Academic patenting in Germany
description
Transcript of Academic patenting in Germany
© Fraunhofer ISISeite 1
Academic patenting in Germany
A new comprehensive approach for the identification and analysis
of academic patents
Friedrich Dornbusch
© Fraunhofer ISISeite 2
1. Brief introduction of the recently developed approach to indentify academic patents in Germany: How does the matching algorithm work? Briefly two descriptive results: What are the main trends in academic
patenting since the abolition of the “Hochschullehrerprivileg” in 2002?
2. First step towards analyzes of the filing behavior of applicants - focusing on the relationship between universities and firms: How does the regional environment influence the filing and respectively
the co-operation behavior (measured by academic patents) between universities and firms?
Content and aim of the presentation
© Fraunhofer ISISeite 3
Like many European countries Germany implemented a Bayh-Dole like IPR-regime (Geuna/Rossi 2011) in 2002 (abolishment of „Hochschullehrerprivileg“): Universities gain the right and responsibility to exert IPR on “their” inventions and to exploit it. Emergence of new exploitation infrastructure and public funding programs (e.g. Schmoch 2007; von Ledebur 2008).
But large shares of academic patents are still not filed by universities (university invented vs. university-owned) (e.g. Geuna/Rossi 2011; Thursby et al. 2009; Lissioni et al. 2008; Geuna/Nesta 2006):
And consequences for the co-operation patterns of universities and simple transferability of Bayh-Dole Act to European countries are still in need of clarification (Bruneel et al. 2010; Valentin et al. 2007; Fabrizio 2007).
Having methodological problems with regard to identification of academic patents in mind, we development of a new approach for identification and analysis of academic patenting in Germany (and other European countries). Detailed description in: Dornbusch, F; Schmoch, U.; Schulze, N.; Bethke, N. (forthcoming) -
Identification of university-based patents: A new large scale approach.
Background
© Fraunhofer ISISeite 4
Previous approaches mainly based on keyword searches (Schmoch 2007; Czarnitzky et al. 2007; 2011; von Ledebur 2009; von Proff et al. 2011) or matching of lists (Thursby et al. 2009; Lissoni et al. 2008; 2009).
In Germany we do not have official lists on academic staff available and the search for the PROF-title is based on estimations:
Making analyzes on institutional level difficult
Basic idea of our approach is to test for identical names of authors of scientific publications with university affiliation and inventors on patent filings. Data sources: PATSTAT and SCOPUS
Main problem: Large datasets danger of homonyms need to use different selection criteria.
New approach towards identification of patents with academic background
© Fraunhofer ISISeite 5
The matching algorithm
Organization matching
Namematching Time windowmatching
Location matching
Classification matching
PATSTAT X Full strings of last-and first name
Priority year First two digits of the postcode
IPC classification =
WIPO 34SCOPUS Author affiliation
= university
Full strings of last-and first name
Publication year:One year time-lag and time-window
First two digits of the postcode
Scopus classification:fine-/ coarse-grained
x uni-inv = 1 if (a names match + b time match + c location match + d subject match)2) Organization 3) Names 4) Time 5) Location 6) Subject
Selection criteria
Recall Precision F-Scores
R=P (F1) P>R (F0,5)R>P (F2)
1-digit pc 0,76 0,63 0,69 0,65 0,73Standard criterion 2-digit pc 0,71 0,77 0,74 0,76 0,72
F-conc 0,71 0,52 0,60 0,55 0,661-digit pc, F-conc 0,64 0,82 0,72 0,78 0,67
High precision 2-digit pc, F-conc 0,59 0,93 0,72 0,83 0,64High recall 2-digit OR (1-digit pc + F-conc) 0,74 0,72 0,73 0,72 0,74
Recent improvement:
NUTS3 including a distance matrix implemented
Detailed description
in a forthcoming methodologi
cal paper
© Fraunhofer ISISeite 6
Old method indicates falling numbers of academic patents.
New method indicates recovering numbers.
Sinking tendency of professors to indicate their title? (Anecdotic evidence)
Results for Germany – Totals ( s t a n d a r d c r i t e r i o n )
0
500
1000
1500
2000
2500
3000
2001 2002 2003 2004* 2005 2006 2007
Num
ber
DPMA NUTS3_30km Benchmark_PROF EPO NUTS3_30km
© Fraunhofer ISISeite 7
Large firms unaffected
SMEs & Private
Other PROs unaffected
University-owned rising.
