents. Nevertheless, the differences observed


Table 2The disciplinary association of classes labs (cross-table disciplines/classes)

Class 1 (%) Class 2 (%) Class 3 (%) Class 4 (%) Class 5 (%) TotalBIO 23 10 20 8 27 16PHARMA 5 10 33 33 5 12MED 68 0 13 42 0 22CHEM 5 0 7 17 23 9PHYS 0 0 13 0 32 9ENG 0 0 13 0 9 4MATH 0 0 0 0 5 1SHS 0 80 0 0 0 8

Total 100 100 100 100 100 81

Table 3The publication (per permanent researcher) performance of labsin classes: cross-table classes/deciles of perf-perm

Class 1(%)

Class 2(%)

Class 3(%)

Class 4(%)

Class 5(%)

1 0 70 13 0 02 5 20 20 0 93 0 10 20 0 184 14 0 13 0 145 18 0 13 8 56 9 0 13 8 147 14 0 7 17 98 9 0 0 17 189 14 0 0 33 5

10 18 0 0 17 9

Total 100 100 100 100 100

between classes 1, 3 and 4 remain valid: the publica-tion performances of class 3 labs are clearly below,class 4 labs overcome the performance of other classeswith regard to the two outcomes scores, patent pro-

Table 4The patent (per permanent researcher) performance of labs inclasses: cross-table classes/deciles of patent-perm (with the fivefirst deciles being joint in one category since a bit more than halflabs do not patent)

Class 1(%)

Class 2(%)

Class 3(%)

Class 4(%)

Class 5(%)

1 64 90 67 17 145 0 0 0 0 146 23 0 0 0 147 5 0 0 8 278 0 0 13 25 149 9 0 7 17 14

10 0 10 13 33 5

Total 100 100 100 100 100

duction counts for the class 1 labs are clearly belowthe other classes ones. Thus the statements on therelative productivities of classes 1, 3 and 4 based onaverage performance scores still hold.

Tables 5 and 6 allow us to compare between labsscores of different classes, by controlling for disci-plinary differences. Let us first focus on the compari-son between labs in classes 1 and 4. It is shown that,for all disciplines but chemistry, labs in class 4 aremore productive than the ones in class 1. For instancebiology labs in class 4 are in average nearly three timesmore productive in terms of publications than the onesof class 1 (12.89 versus 4.34 papers), while pharmacyones are twice as productive (6.91 versus 3.49). Asregards to patent production, the differences betweenlabs of the two classes are even more remarkable. Letus now consider class 3. The publication performanceof these labs controlling for disciplines remains clearlybelow the ones of classes 1 and 4. On the contrary,results in terms of patent production (Table 6) speakfor a careful use of class 4 average patent productionperformance. Indeed, patenting scores for the labs ofthis class are specifically due to labs in pharmacy andengineering sciences while labs in other fields have nopatents.

6. The determinants of publication performance

This section is dedicated to specifically studying thedeterminants of laboratories publication performance.For that purpose we have run an OLS linear regres-sion on perf-perm which is the publication perfor-mance per permanent researcher of the lab. Such studycannot provide the same richness than the typology,


Table 5The publication per permanent researcher performance of labs in classes and disciplines (mean values of perf-perm and standard errors inparentheses)

Class 1 Class 2 Class 3 Class 4 Class 5

BIO 4.341 (2.525) 0.526 (0.526) 3.583 (1.527) 12.892 () 5.012 (1.892)MED 6.585 (3.532) () 3.542 (0.181) 7.348 (3.436) ()PHARMA 3.494 () 2.217 () 2.925 (1.301) 6.914 (1.127) 3.597 ()CHEM 7.349 () () 3.077 () 6.112 (3.302) 6.681 (2.180)ENG () () 1.453 (0.380) () 2.803 (1.669)MATH () () () () 2.456 ()PHYS () () 2.746 (2.724) () 3.928 (1.568)SHS () 0.704 (0.697) () () ()

Table 6The patenting scores per permanent researcher performance of labs in classes and disciplines (mean values of patent-perm and standarderrors in parentheses)

Class 1 Class 2 Class 3 Class 4 Class 5

BIO 0.025 (0.559) 0 () 0 () 4 () 0.652 (.649)MED 0.161 (0.308) () 0 () 0.383 (0.291) ()PHARMA 0.143 () 1.5 () 0.616 (0.651) 1.359 (0.975) 0.035 ()CHEM 0 () () 0 () 0.625 (0.883) 0.419 (0.382)ENG () () 0.75 (1.061) () 0.382 (0.295)MATH () () () () 0.063 ()PHYS () () 0 () () 0.146 (0.175)SHS () 0 () () () ()

