Post on 03-Feb-2016
description
transcript
LKS or markets? Social network analysis and the geography of knowledge diffusion
F. Lissoni (Università di Brescia and KITES-Bocconi Univ.)
DIMETIC Doctoral European Summer School Session 2: REGIONAL AND POLICY DIMENSIONS OF INNOVATION AND GROWTH
Pecs, June 29th - July 10th 2009
Localized Knowledge Spillovers
DIMETIC in Pecs - Flix on LKS 2
LKS: A (quick) critical survey
Localized Knowledge Spillovers are defined in the literature as …
Knowledge externalities – positive effects of scientific or technical discoveries on the productivity of firms, which neither made the discovery themselves, nor licensed their use from the holder of the intellectual property right
…bounded in space – firms located nearby the sources of knowledge externalities benefit more than firms located elsewhere, thus introducing innovations at a faster rate
LKS as a “local public good”:
Widespread use in econometric literature on geography of innovation, based upon the “knowledge production function”
Great reliance on the concept of “tacit knowledge”, first popularized (as “contextual” or “rooted”) by Italian literature on Industrial Districts
DIMETIC in Pecs - Flix on LKS 3
“The theory of knowledge spillovers, derived from the knowledge production function, suggests that the propensity for innovative activity to cluster spatially will be the greatest in industries where tacit knowledge plays an important role. (…) it is tacit knowledge, as opposed to information, which can only be transmitted informally, and typically demands direct and repeated contacts” (Audretsch , 998, p.23 - Italics are mine)
DIMETIC in Pecs - Flix on LKS 4
“[In] each local system an integration between “codified knowledge” and “contextual knowledge” is realized. [...]“[The] coding and decoding of knowledge often involves a set of skills which cannot be set out in a simple standardized code. Rather, it is a matter of complex and often indefinite, and not rarely “indescribable” skills which can be acquired only by direct experience, by repeated practice, or by the process of “seeing at work”” (Becattini and Rullani, 1996; p.162-164 - Italics are mine)
This local know-how is passed on by doing things and seeing how other people do things, through informal chit-chat. […] Above all, this form of knowledge is necessarily rooted in a specific area in which people are linked by the bonds of a shared history or values, where specific institutions work to the benefit of people and where codes of behaviour, lifestyles, employment patterns and expectations are inextricably implicated in productive activity.” (Brusco, 1996; pp. 149-150 - Italics are mine)
DIMETIC in Pecs - Flix on LKS 5
LKS as a by-product of the knowledge production function approach to innovation
aiii KKXY 1
The classic knowledge production function (Griliches, 1992) Output of firm i depends not only on own research
efforts, but also on the pool of publicly available knowledge (wij captures the fraction of knowledge in j borrowed by i)
jija KwK
DIMETIC in Pecs - Flix on LKS 6
The geographic knowledge production function [Jaffe (AER, 1989); Audretsch-Feldman (AER, 1996); Acs et al. (RESTAT, 1994)]
sisisisi URIRDI 21.
I = region s / industry i innovative output IRD = Private firms R&D expenditures UR = University R&D expenditures
Methodological problems: Spatial unit of observation (i.e. state, county, city) Spatial econometric issues (i.e. spatial autocorrelation) Localisation of production (i.e. correlation with other variables)
DIMETIC in Pecs - Flix on LKS 7
Many positive findings of “spillovers” and LKS via the knowledge production function approach spatial proximity and its dual interpretation
• Spatial proximity and knowledge tacitness– Knowledge produced by firms (and/or) universities (partly)
spills over positive externality– Knowledge that spills over is mainly ‘tacit’- i.e. highly
contextual, difficult to codify-, more easily transferred through face-to-face contacts
– Tacit knowledge is a pure public good, but a local one, i.e. most readily available to firms located nearby the sources of knowledge
• Spatial proximity raises the likelihood of establishing a contact among agents (epidemic logic)
DIMETIC in Pecs - Flix on LKS 8
Two key problems with these explanations:
1. The odd role of knowledge: tacit but public
shouldn’t be easy to keep tacit knowledge secret (at least in part)? and sell it?
shouldn’t sharing (via repeated interaction) be a voluntary act? at a price?