Academic patenting in Germany – Shares by different appl icant types ( s t a n d a r d c r i t e r i o n )
0.00
0.10
0.20
0.30
0.40
0.50
0.60
2001 2002 2003 2004 2005 2006 2007
Shar
es
Large firms SME Private University-owned Other PROs
© Fraunhofer ISISeite 8
First steps towards analyzes of fil ing behavior in academic patenting
© Fraunhofer ISISeite 9
Universities as local knowledge hubs (Youtie/Shapira 2008), sources for localized knowledge spillovers and collaborations (e.g. Jaffe 1989; Anselin et al. 1997; Laursen et al. 2011).
Counter question: How does the profile of universities local environment influence their filing and respectively co-operation behavior (measured in academic patents) with firms distinguishing between SMEs and large firms?
Testing for: Geographical distance: Due to higher resource endowments large firms are
more likely to bridge greater distances in order to get access to outstanding university research and smaller firms more likely to depend on local knowledge provided by universities (e.g. Bodas-Freitas et al. 2010; Tödtling 2009; Torre 2008; Asheim/Coenen 2005).
Local knowledge base: The pool and type of local knowledge (embodied in employees of local firms) is likely to influence whether if the university finds cooperation partners with adequate absorptive capacity in the region. In doing so, co-operations with SMEs are expected to underlie stronger influences of the knowledge base than with large firms (e.g. Ostergard 2009; Asheim et al. 2007; Agrawal et al. 2006).
Local technological profile: Besides the knowledge pool test for the local technological profiles influence on the co-operation and filing behavior in academic patenting.
Influence of the local environment of universit ies on the fil ing behavior
© Fraunhofer ISISeite 10
H1: The chance for cooperation with MNEs rises with rising distance. H2: The larger the knowledge base (in the form of highly qualified personnel) in the
region the larger the tendency to cooperate with local firms, especially SMEs. H3: The type of local technological regime influences the tendency to
cooperate with SMEs and MNEs in different ways (exploratory hypothesis).
Dataset on level of single academic patent applications indicating different applicant types (UNI, SME, MNE)
Complemented with: Official sources: Eurostat, Destatis, Bundesinstitut für Bau-, Stadt- und
Raumforschung (BBSR). EUMIDA-dataset for university characteristics. Additional patent information from PATSTAT. Additional bibliometric information on university level (SCOPUS).
Hypotheses and Data
© Fraunhofer ISISeite 11
dV: uni/sme/mne (categorial)
Independent Variables: Distance in km (H1) Number of persons in ht-sectors (H2) Field specific patent intensity (patents/inhabitants) (H3)
Additionally controlling for: Agglomeration effects: Dummies for core regions, concentrated regions,
peripheral regions, local firm size structure. University characteristics: Size, scientific regard, third party funds,
publication intensity. Patent characteristics: Non patent literature (proxy for intensity of the
science link), patent family size.
Variables
© Fraunhofer ISISeite 12
Selection of basic dataset by high precision criterion in order to maximize the validity.
Excluding Fachhochschulen (polytechnics) - only universities in the dataset. Priority year 2007
Drop of Patents appearing more than twice to avoid over overestimation of single patents.
Dataset contains the involved universities and thetype of applicant (uni/sme/mne ) (N=1201).
Dataset c leaning
Patent_CNT Freq. Percen
t Cum.
1 921 72.63 72.632 280 22.08 94.723 39 3.08 97.79
4 28 2.21 100.00
Total 1,268
100.00
dV Freq. Percent Cum.
0 = Uni 373 31.06 31.061 = SME 200 16.65 47.71
2 = MNU 628 52.29 100.00
Total 1,201
100.00
© Fraunhofer ISISeite 13
Summary statist ics
Data sources: Eurostat, Destatis, BBSR, EUMIDA Dataset, PATSTAT, SCOPUS.