Table 7Description of the variables used in the multiple linear regression

Variable Description

Perf-perm Average publication performance of permanents in the labPerm-res Number of permanent researchers (professors and full-time researchers)PhD-pm Number of PhD students per permanent researchersPost-doc-av Number of post-docs per permanent researchersNon-res-pm Number of non-researchers per permanent researchersPromotion-pos Share of promoted (holding senior positions) among permanent researchersFull 1 Dummy variable, first quartile of the share of full-time among permanent researchersFull 2 Dummy variable, second quartile of the share of full-time among permanent researchersFull 3 Dummy variable, third quartile of the share of full-time among permanent researchersFull 4 Dummy variable, fourth quartile of the share of full-time among permanent researchersAge 1 Dummy variable, first quartile of the average age of permanent researchersAge 2 Dummy variable, second quartile of the average age of permanent researchersAge 3 Dummy variable, third quartile of the average age of permanent researchersAge 4 Dummy variable, fourth quartile of the average age of permanent researchersPatent-perm Number of patent (applications) invented per permanent researchers of the labPub-internat Share of internationally co-authored publication occurrences among all publications of the labPub-indus Share of publications co-authored with industrial researchers among all publications of the labSHS Dummy for human and social sciencesBMP Dummy for biology, medicine and pharmacyEMP Dummy for engineering, mathematics and physicsCHEM Dummy for chemistry


but it may usefully complement it by evidencing themain relations between per capita publication per-formance and other variables of interest that tran-scend the whole population. The variables used aredescribed in Table 7 and the regression results arepresented in the Table 8. Even if the regression ex-hibits a high goodness-of-fit score, these results shouldstill be taken cautiously due to the limited number ofobservations.

Among independent variables, there is only onesize variable, namely per-res which gives the num-ber of permanent researchers. We find a significant andnegative coefficient for that variable, suggesting smallsized laboratories are more productive. This result isconsistent with the results previously obtained in theliterature (cf. Bonaccorsi and Daraio, 2002; Mairesseand Turner, 2002).

Concerning the age of the permanent researchers,the results are also consistent with the standard resultsin the literature. We find that only the third quartileof labs with respect to average age (Age 3) pub-lish significantly more than the first quartile Age 1.Such result is compatible with an inversed-U shapeof productivity according to age. Nevertheless, suchresult should still be taken cautiously since we onlyuse cross-section data and that it is therefore difficultto distinguish between age and cohort effects (fora discussion on that issue see Stephan, 1996). Thehypothesis of a potential effect of age mix on pro-ductivity has also been tested. Standard deviation ofpermanents ages has been included in the regression.Nevertheless, it never appeared to be significantly re-lated to publications. Since dropping it was increasingthe fit of the model, we decided not to include it inthe final regression.

Considering the professional status of the perma-nents, we find that the share of promoted amongpermanent researchers (Promotion-pos) significantlyincreases the publication performance of the labs.Two explanations may support the positive coefficientfor promotion. The scientists who have the higherabilities for academic performance are likely to be thepromoted ones (selection effect). Moreover, promo-tion may improve the professional status of scholarssubstantially, increasing their productivity just be-cause they can better exploit unmeasured externalresources (status effect) thus bringing in the labs somecritical resources.

Table 8OLS estimates of a multiple linear regression of the publicationper permanent researcher performance of labs (perf-perm)Dep var: perf-perm Coefficient Standard error

Perm-res 0.036 0.018PhD-pm 0.274 0.618Post-doc-av 0.511 0.410Non-res-pm 1.401 0.346Promotion-pos 3.907 1.579Full 1 Ref.Full 2 1.502 0.718Full 3 1.319 0.762Full 4 0.879 0.831Age 1 Ref.Age 2 1.059 0.770Age 3 1.318 0.714Age 4 0.484 0.738Patent-perm 1.300 0.447Pub-internat 0.518 1.551Pub-indus 7.964 4.864SHS Ref.BMP 2.885 1.076EMP 2.495 1.186CHEM 3.480 1.189354Constant 1.905 1.446547Adjusted R2 0.5423NB: Standard errors are in parentheses. , and indicatethat coefficients are statistically significant at the 0.01, 0.05 and0.10 levels, respectively. For the categorical variables, coefficientshould be understood as compared with the first modality whichis taken into reference (Full 1, Age 1 and SHS).

One important result is that it is only the secondand (in a lesser extent) the third quartiles of labs ac-cording to the share of full-time among their perma-nents (Full 2 and Full 3) that significantly publishmore than the ones of the first quartile (Full 1). Thistends to confirm the first insights of the typology in-dicating that it is by mixing together permanents thathold full-time research and teach-and-research posi-tions that one may globally sustain publication produc-tivity. The labs that have between 14 and 43% of thetheir permanents holding full-time positions (Full 2equals one), tend to publish even more than the labsthat have higher share of full-time researchers.