2. Little enquiry on knowledge transmission means (many examples, often contradictory):
2a. local labour markets for scientists & engineers they may/may not generate externalities
2b. networks of scientists and technologists social or commercial?
2c. occasional meetings (social networks at large) how can people from different background share tacit knowledge?
LKS spatial proximity and its dual interpretation (cont.)
DIMETIC in Pecs - Flix on LKS 9
Example of transmission mean 1: Labour market for scientists & engineers (Almeida-Kogut, MS 1999)
Recent confirmatory work by Lee Fleming and associates: effects of “non compete” rule in labor markets of different US states
We come back to it later (some technicalities need first to be explained)
“knowledge spillovers” significantly more localised in Silicon Valley (SV) than in other US regions
Intra-regional mobility in SV is much higher than in other regions
Inventors’ intra-regional mobility has a significant and positive effect on the probability to observe knowledge spillovers
DIMETIC in Pecs - Flix on LKS 10
• Positive relationship between biotech firms’ innovative performance and scientific strength of local Universities, as proxied by number of scientific papers authored by ‘star’ scientists
• HOWEVER, when ‘stars’ are divided into firm scientists (linked) and purely academic scientists (untied), the explanatory power of the latter vanishes
Knowledge embodied in scientists has high degrees of natural excludability
Only way to build upon that knowledge is to enter into contractual arrangements or start-up a new firm
Scientists that do so retain their affiliation and get jobs within commuting distance
Example of transmission mean 2:Markets for technologies (Zucker et al, EI 1998)
DIMETIC in Pecs - Flix on LKS 11
1. Patent citations as “paper trail”
“Knowledge flows leave a paper trail in the form of citations in patents ….to the extent that regional localization of spillovers is important, citations should come disproportionately from the same state/area as the originating patent” (Jaffe, Trajtenberg and Henderson, QJE 1993: 578)
“Knowledge flows (…) are invisible; they leave no paper trail by which they may be measured and tracked, and there is nothing to prevent the theorist from assuming anything about them she likes” (Krugman, 1991: 53)
Most influential attempt to measure LKS directly (no KPF). Reaction to trade theorists’ skepticism
The silver bullet? LKS and patent citations
DIMETIC in Pecs - Flix on LKS 12
The meaning of patent citations
PatentApplication (Y)
Citation ofprevious patents (X)
Searchreport
Citations added bythe examiner
A citation of Patent X by Patent Y means that X represents a piece of previously existing knowledge upon which Y builds paper trail of knowledge flow
DIMETIC in Pecs - Flix on LKS 13
JTH’s experiment with patent citations
Sample of citedpatents
(i.e. originators ofknowledge flows)
All subsequentpatents citing them
(i.e. benefit fromKnowledge flows)
For each pair, citing-cited, compare geographic address of inventors: if they come from the same region, then co-location=1, else co-location=0
Compute the fraction of pairs, citing-cited, for which co-location=1 if fraction is high then knowledge flows are highly “localised”
Compared to “what”? What is the “expected” fraction?