Notes:
1 = Units are indicated in thousands; 2 = Units are indicated in hundreds
3 = SME (employees<499); MNU (employees >500)
Variable Obs Mean Std. Dev. Min Max
AVG_DIST_INVTEAM1 1080 82.94 107.28 0.00 728.87
HITEC_EMP1 1201 28.66 18.24 1.48 60.76
MEDHITEC_EMP1 1201 108.06 70.10 8.95 297.46
MEDLOTEC_EMP1 1201 49.13 30.63 8.95 158.56
LOWTECH_EMP1 1201 54.02 24.10 13.25 98.95PATINT_ELECT_ENG1 1201 0.38 0.25 0.04 0.79PATINT_INSTR1 1201 0.28 0.14 0.05 0.52PATINT_CHEM1 1201 0.40 0.24 0.09 1.24PATINT_MED1 1201 0.53 0.31 0.07 1.27PATINT_OTHER1 1201 0.13 0.06 0.01 0.27AGGLO_DUM 1201 1.50 0.78 1.00 3.00
SME_REG2 1201 14.44 6.07 2.93 29.10
MNE_REG2 1201 0.65 0.37 0.07 1.54SCIREG_UNI 1176 8.96 10.40 -37.07 36.69PUB_INT_UNI 1201 0.02 0.01 0.00 0.04
TH_P_FUNDbySTAFF1 1201 38.82 9.55 11.41 65.43
STAFF_UNI1 1201 2749.61 1276.56 31.00 5349.00PAT_FAM_SIZE 1201 2.54 1.85 1.00 15.00PAT_NPL_DUM 1201 0.44 0.50 0.00 1.00uni_sme_mne 3 1201 1.21 0.89 0.00 2.00
© Fraunhofer ISISeite 14
Pre l im inary mode l : Mu l t inomina l Log i t – dV: un i / sme/mnu
Level of significance: *** = 0.01; ** = 0.05; * = 0.10Notes: 1 = Units are indicated in thousands; 2 = Units are indicated in hundreds
Higher distance in inventor teams is positively associated with filings of MNEs.
Higher numbers of employees in local hi-tech sectors positively influence the filings of firms.
Higher numbers in med- and low-tech show negative effect on SMEs to file an academic patent.
Higher numbers in low-tech show a positive effect on UNIs to file an academic patent.
Different local technology regimes:
Referring to the BASE, SMEs have negative effect in INSTR and MED.
Turns into positive for UNIs in marginal effects.
Scientific excellence negative for firm filings, but publication intensity positive for MNEs.
AVG_DIST_INVTEAM 0.465 *** 0.509 *** -0.101 *** 0.018 0.084 ***se 0.154 0.150 0.027 0.016 0.029
HITEC_EMP 0.032 ** 0.047 *** -0.009 *** 0.000 0.009 ***se 0.016 0.014 0.003 0.002 0.003
MEDHITEC_EMP 0.001 0.023 *** -0.004 *** -0.002 ** 0.006 ***se 0.007 0.006 0.001 0.001 0.001
MEDLOTEC_EMP -0.028 * 0.022 * -0.002 -0.006 *** 0.008 ***se 0.015 0.012 0.002 0.002 0.003
LOWTECH_EMP -0.061 *** -0.017 0.005 ** -0.007 ** 0.001 0.023 0.015 0.003 0.003 0.004
PATINT_ELECT_ENG 1.628 5.799 *** -0.985 *** -0.301 1.286 ***se 1.322 1.544 0.258 0.194 0.367
PATINT_INSTR -6.715 *** -8.830 *** 1.696 *** -0.120 -1.576 ***se 2.465 1.983 0.332 0.329 0.455
PATINT_CHEM 1.129 * 1.646 *** -0.311 *** 0.006 0.305 ***se 0.677 0.444 0.079 0.094 0.112
PATINT_MED -2.851 ** -1.133 0.310 ** -0.287 -0.023 se 1.368 0.831 0.154 0.178 0.214
PATINT_OTHER 10.099 ** 10.262 *** -2.078 *** 0.453 1.625 * se 4.762 3.763 0.659 0.668 0.932
DUM_CORE_REGse
DUM_URB_REG -0.909 *** 0.466 * -0.048 -0.125 *** 0.173 ***se 0.279 0.273 0.048 0.022 0.054
DUM_PERI_REG 1.368 *** 0.633 * -0.148 *** 0.152 * -0.004 se 0.477 0.336 0.045 0.091 0.094
SME_REG 0.001 -0.180 *** 0.028 ** 0.016 -0.045 ***se 0.092 0.069 0.013 0.012 0.016
MNU_REG 6.942 *** -1.906 -0.023 1.121 *** -1.098 ** se 2.160 1.847 0.316 0.295 0.464
SCIREG_UNI -0.036 ** -0.024 ** 0.005 *** -0.003 -0.003 se 0.015 0.012 0.002 0.002 0.003