The effects of non-permanent researchers andnon-researchers have been recorded by using theirratios with respect to the number of permanent re-searchers: Phd-pm, Post-doc-pm and Non-res-pm are,respectively, the number of PhD students, the numberof post-docs and the number of non-researchers per


permanent researchers in the lab. Among these vari-ables, only Non-res-pm appears to be significant. Thenon-researchers increase significantly the publicationoutcome of research laboratories.

As expected, laboratories disciplines affect sig-nificantly publication performance. We chose tohave quite aggregated disciplinary dummies. Thefour-dummies configuration (see Table 8) has beenretained since numerous iterations showed that it isthe one that allows for the highest adjusted-R2. Wefind that all disciplines publish more than the socialand human sciences (SHS), with the highest coeffi-cient recorded for chemistry (3.480). The disciplinarydummies essentially serve for control purposes sinceproductivity naturally varies and SCI and SSCI cov-ering are of unequal quality across fields.

7. Discussion and conclusion

This paper offers a first empirical investigation ofan original dataset describing the research activity ofa large European university which allows us to anal-yse the organization of research at the laboratory level.We study the labour force composition of labs, howthese various inputs are correlated with the variousoutputs, and the output structure of the labs. Secondly,we construct a typology of labs, which identifies dif-ferent styles of research organization and their out-comes. Finally, we run an econometric model to anal-yse the determinants of collective publication. The useof three different, but complementary empirical meth-ods allows us to reach an important statement: labshave different organizations, which influence their sci-entific performance.

In the correlation studies we observe that full-timeresearchers are often associated to university profes-sors who hold teach-and-research positions and thateach type of permanent scientist attracts different kindof non-permanent researchers. PhDs appeared to bemore intensively associated to university professorsfollowing primarily a logic of personal contacts andonly secondarily a logic of excellence. The latter ap-plies primarily to post-docs who are correlated withfull-time scientists.

University professors tend to decrease all thescores in average outcomes. Concerning publication,this result seems to be induced by the presence of

non-publishing university professors. We have no ev-idence of a distinctive behaviour between publishingprofessors and full-time researchers. The contributionof non-promoted permanent researchers is weak. Thenumber of non-permanent researchers affects the dis-tribution of outputs among the permanent researchersof the lab: PhDs and foreign post-docs increased withthe kurtosis and the variance of the average publica-tion performance of permanent researchers. That mayindicate that they are attracted by some publicationstars and/or that they work preferentially with themost productive researchers, thus specifically enhanc-ing their productivity. Non-permanent researchersand especially French post-docs are important for thepatenting activity of labs. Full-time researchers are themost important contributors to patent and publicationproduction.

About the outcome structure, we found quite sur-prisingly that the share of international collaborationsis not associated with a higher average publicationperformance. Those who collaborate more inten-sively with international co-authors are not neces-sarily those who publish more. On the contrary, theaverage performance in international collaborationsgoes along with a high performance in terms ofcollaborations with industrial partners. One impor-tant result is that the intensity of patenting activityis correlated with all publication intensity measures(both per permanent researcher): strongly with theintensity of publications with industrial partners andweakly with international partners. Moreover, whileit grew over the period covered (19932000), thepatenting activity seems to become more and moreintensively linked to and supported by publicationperformance.

These first analyses of the combination of the dif-ferent types of personnel in labs, of their contributionto the scientific production and of the correlation be-tween outcomes led us to build a typology based onthe simultaneous use of input and output variables. Weobtained five classes of labs, which illustrate five dif-ferent styles of research organization and production.A class (class 2) groups together all the labs belongingto the social sciences, geography and humanities andanother (class 5) covers the big labs. The first classcounts the highest share of full-time researchers, fewnon-permanent researchers, many non-researchers;they publish more than the average but patent much


less than the average. In class 3 contrary to class 1 wefind a high share of teach-and-research positions withmany PhDs and non-promoted permanent researchers.This combination explains the low publication andpatenting performances (cf. Section 4.3). Class 4 in-volves an equal share of full-time researchers andprofessors, permanent scientists are promoted andrather young, assisted by many non-permanent col-leagues and working on an inter-disciplinary basis.This specific association generates the highest scoresin terms of both patents and publications.

