If Star Scientists do not patent:
The Effect of Productivity, Basicness and Impact on The
Decision to Patent in the Academic World
*Mario Calderini, *Chiara Franzoni e **Andrea Vezzulli
*DISPEA, Politecnico di Torino, Turin
**CESPRI, Università Commerciale L. Bocconi, Milan
Academic Patenting.
Rivalry vs. Complementarity Hp
RIVALRY
• the pursuit of market goals may favor a re-arrangement of
academic research agendas in favor of short-term exploitable
trajectories of research
• the rules of market competition may not be compatible with the
social norms of priority and free circulation of knowledge
(Dasgupta and David, 1985; Heller and Eisenberg, 1998)
COMPLEMENTARITY
• feedback from industrial work may be so rich to enable advances
in knowledge or raise new quests for fundamental inquires
(Rosemberg, 1982; Mansfield, 1995)
• Pasteur’s Quadrant: in some areas considerations of use and
fundamental understanding can be pursued at the same time
(Stokes, 1997)
Empirical Evidence
• CROSS-SECTION:
most productive scientists in terms of publications are also more
productive in terms of patents (Agrawal and Henderson, 2002; Stephan
et al., 2007; Van Looy et al., 2004; Carayol, 2007)
• LONGITUDINAL:
academic inventors are likely to experience a (temporary) increase in
number of articles published in coincidence with
the patent event (Azoulay et al., 2006; Breschi et al., 2007).
patents are preceded by a flurry of publications (Azoulay et al., 2007),
although propensity might be decrease for stars (Calderini et al., 2007).
• FIELDS:
Life Sciences, Computer Sciences, Engineering, Physics, Chemistry
OPEN ISSUES:
Quality? How about ENGINEERING vs. SCIENCE?
Sample and Data
SAMPLE
• Names of 1323 Italian publicly-funded scientists in 2001
• Material Sciences
DATA
• Longitudinal data on all publications (ISI) and patents
• (EPO/USPTO) made by each scientist from the age of 23
• 1970 – 2001
• 20,856 scientific papers published
• 941 journals: Impact Factor (JCR) and Level (Chi/research
report)
• 305 patents assigned to academic inventors
Politecnico di Torino
# Inventors and # Patents per type of assignee
• 83–87% patents (accounting for 80-81% inventors) was assigned to a
firm (academic privilege)
• “serial inventors”
(1)
(2)
# Inventors
with at least
one patent in
the category
# Patents
(1)
# Inventors
with patents
only in the
category
# Patents
(2)
Inventor
9
9
6
6
research institution
24
36
13
19
Firm
106
252
95
219
research institution & firm
4
5
4
5
inventor & firm
1
1
1
1
131
303
119
250
Assignee type
Grand Total
Politecnico di Torino
Variables: Productivity, Basicness, Impact
• PRODUCTIVITY: 3-years moving average of the number of
articles published by each individual
L
 articles
l 0
i ( t l )
L 1
• BASICNESS: 3-years moving average of the rank (Level) of the
journals where the individual published
Pit
 level
p 1
itp
L
 articles
l 0
i ( t 1)
• IMPACT: 3-years moving average of the Impact Factor of the
journals where the individual published
Pit
 impact _ factor
itp
p 1
L
 articles
Politecnico di Torino
l 0
i ( t 1)
PRODUCTIVITY: 3-years moving average of the
number of articles published by each individual
8
7
Productivity
6
5
4
3
2
1
0
1
3
5
7
9
11 13
15 17
19 21
23 25 27
29 31
33 35
years of career
mean
Politecnico di Torino
st.dev
median
L
 articles
l 0
L 1
i ( t l )
Pit
 level
BASICNESS: 3-years moving average of the rank
(Level) of the journals where the individual published  articles
p 1
itp
L
l 0
3.5
3
Basicness
2.5
2
1.5
1
0.5
0
1
3
5
7
9
11 13 15 17 19 21 23 25 27 29 31 33 35
years of career
mean
Politecnico di Torino
st.dev
median
i ( t 1)
IMPACT: 3-years moving average of the Impact
Factor of the journals where the individual published
Pit
 impact _ factor
itp
p 1
L
 articles
l 0
1.8
1.6
1.4
Impact
1.2
1
0.8
0.6
0.4
0.2
0
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35
years of career
mean
Politecnico di Torino
st.dev
median
i ( t 1)
Model estimate
Proportional Hazard assumption (hp: all individuals have
identical shape of hazard).
