SMALL WORLDS IN NETWORKS OF INVENTORS AND THE ROLE OF SCIENCE: AN ANALYSIS OF FRANCE

Transcription

1 SMALL WORLDS IN NETWORKS OF INVENTORS AND THE ROLE OF SCIENCE: AN ANALYSIS OF FRANCE FRANCESCO LISSONI (1), PATRICK LLERENA (2), BULAT SANDITOV (3) (1) Brescia University & KITeS Bocconi University, (2) BETA University of Strasbourg, (3) UNU-MERIT, Maastricht University OECD-KNOWINNO1 workshop on Measuring the use and impact of Knowledge Exchange Mechanisms - Alexandria, VA; November 2011

2 Research questions 1. Structure of inventors networks in France: Are they small worlds (tightly knit communities of inventors & few shortcuts between communities)? Methodological issue: how to test small world properties in an affiliation network? (Uzzi, 2007) 2. Role of academics and CNRS researchers in inventors networks Do they contribute to small-world structure (bridging distant communities of inventors)? If yes, is it a side effect of team size or the result on crossorganizational mobility?

3 Background: social structure in S&T 1. Sociology of science: scientific progress through tight cliques and bridges btw them Invisible Colleges (ICs) (De Solla Price 1963, Crane 1972) ICs revisited by Small World theory (Newman, 2001; Moody, 2004; Goyal el al., 2006) co-authorship and co-citation analysis of scientific publications differences across disciplines, due to differences in team size and consensus 2. Innovation studies /1: do small worlds enhance innovation? Theory (Cowan & Jonard, 2003;2004) (Controversial) empirical evidence: Fleming et al. (2007), Breschi & Lenzi (2011): co-invention networks (patent data) 3. Innovation studies /2: do academics contribute to small worlds? Empirical evidence (centrality of academics): Balconi et al. (2004) on coinvention networks (patent data)

4 Data Sources EP-INV database (KITeS-Bocconi University) Patent applications at EPO reclassified by applicant and inventor Subset of inventors with address in France KEINS database on academic inventors (Lissoni et al. 2006) Matching inventors with a list of university professors in service in Phone-call & checking of matches, with at least one patent after 1993 Dataset on CNRS inventors (Guarisco, 2009; Thibaut, 2009) KEINS methodology applied to CNRS researchers on duty in 2007 NB CNRS is the largest French public research organization PRO); and the weight of French PROs in their national science system is unparalleled in any other country

5 Q: Do academics and CNRS researchers contribute to patenting? A: Yes, esp. in science-based fields

6 Academic & CNRS inventors by patent field, TECHNOLOGICAL FIELDS INVENTORS PATENTS All Acad+CNRS All Acad+CNRS Electrical Eng. Electronics % % Scientific Instruments % % Chemicals. Materials % % Pharmaceuticals. Biotechnology % % Industrial processes % % Mech. Eng. Machines. Transport % % Consumer goods. Civil eng % %

7 Science intensity of patent fields*, per year;

8 Q1: What are networks of inventors? Q2: Are they small worlds? In which fields? Q3: Any intuitive connection between small world properties and science-intensity (presence of academics & CNRS) of the field?

9 NETWORK OF INVENTORS Patents 2-mode (affiliation) network Inventors 1-mode network

10 Small worlds: Watts & Strogatz (1998) Highly connected (largest component: large both with respect to entire network and 2 nd largest) Clustered BUT Short avg length Clustering: C = [3 #(/\)]/ #(/\) Small-world ratio: Q=(C obs /С rg )/(L obs /L rg )

11 Networks of inventors: size of the main components, and distribution of academic inventors, Electr. eng. Electronics Instruments Chemicals & Materials Pharma & Biotech. Industrial processes Mech. Eng., Transport Consumer goods; ALL COMPONENTS & ISOLATES LARGEST COMPONENT 2nd LARGEST COMP All inventors Academic All inventors Academic All inventors Academic All inventors Academic All inventors Academic All inventors Academic All inventors Academic