PUB_INT_UNI -0.229 0.457 ** -0.061 -0.073 * 0.134 ** se 0.230 0.233 0.038 0.039 0.061
TH_P_FUNDbySTAFF 0.009 -0.030 ** 0.004 ** 0.004 * -0.008 ***se 0.015 0.012 0.002 0.002 0.003
STAFF_UNI -0.059 0.073 -0.009 -0.015 0.023 se 0.152 0.170 0.029 0.023 0.041
PAT_FAM_SIZE 0.033 0.266 *** -0.043 *** -0.020 ** 0.063 ***se 0.079 0.077 0.014 0.008 0.015
PAT_NPL_DUM -0.466 * -1.015 *** 0.181 *** 0.026 -0.207 ***se 0.280 0.235 0.041 0.039 0.054
FIELD CONTROLSse
YES
REFERENCE REFERENCE
MARGINAL EFFECTSSME (1) MNE (2) UNI (0) SME (1) MNE (2)
YES
COEFFICIENTS (BASE UNI (0))N = 1056 | r²_p = 0.221LR(32): 466.502 | Prob > LR: 0
Regi
onDi
stU
nive
rsity
Pate
nt
© Fraunhofer ISISeite 15
Strengthened position of universities: Rise of universities largely on the expense of privately and SME-owned
patents. Due to resource constraints: “bargaining positions” / “business cycle
effect” … ? The type of local knowledge pool significantly affects filing of and co-
operation with SMEs (controlling for agglomeration effects): Indication that local networks (communities of practice) are important for
universities and SMEs to co-operate These are most likely to be established in hi-tech sectors, due to absorptive capacity and cognitive/science proximity of firm’s employees to university inventors.
Strong effect for distance in inventor teams of MNE-owned patents and strong effect for publication intensity. Indicating MNEs ability to screen distant universities profile?
Prel iminary results and conclusions
© Fraunhofer ISISeite 16
Theoretical (more fine grained) derivation of hypotheses
Specification of the regression approach: “Interpretation problem”: dV „university owned“ as non “collaboration” ? Influence of local factors with more fine-grained classifications of the
technological profile and a specialization index. Perhaps use of local technology cycle times Testing: “Fit” of universities (science/innovation) and regions profile:
Ideas? Testing: Interaction effects for local firm size structure and knowledge
base. Deepening interpretation of results and drawing of conclusions.
Just an idea: Conducting two or three case studies at German universities within different regions and with different profiles to verify the results qualitatively.
(Much) work remaining
© Fraunhofer ISISeite 17
Thank you for your attention!
© Fraunhofer ISISeite 18
Backup
© Fraunhofer ISISeite 19
Previous approaches: Keyword searches (Schmoch 2007; Czarnitzky et al. 2007; 2011; von Ledebur 2009; von Proff et al. 2011).; Matching lists (Thursby et al. 2009; Lissoni et al. 2008; 2009).
(Germany: No official lists existing and search for PROF-title based on estimations.)
New approach: Test for identical names of authors with university affiliation and inventors
on patents. Data sources: PATSTAT and SCOPUS
Main advantages: Enables semi-automated generation of matching lists. Not dependent on indication of PROF-title (no estimations needed). All research relevant staff included (no estimations needed). Enables analyzes on institutional level. Can be applied to different countries, enabling systematic analyses and
comparisons . Main problem: Large datasets danger of homonyms use of different
selection criteria.