i (t )  0 (t ) exp( xi (t ))
Estimate by Partial Likelihood method (Cox, 1972), which avoids
imposing a specific distribution for T (baseline cancels out).
i (t ) 0 (t ) exp( xi (t )) exp( xi  )


 j (t ) 0 (t ) exp( x j (t )) exp( x j  )
Politecnico di Torino
Results: all publication indicators have a curvilinear effect on
the probability of experiencing an event
Dep. Variable
(1)
(2)
(3)
(4)
(5)
(6)
All patents
(131 failures)
Firm-Assigned Patents
(106 failures)
All patents
(131 failures)
Firm-Assigned Patents
(106 failures)
All patents
(131 failures)
Firm-Assigned Patents
(106 failures)
Coeff.
St. Error
Coeff.
St. Error
Coeff.
St. Error
Coeff.
St. Error
Coeff.
St. Error
Coeff.
St. Error
gender
0.460
(0.213) **
0.417
(0.235) *
0.467
(0.214) **
0.436
(0.236) *
0.452
(0.213) **
0.409
(0.235) *
exptto
0.010
(0.004) ***
0.014
(0.004) ***
0.011
(0.004) ***
0.014
(0.004) ***
0.010
(0.004) ***
0.014
(0.004) ***
instdim
-4.0 e-04 (2.2 e-04) *
-4.7 e-04 (2.4 e-04) **
productivity
0.219
(0.090) **
0.264
(0.099) ***
productivity^2
-0.019
(0.009) **
-0.022
(0.010) **
-4.1 e-04 (2.2 e-04) *
-4.9 e-04 (2.3 e-04) **
basicness
0.758
(0.247) ***
1.070
(0.271) ***
basicness^2
-0.170
(0.063) ***
-0.248
(0.069) ***
-4.1 e-04
(2.2 e-04) *
-4.9 e-04 (2.4 e-04) **
impact
0.365
(0.176) **
0.400
(0.186) **
impact^2
-0.065
(0.043)
-0.060
(0.043)
Obs.
19459
19806
19459
19806
19459
19806
Log likelihood
-869.551
-698.235
-867.624
-694.204
-870.290
-699.332
Prob > chi2
0.004***
0.001***
0.001***
0.000***
0.007***
0.001***
*p≤0.1,**p≤0.05,***p≤0.001
Politecnico di Torino
Results: all publication indicators have a curvilinear effect on
the probability of experiencing an event
Indep var: i (t)
Obs. 19459 (131 failures)
productivity
0.219
(0.090) **
- 0.019
(0.009) **
0.758
(0.247) ***
- 0.170
(0.063) ***
0.365
(0.176) **
productivity^2
basicness
basicness^2
impact
impact^2
- 0.065
Politecnico di Torino
(0.043)
Results: publication&basicness and publiation&impact have a
threshold effect on the probability of experiencing an event
Politecnico di Torino
Results:
Effect of Productivity & Basicness and Productivity & Impact
Indep var: i (t)
Obs. 19459 (131 failures)
productivity (+45%)
0.374
(0.141)
***
basicness (+18%)
0.163
(0.069)
**
prod x basic (-11%)
- 0.118
(0.045)
***
productivity (+18%)
0.166
(0.066)
**
impact (+20%)
0.179
(0.074)
**
- 0.082
(0.034)
**
prod x impact (-8%)
Politecnico di Torino
Productivity_Basicness and Productivity_Impact Effects
Politecnico di Torino
Curvilinear effects
BASICNESS
IMPACT
Threshold 3.49 – 77th centile
10
Threshold 2.34 68th centile
3
mean basicness - 1sd (0.08)
mean basicness (1.78)
mean basicness + 1sd (3.49)
9
mean impact - 1sd (0)
mean impact (1.03)
mean impact + 1sd (2.34)
2.5
8
7
hazard rate
hazard rate
2
6
5
4
1.5
1
3
2
0.5
1
0
0
0
0.4
0.8
1.2 1.6
2
2.4 2.8
3.2 3.6
4
4.4 4.8
productivity
Politecnico di Torino
5.2
5.6 6
0
0.4 0.8
1.2 1.6
2
2.4 2.8
3.2 3.6 4
productivity
4.4 4.8
5.2 5.6 6
Other results
.