Step 1)
Step 2)
DIMETIC in Pecs - Flix on LKS 14
The “expected” fraction of localised citations
Sample of citedpatents
(i.e. originators ofknowledge flows)
All subsequentpatents citing them
(i.e. benefit fromKnowledge flows)
For each citing patent a “control” patent with same technology class, but not citing the same patent (mimics the existing spatial distribution of innovative activities)
Repeat steps 1) and 2) as above, i.e. compute the fraction of control-cited pairs for which co-location=1 this fraction provides the “expected” value
DIMETIC in Pecs - Flix on LKS 15
Actual vs. expected co-location of citations
Geographical matching with cited patents (%)
Citing Controls t-statistic
Geographical level
Country 69.3 58.5 7.2
State 10.5 4.1 7.9
Metropolitan area 6.9 1.1 9.6
Evidence that citations (spillovers) are more geographically localised than expected
So
urce
: JTH
DIMETIC in Pecs - Flix on LKS 16
2. Replicas of JTH experiments: market-based diffusion channels
MARKETS FOR TECHNOLOGY (Mowery-Ziedonis, 2004)Transfer through licensing goes along with localized transfer of tacit knowledge Evidence that spillovers (i.e. citations without licensing) are less localised than knowledge flows mediated by licenses
LABOUR MARKETS (Almeida-Kogut, MS 1999) Patent citations are significantly more localised in Silicon
Valley (SV) than in other areas Intra-regional mobility in SV is much higher than in other
regions Inventors’ intra-regional mobility has a significant and positive
effect on the probability that a patent will build upon a major patent from the same region
DIMETIC in Pecs - Flix on LKS 17
DIMETIC in Pecs - Flix on LKS 18
DIMETIC in Pecs - Flix on LKS 19
DIMETIC in Pecs - Flix on LKS 20
• Revisitation of JTH experiment , but controlling for “market-induced” social ties among inventors
• Key intuition: JTH is just a “localization test” → Absent direct measures of knowledge exchanges, spatial proximity captures ALL the mechanisms through which knowledge is transmitted: what’s left of spatial proximity when we control for social proximity?
• How to control for social proximity? NETWORK OF INVENTORS!
3. BL replication of JTH experiments: telling spatial and social proximity apart
DIMETIC in Pecs - Flix on LKS 21
• Revisitation of JTH experiment , but controlling for “market-induced” social ties among inventors
• Key intuition: JTH is just a “localization test” → Absent direct measures of knowledge exchanges, spatial proximity captures ALL the mechanisms through which knowledge is transmitted: what’s left of spatial proximity when we control for social proximity?
• How to control for social proximity? NETWORK OF INVENTORS!
3. BL replica of JTH experiments: telling spatial and social proximity apart
DIMETIC in Pecs - Flix on LKS 22
cross-firm inventors
Network of inventors: co-invention & mobility
DIMETIC in Pecs - Flix on LKS 23
Data
SOURCE: • EPO-CESPRI database All patent applications to the European Patent
Office (EPO), by company & inventor
SAMPLE:• All patents of US applicants with at least one US inventor in:
– Biotechnology (IPC C12M to -S)– Drugs (IPC A61K)– Organic chemistry (IPC C07, excl C07B)– (63188 inventors; 66349 patents)
• Location of patents by inventors’ MSAs and States• Time coverage 1978-2002 → co-invention network built assuming a life
span of 5-years for collaboration ties
DIMETIC in Pecs - Flix on LKS 24
Network of US inventors: quite (almost) a small world!
*last year of 5-years time window, e.g. 1999 refers to collaborations from 1995 to 1999
* * *
DIMETIC in Pecs - Flix on LKS 25
Example of small world network
DIMETIC in Pecs - Flix on LKS 26
Few inventors move across firms… even less in space!