New approach towards identification of patents with academic background
© Fraunhofer ISISeite 20
Impact of different select ion cr iter ia
0
1000
2000
3000
4000
5000
6000
7000
1996 1997 1998 1999 2000 2001 2002 2003 2004* 2005 2006 2007
Num
ber
C-conc F-conc 1-digit pc
2-digit pc OR (1-digit pc + F-conc) 2-digit pc Benchmark
1-digit pc + F-conc 2-digit pc + F-conc
© Fraunhofer ISISeite 21
Academic patenting in Germany by region – totals ( y e a r s ’ 0 5 t i l l ’ 0 7 b y s t a n d a r d c r i t e r i o n )
=0=1-10=11-100=101+
© Fraunhofer ISISeite 22
Complemented with: Data from offical sources like Eurostat and Destatis. Data from EUMIDA for university characteristics. Additional patent information from PATSTAT. Additional bibliometric information on university level (SCOPUS).
Example for the dataset structure
© Fraunhofer ISISeite 23
Pre l im inary mode l : Mu l t inomina l Log i t – dV: un i / sme/mnu
Level of significance: *** = 0.01; ** = 0.05; * = 0.10Notes: 1 = Units are indicated in thousands; 2 = Units are indicated in hundreds
CATEGORIES (BASE UNI (0))
AVG_DIST_INVTEAMse
HITEC_EMP 0.024 0.045 *** -0.008 *** -0.001 0.009 ***se 0.015 0.013 0.002 0.002 0.003
MEDHITEC_EMP 0.004 0.021 *** -0.003 *** -0.001 * 0.005 ***se 0.006 0.006 0.001 0.001 0.001
MEDLOTEC_EMP -0.027 * 0.025 ** -0.002 -0.006 *** 0.009 ***se 0.015 0.012 0.002 0.002 0.003
LOWTECH_EMP -0.053 ** -0.006 0.003 -0.007 ** 0.004 0.024 0.017 0.003 0.004 0.004
PATINT_ELECT_ENG 2.314 * 6.367 *** -1.063 *** -0.284 1.348 ***se 1.369 1.426 0.235 0.210 0.344
PATINT_INSTR -6.278 *** -8.306 *** 1.541 *** -0.100 -1.442 ***se 2.431 1.848 0.320 0.337 0.429
PATINT_CHEM 1.271 * 1.367 *** -0.265 *** 0.051 0.214 ** se 0.740 0.438 0.086 0.099 0.104
PATINT_MED -2.958 ** -0.710 0.246 -0.358 * 0.112 se 1.321 0.859 0.151 0.186 0.224
PATINT_OTHER 8.878 * 7.798 ** -1.588 *** 0.524 1.064 se 4.682 3.594 0.613 0.705 0.921
DUM_CORE_REGse
DUM_CONC_REG -0.770 *** 0.675 ** -0.079 -0.134 *** 0.213 ***se 0.276 0.322 0.050 0.021 0.058
DUM_PERI_REG 1.272 *** 0.333 -0.109 ** 0.182 ** -0.073 se 0.448 0.349 0.049 0.089 0.092
SME_REG 0.037 -0.185 *** 0.026 ** 0.023 ** -0.049 ***se 0.092 0.069 0.013 0.012 0.015
MNU_REG 5.187 *** -2.601 0.145 1.001 *** -1.146 ** se 1.964 1.842 0.296 0.309 0.473
SCIREG_UNI -0.025 * -0.020 0.004 * -0.002 -0.002 se 0.015 0.013 0.002 0.002 0.003
PUB_INT_UNI -0.375 0.355 -0.035 -0.089 * 0.124 * se 0.260 0.255 0.038 0.049 0.071
TH_P_FUNDbySTAFF 0.002 -0.032 *** 0.005 ** 0.003 -0.008 ***se 0.015 0.012 0.002 0.002 0.003
STAFF_UNI -0.100 -0.065 0.014 -0.008 -0.006 se 0.155 0.156 0.026 0.025 0.039
PAT_FAM_SIZE 0.096 0.312 *** -0.051 *** -0.016 * 0.068 ***se 0.080 0.074 0.013 0.009 0.015
PAT_NPL_DUM -0.553 ** -1.041 *** 0.184 *** 0.019 -0.203 ***se 0.271 0.241 0.039 0.041 0.057
FIELD CONTROLS YESse
N = 1176 | r²_p = 0.205LR(32): 481.914 | Prob > LR: 0
MARGINAL EFFECTSSME (1) MNE (2) UNI (0) SME (1) MNE (2)
Dist
Regi
on
REFERENCE
Uni
vers
ityPa
tent
YES
REFERENCE