Male gender: +140% hazard, but not significant for restricted event of
patenting with a firm.
.
No time/cohort effect: probability to patent has not changed over time.
.
Experience of TTOs increases the hazard to patent.
.
Probability to patent is higher in low-industry environments.
.
Probability to patent with firms decreases with the size of institutions.
.
Estimates on the restricted event to patent with a firm confirm all curvilinear
effects.
Politecnico di Torino
Restricted event to patent with a firm:
all curvilinear effects hold
Indep var: i (t)
(106 failures)
productivity
productivity^2
basicness
basicness^2
impact
0.264
(0.099) ***
- 0.022
(0.010) **
1.070
(0.271) ***
- 0.248
(0.069) ***
0.400
impact^2
- 0.060
Politecnico di Torino
(0.186) **
(0.043)
Restricted event to patent with a firm.
All results hold. Effects increse in magnitudo.
Indep var: i (t)
(106 failures)
Productivity (+53%)
0.427
(0.146)
***
basicness (+20%)
0.181
(0.077)
**
- 0.131
(0.048)
***
productivity (+20%)
0.184
(0.071)
**
Impact (+23%)
0.209
(0.076)
***
- 0.086
(0.034)
**
prod X basicness (-12%)
prod x impact (-8%)
Politecnico di Torino
Conclusions
.
Performances of scientists are a strong predictor of the likelihood to patent.
.
All bibliometric indicators had a curvilinear effect: are there different career
trajectories?
i) low to medium levels of the indicators: any increase in performances increases
the probability to patent:
(e.g. higher productivity=more results to exploit; higher impact=higher
reputation&visibility; higher level=more pervasive results)
ii) high levels of the indicators: any increase in performances decreases the
probability to patent:
(e.g. higher productivity, higher impact, higher basicness= more funds for
untargeted research)
Strength of those effects may depend on: national system of research funding,
technological regimes, type of firms in the region.
Politecnico di Torino
Discipline counts? Research Hypothesis
• not all disciplines earn equal benefits from serving
practical ends.
• Whereas science is aimed at the understanding of
phenomena, engineering is applied in scope, i.e.
aims to solve problems of industrial (practical)
relevance, although by means of a rigorous scientific
method (see Walter G. Vincenti, 1990).
• NB: Here “applied” is used in its epistemological, rather
than hierarchical meaning. Investigation is scoped to
problems, but the process of knowledge creation may
not necessarily be deductive (from basic disciplines), as
the conventional wisdom suggests.
• HP: working on practical problems such as those posed
by inventing a new functional tool can be in principle
more fertile of ideas for engineering than for science.
Dataset: Patents
• 83–87% patents (80% inventors) was assigned to a firm
•“serial inventors”
sc_field
Cumpat_2001 Std. Dev. Freq.
CHEMISTRY
0.224
0.941
917
ENGINEERING
0.298
1.126
309
PHYSICS
0.143
0.550
35
OTHER
0.000
0.000
13
Total
0.237
0.977
1274
• Kruskal-Wallis Test confirms equality of populations for
total patents invented in the overall observation period
Dataset:
Chemists vs. Engineers
The majority of our materials scientists was a chemist or an
engineer of materials. We run separate analysis for subgroups.
n.