DIMETIC in Pecs - Flix on LKS 27
Low (spatial) mobility of (firm) mobile inventors correlation between social and spatial distance
0
500
1000
1500
2000
2500
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
Social distance (d)
Sp
atia
l dis
tan
ce (
Km
)
DIMETIC in Pecs - Flix on LKS 28
Mapping inventors’ distance onto patents
• Given a pair of patents (e.g. citing-cited/control-cited)– They are socially connected if at least two inventors from respective
teams are part of the same network component– The geodesic distance between them is equal to the minimum
geodesic distance between inventors in the respective teams• Distance 0: same inventor in both teams• Distance 1: at least two inventors have previously co-invented• Distance 2: at least two inventors have a common co-inventor
etc. etc.• Distance is infinite if inventors belong to disconnected
components (i.e. they are not mutually reachable)
DIMETIC in Pecs - Flix on LKS 29
JTH-like sampling network properties of samples: “citing” patents are closer to “cited” ones than controls (i)
DIMETIC in Pecs - Flix on LKS 30
JTH-like sampling network properties of samples: “citing” patents are closer to “cited” ones than controls (ii)
DIMETIC in Pecs - Flix on LKS 31
Replicating JTH, and controlling for networks of inventors geograph. matching at MSA/State level, by social distance
DIMETIC in Pecs - Flix on LKS 32
[same for [same for ncnc]]
DIMETIC in Pecs - Flix on LKS 33
DIMETIC in Pecs - Flix on LKS 34
DIMETIC in Pecs - Flix on LKS 35
Interpretation• Cross-firms mobility of workers drive the diffusion of knowledge cross-
firm citations are in fact self-citations by mobile inventors• Mobile inventors contribute to create (spatially localized) networks, and
knowledge is further passed on through these networks (but only as SHORT distances)
• Spatial proximity is largely a mere proxy of social ties, BUT NOT “any” social ties professional ones (co-invention) Once controlling for such social ties, the role of spatial promixity fades away
DIMETIC in Pecs - Flix on LKS 36
Other controls for social ties:
PREVIOUS WORK EXPERIENCES:Agrawal A.K., Cockburn I.M., McHale J. (2007), “Gone But Not Forgotten: Labor Flows, Knowledge Spillovers, and Enduring Social Capital”, Journal of Economic Geography (forthcoming)
ETHNIC TIESAgrawal A.K., Kapur D., McHale J. (2004), “Defying Distance: Examining the Influence of the Diaspora on Scientific Knowledge Flows”, mimeo (NBER WP?) these ties could prove to be less professional, and more “social in strict sense” (externalities)
DIMETIC in Pecs - Flix on LKS 37
A flaw in JTH methodology…?
• Thompson and Fox-Kean (AER, 2005): JTH method of selecting control patents may induce spurious evidence of localised spillovers – Control and citing patents may be unrelated and thereby control
patents fail to control for pre-existing spatial concentration of innovations
• 3-digit USPC main classes too broad • Matching should cover also secondary classes
• JTH results vanish when narrower than 3-digit classes are selected and citing-controls match also at the secondary class level
DIMETIC in Pecs - Flix on LKS 38
EP671165 Collagen-based injectable drug
delivery system and its use. Applicant: COLLAGEN CORP (US)
Palo Alto (CA)
Primary IPC: A61K9/00
CITING PATENT
EP671165 Collagen composites for controlled
drug release Applicant:VITAPHORE CORP (US)
Menlo Park (CA)
Primary IPC: A61K9/00
CITED PATENT
EP671165 Calcium dietary supplement
Applicant: CYANAMID CORP (US) Wayne (NJ)
Primary IPC: A61K33/10
CONTROL PATENT
Citing and cited patents come from the same industry. Control patent, while matching the citing one at the 4-digit IPC, comes from a different industry. Control patent should be selected from within A61K9/00
DIMETIC in Pecs - Flix on LKS 39
DIMETIC in Pecs - Flix on LKS 40
DIMETIC in Pecs - Flix on LKS 41
… or just one more hint of the role of social ties?
• Increase technological proximity between citing and control patents Select patents produced by inventors belonging to the same technological community, which are possibly
– linked by strong social connections…
– … and (therefore) spatially co-located
• What is flawed is not the JTH methodology, but the interpretation of localised patent citations as evidence of pure “Marshallian” externalities (spillovers)
• Only citations mediated by market mechanisms or collaborative networks are likely to be spatially localised. Unmediated citations do not show any localisation pattern.
DIMETIC in Pecs - Flix on LKS 42
% of personal/socially connected patents in the citing and control sample; by technology matching criteria
0
5
10
15
20
25
30
35
4-digit 12-digit 12-digit (main) & 4-digit(secondary)
12-digit (main &secondary)
% c
onne
cted
Citing
Control
DIMETIC in Pecs - Flix on LKS 43
Summary and conclusions• Geography matters because inventors are mobile across firms, but not in
space, and the resulting social networks are also localised
• However, social networks of inventors may be market-mediated: inventors are professionals: they sell their knowledge, not just share it (with other inventors and companies) inventor-network ties originate either from team work: what rules for knowledge sharing?