12 Benchmark random graph (BRN) One-Mode random graphs are not an appropriate benchmark random structure for inventors networks Each inventor connects with the whole research team (rather than with individual inventors) Benchmark random graph (BRN): Keep number of number of inventors per patent and number of patents per inventor, and randomly rewire patent-inventor links Project bipartite graph onto the set of inventors

13 Observed networks vs. simulated benchmark random graph (BRN), by patent field (all patents, ) Electrical eng. & Electronics Instruments Chemicals. Materials Pharm & Biotech Industrial processes Mechanical eng. & Transport Consumer goods & Civil eng. observed simulated observed simulated observed simulated observed simulated observed simulated observed simulated observed simulated

14 W-S model for bipartite graph Rewiring starts here Rewiring starts here

15

16 Q1: Do academic and CNRS inventors contribute to small world properties of networks they belong to? Q2: If yes, does it depends on the scientific mode of production (large teams), their high productivity, or their mobility?

17 Industrial processes Academic CNRS Academic &CNRS inventors as small world catalysts: Centrality Inventors Electrical eng & Electronics Academic CNRS Instruments Academic CNRS Chemicals. Materials Academic CNRS Pharma & Biotech Academic CNRS

18 Productivity & Team size TECHNOLOGICAL FIELDS Productivity All Acad.+ CNRS All Team Sizes Acad.+ CNRS (obs.) Acad.+ CNRS (baseline) Electrical Eng. Electronics Scientific Instruments Chemicals. Materials Pharmaceuticals. Biotechnology Industrial processes Mech. Eng. Machines. Transport Consumer goods. Civil eng

19 Mobility : Number of distinct applicants normalized by the number of inventor s patents TECHNOLOGICAL FIELDS All inventors Acad. inventors CNRS inventors Electrical eng. Electronics Instruments Chemicals. Materials Pharmaceuticals. Biotechnology Industrial processes Mech. Eng.; Transport Consumer goods. Civil eng

20 Academic &CNRS inventors as small world catalysts: Node-deletion test Electr. eng. & Electronics Type removed Nr removed C1 C1 L L/L Acad+CNRS % Controls % Instruments Acad+CNRS % Controls % Chemicals. Materials Acad+CNRS % Controls % Pharma & Biotech Acad+CNRS % Controls % Ind. processes Acad+CNRS % Controls %

21 Summary Networks of inventors in France appear to be small worlds...in science-intensive technological fields in which academic and CNRS inventors are an important presence Academic and CNRS inventors & small worlds They connect otherwise disconnected components They bridge between distant communities of inventors shortening distances They provide shortcuts due to inter-organizational mobility of inventors

22 Policy conclusions & further research Our research does not say anything on desirability of academic patenting It simply exploits evidence of extensive academic patenting to produce relational data on academics and other inventors It points out that academic inventors are in a key structural position in the network What does follows from academic inventors structural position? Further research on: Dissemination effects (citation analysis: Rosell & Agrawal, 2007; Wes Cohen s presentation at the Patent Statistics for Decision Makers Conference, next Thursday) Leverage and brokerage: access to resources (both cognitive and financial: Lissoni, 2010; evidence on academic inventors superior scientific productivity)

23 More on inventor data (networks, academic, and more) 1. Lunch Talk - Inventor data disambiguation & sharing (1-1.45pm) Sponsored by the European Science Foundation ("APE-INV" project on ) F. Lissoni New trends in inventor data production and use V. Torvik Probabilistic disambiguation of author-inventor names across PubMed and USPTO data 2. As part of PATSTAT user meeting (from 2pm, today) M. Pezzoni & A. Maurino "PATSTAT linkable new database on inventor names" G. Tarasconi "Metadata structure proposal for sharing address cleaning patterns for PATSTAT"