917
309
35
13
1274
average
age
35.687
35.482
35.032
35.394
35.614
pat
CHEMISTRY
0.014
ENGINEERING
0.019
PHYSICS
0.008
OTHER
0.000
Total
0.015
Chi-sq with ties
2.627
7.934**
Test: Equality of populations (Kruskal-Wallis test)
cumpat publ_m publbas4
ifac
0.224
0.298
0.143
0.000
0.237
2.367
1.166
2.676
1.884
2.258
2.191
1.457
0.771
1.014
0.327
1.267
0.537
0.582
0.490
0.908
0.712
653.91*** 8616.1*** 3427.7***
Modeling
dependent Variables:
(3 models)
A. QUANTITY number of publications
B. BASICNESS number of basic publications
(IpIQ basicness index=4)
C. IMPACT impact factor
Independent Variables:
Controls:
postpat dummy=1 if invented in previous year
gender, region of affiliation, seniority, experience
of TTO, field, coauthorship)
PROBLEMS IN DATA TREATMENT
1.
2.
3.
4.
Endogeneity > Inverse prob. of treatment weights (Azoulay et al., 2006;
Breschi et al., 2006)
A and B are positive integers with excess zeros > Zero inflated Negbin
C can be measured only when publications are not zero (left truncation) >
Heckman selection equation
Patterns of publications are Subfield-specific. Consequently, each indicator in
was normalized by subfield in the multivariate analysis.
Analysis .A: Publications
QUANTITY:1. count of publications > Zero Infl NegBin
2. log[publications+1] > OLS Fixed Effects
3. as in 2, but publications are weighted by
coauthors
Coefficients estimated for postpat (dummy=1 if author patented in the previous year
Comparison of 3 alternative model estimates
ZINB
(Dep var:
postpat)
OLS_FE
(Dep var: Lpubl)
OLS_FE weighted
for co-authors (Dep
var: lpubl)
ALL
0.018
0.086**
0.037*
ENGINEERS
0.669**
0.278*
0.542
-0.216**
-0.027**
CHEMISTS
-0.329***
Analysis B: Number of basic publications (Level 4
IpIQ journals)
BASICNESS:1. after patent dummy (postpat) > Zero Infl NegBin
2. log[publbas+1] (lpublbas4) > OLS Fixed Effects
3. as in 2, but basic publs are weighted by coauth.
Coefficients estimated for postpat (dummy=1 if author patented in the previous year
Comparison of 3 alternative model estimates
ZINB
(Dep var:
publbas4)
OLS_FE
(Dep var:
Lpublbas4)
OLS_FE
weighted for
co-authors
(Dep var:
lpublbas4)
ALL
-0.090
-0.021
-0.019
ENGINEERS
-0.360
-0.004
-0.015
CHEMISTS
-0.451**
-0.072**
-0.372***
Analysis C: Impact (Journal Impact Factor)
IMPACT:
standardized Impact Factor (stdifac) :
[(IF-mean(IF)/std.dev(IF)] > Heckman (postpat)
Coefficient estimate for postpat (dummy=1 if author patented
in the previous year)
Inverse Probability of Treatment Weighted
HECKMAN_ML
(Dep var:
postpat)
ALL
0.159***
ENGINEERS
0.281**
CHEMISTS
0.080
Heckman selection equation. Standardized Impact Factor,
conditional to having made at least one
publications(accounts for left truncation at zero)
Conclusions
Our estimate of the post-patent productivity, impact and
basicness of publications of a sample of Italian Material
Scientists showed that:
• In the overall sample, productivity is not affected (or slightly
positively affected) by patenting
• When separated into subfields,
1. Engineers experience an increase of publications after patenting
2. Chemists experience a decrease of publications after patenting
3. Engineers experience an increase of Impact Factor and hold basic
publications unchanged.
4. Chemists experience a decrease of basic journal publications, and
hold Impact Factor unchanged.
5. The increase of IF occurs at negative marginal return (neutralized
after the 4th patent), but this effect is unlikely to occur in practice
Scarica

Professor Mario Calderini - If Star Scientists do not Patent