• Mobility of skilled workers has a key role as driver of knowledge diffusion direct: spreads/sell knowledge across firmsindirect: connects teams and reduces social distance
• By imposing a narrower technological focus, one just deals with a socially connected and co-localised community of scientists no difference in the co-localisation pattern between citing and control
DIMETIC in Pecs - Flix on LKS 44
Research agenda: What does remain of the ‘spillovers’ story?
• What do you mean by “mobile”? Causes and consequences of labour mobility- real mobility (job changes)- contract R&D and consulting (also from academics)- collaborative research (patent co-assignment; academic patents)- M&AsLaforgia F., Lissoni F. “What do you mean by ‘mobile’? Multi-applicant inventors in the European Biotechnology Industry”, in; Malerba F., Vonortas N. (eds.), Innovation Networks in Industries, Edward Elgar, 2009 (forthcoming)
• The economics of science and academics’ participation to the markets for technologies ( this my current line of research)
NEED TO COLLECT DATA ON INVENTORS (and not just on patents or their applicants)
INVENTOR DATA: MANY (RELATED) POSSIBILITIES- network- geography- name-matching academic inventors!!!!!
DIMETIC in Pecs - Flix on LKS 45
Lissoni F., and P.Lotz, J. Schovsbo, A. Treccani (2009) “Academic Patenting and the Professor’s Privilege: Evidence on Denmark from the KEINS database”, Science and Public Policy (forthcoming)
Lissoni F., and P.Llerena, M.McKelvey, and B.Sanditov. “Academic Patenting in Europe: New Evidence from the KEINS Database”, Research Evaluation 16, 2008, pp 87-102 Reprinted in: M. McKelvey and M.Holmén (eds.), European Universities Learning to Compete: From Social Institution to Knowledge Business, Edward Elgar, Cheltenham
Soon to come: papers on the Netherlands and the UK
DIMETIC in Pecs - Flix on LKS 46
DIMETIC in Pecs - Flix on LKS 47
DIMETIC in Pecs - Flix on LKS 48
Weight of academic patents on the total patents by domestic inventor, by country and type of ownership (1994-2001)
1,0%
4,0% 4,0%4,3%
6,0%
3,4%
0,3% 0,4%
6,2%
0,3%
0,0%
1,0%
2,0%
3,0%
4,0%
5,0%
6,0%
7,0%
University ow ned academic patents All academic patents
The Netherlands US France Italy Sw eden
DIMETIC in Pecs - Flix on LKS 49
DIMETIC in Pecs - Flix on LKS 50
Balconi M., Breschi S., Lissoni F., “Networks of inventors and the role of academia: an exploration of Italian patent data” , Research Policy 33/1, 2004, pp. 127-145
Academic scientists are more central in network of inventors (they are at average shorter distance to other inventors)NB: Academic inventors are “mobile” because they patent for different firms
Ongoing research + previous findings in US literature
University patents are more highly cited than non-university ones
RESEARCH QUESTIONS: is higher citation rate of university patents explained (also) by academic inventors’ position in the network?
DIMETIC in Pecs - Flix on LKS 54
Average of the received citations in the Netherlands, Italy, France and Sweden. 1994-2002
2,13 2,11 2,162,04
1,92
2,242,30
1,89
0,00
0,50
1,00
1,50
2,00
2,50
THE NETHERLANDS ITALY FRANCE SWEDEN
Academic patents Not academic patents
Citations to academic patents
DIMETIC in Pecs - Flix on LKS 55
Citations to academic patents
DIMETIC in Pecs - Flix on LKS 56
A big project on inventorsESF-APE-INV, 2009-2013:
• creation of a European database on inventors studies on mobility/networks
• Identification of “academic inventors” (university staff who are inventors) studies on technology transfer and networks
2 workshops per year
Data available to all who contribute to their creation (after the end of the project to all researchers)
Short/Long mobility grants for European PhD students and PostDocs
Partners (by now): KITES-Bocconi, ULB, KU Leuven, EPFL, Goteborg Univ., Beta-Strasbourg ( OST), Munich Univ (Ludwig-Maximilian), CBS, and many others