Finally we have complemented the typology andthe correlation analyses by focusing on the deter-minants of the collective publication performances(using an OLS regression). Our most robust con-clusions that are supported by the different analysescould be summarized as follows. Firstly, all method-ologies used underline that the most publishing labsare also patenting. This tends to reject the crowdingout hypothesis between publications and patenting.Nevertheless our study is not totally formally conclu-sive in that direction since we could not control forlaboratory specific unobservable effects. Secondly,the mid-career promotion system in France appearsas a consistent screening process since the share ofpromoted permanents always favour both publicationand patent counts. Thirdly, we find that full-time andteach-and-research positions are complements sincea right proportion (around an equal share) of bothpositions within labs induces a high performance interms of publications. The presence of full-time re-searchers may contribute maintaining incentives forthe university professors to perform research. Thiscomplementarity may also be due to their respectivepower of attractiveness: full-time researchers attractpost-docs and university professors increase the num-ber of PhD candidates. Such insight as well as severalother complementary ones highlight the often ignoredrole of non-permanent researchers. This last result iscertainly an important insight of our study to showthat policy makers and managers of laboratories anduniversities should design strategies that explicitlytake into account their ability to attract the grow-ing population of non-permanent researchers. In theemerging European research space, the attractivenessof labs for young researchers, who constitute a highlymotivated human resource, may well make the differ-ence in terms of performance.

Acknowledgements

This work is part of a larger project on knowledgeproduction at ULP developed by a team of researchersat BETA led by P. Llerena. We are grateful to him andto all other members of the team whose support wascrucial at some points: Joaquin Azagra-Caro, MoniqueFlasaquier, Caroline Hussler, Rachel Levy, Thuc UyenNguyen Thi, Musa Topaloglu, Panos Vassiliadis, andSandrine Wolff. Acknowledgements should be ex-tended to the administrative departments of ULP,the technology transfer office (ULP-Industrie), andto the French patent office (INPI) for allowing us toget access to their databases. A. Bonaccorsi (SantaAnna School of Advanced Studies, Pisa) and F. Lis-soni (CESPRI, Universit Bocconi, Milano) shouldbe acknowledged for their help in data collection.Thanks also to P. Roux who provided us technicalassistance. A preliminary version of this paper waspresented at the third EMAEE (European Meeting onApplied Evolutionary Economics) Conference TheKnowledge-Based Economies. New Challenges inMethodology, Theory and Policy, Augsburg, Ger-many, April 1012, 2003.

References

Adams, J., Griliches, Z., 1996. Measuring science: an exploration.Proceedings of the National Academy of Science 93, 1266412670.

Adams, J., Griliches, Z., 1998. Research productivity in a systemof universities. Annales dEconomie et de Statistique 49/50,127162.

Allison, P.D., Long, J.S., 1990. Departmental effects on scientificproductivity. American Sociological Review 55, 469478.

Arora, A., David, P.A., Gambardella, A., 1998. Reputation andcompetence in publicly funded science: Estimating the effectson research group productivity. Annales dEconomie et deStatistique 49/50, 163198.

Benzcri, J.-P., 1992. Correspondence Analysis Handbook. MarcelDekker, New York.

Bonaccorsi, A., Daraio, C., 2002. The organization of science.Size, agglomeration and age effects in scientific productivity.In: Proceedings of the Conference Rethinking Science Policy:Analytical Frameworks for Evidence-based Policy. SPRU,Brighton, March 2123.

Carayol, N., 2003a. An Economic Theory of Academic Com-petition: Dynamic Incentives and Endogenous CumulativeAdvantages, Mimeo, BETA, University Louis Pasteur,Strasbourg.


Carayol, N., 2003b. Objectives, agreements and matching inscience industry collaborations: reassembling the pieces of thepuzzle. Research Policy 32, 8871148.

Cole, S., Cole, J., 1973. Social Stratification in Science. Universityof Chicago Press, Chicago.

Coup, T., 2003. Science is golden: academic R&D and universitypatents. Journal of Technology Transfer 28, 3146.

Crow, M., Bozeman, B., 1987. R&D laboratory classification andpublic policy: the effects of environmental context on laboratorybehavior. Research policy 16, 229258.

Dasgupta, P., David, P.A., 1994. Toward a new economics ofscience. Research Policy 23, 487521.

David, P.A., 1994. Positive feedbacks and research productivity inscience: reopening another black box. In: Granstrand, O. (Ed.),Economics of Technology. Elsevier, Amsterdam.

Diamond, A.M., 1986. The life-cycle research productivity ofmathematicians and scientists. The Journal of Gerontology 41,520525.

Foltz, J., Barham, B., Kim, K., 2000. Universities and agriculturalbiotechnology patent production. Agribusiness 16, 8295.

Fox, M.F., 1992. Research, teaching and publication productivity:mutuality versus competition in academia. Sociology ofEducation 65, 293305.

Download - Carayol & Matt 2004 - Does Research Organization Influence Academic Production

Top Related