Research Policy 31 (2002) 399–418

Academic patent quality and quantity before and after the

Bayh–Dole act in the United States
David C. Mowery a , Arvids A. Ziedonis b,∗
aHaas School of Business, University of California, Berkeley, CA 94720-1900, USA
b The Wharton School, University of Pennsylvania, Philadelphia, PA 19104-6370, USA
Received 12 May 2000; received in revised form 28 June 2000; accepted 26 February 2001

Abstract
This paper summarizes the results of empirical analyses of data on the characteristics of the pre- and post-1980 patents
of three leading US academic patenters—the University of California, Stanford University, and Columbia University. We
complement the analysis of these institutions with an analysis of the characteristics of the patents issued to all US uni-
versities before and after 1980. Our analysis suggests that the effects of the Bayh–Dole act on the content of academic
research and patenting at Stanford and the University of California were modest. The most significant change in the
content of research at these universities, one associated with increased patenting and licensing at both universities be-
fore and after 1980, was the rise of biomedical research and inventive activity, but Bayh–Dole had little to do with this
growth. Both UC and Stanford University administrators intensified their efforts to market faculty inventions in the wake
of Bayh–Dole. This enlargement of the pool of marketed inventions appears to have reduced the commercial “yield” of
this population at both universities. But we find no decline in the “importance” or “generality” of the post-1980 patents
of these two universities. The analysis of overall US university patenting suggests that the patents issued to institutions
that entered into patenting and licensing after the effective date of the Bayh–Dole act are indeed less important and less
general than the patents issued before and after 1980 to US universities with longer experience in patenting. Inexperienced
academic patenters appear to have obtained patents that proved to be less significant (in terms of the rate and breadth
of their subsequent citations) than those issuing to more experienced university patenters. Bayh–Dole’s effects on entry
therefore may be as important as any effects of the act on the internal “research culture” of US universities in explaining
any decline in the importance and generality of US academic patents after 1980. © 2002 Elsevier Science B.V. All rights
reserved.
Keywords: Patents US universities; Bayh–Dole act

1. Introduction act has contributed to the rapid emergence of new

high-technology firms and high rates of growth in the
The Bayh–Dole act of 1980 has been widely cited US economy during the 1990s. 1 An earlier paper
as an important initiative in postwar US technology
policy, and some observers have asserted that the 1 See the recent staff report of the US Congressional Joint

Economic Committee, which stated that “One important reform

which helped spur quick adoption was the Bayh–Dole act of
∗ Corresponding author. Tel.: +1-215-898-7519. 1980, which gave universities greater incentives to commercialize
E-mail address: ziedonisa@wharton.upenn.edu (A.A. Ziedonis). technology. The act allowed universities to patent the results of

0048-7333/02/$ – see front matter © 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 0 4 8 - 7 3 3 3 ( 0 1 ) 0 0 1 1 6 - 0
400 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
(Mowery et al., 2001) examined the effects of the
Bayh–Dole act on the patenting and licensing activ-
ities of Columbia University, the University of Cal-
ifornia, and Stanford University. Comparison of data
from these three universities allowed us to consider
the effects of these new federal policies on university
patenting and licensing in a “before and after” analy-
sis of the behavior of two research universities active
in patenting and licensing before 1980 (the Univer-
sity of California and Stanford), as well as that of a
post-1980 entrant (Columbia) that now ranks among
the leading institutional recipients of licensing and
royalty income (AUTM, 1994).
The paper by Mowery et al. (2001) concluded that
the effects of Bayh–Dole on the content of academic
research at Stanford and the University of California
were modest. The most significant change in the con-
tent of research at these universities, one associated
with increased patenting and licensing at both univer-
sities before and after 1980, was the rise of biomed-
ical research and inventive activity, but Bayh–Dole
had little to do with this growth. Indeed, the rise in
biomedical research and inventions in both of these
universities predates the passage of Bayh–Dole.
This paper extends our previous analysis in an
examination of the effects of the Bayh–Dole act on
universities’ patenting behavior and on the entry of in-
experienced academic patenters into this activity after
1980. We analyze the characteristics (“importance”
and “generality”, which we define below) of the pre-
and post-1980 patents of these three universities, as
well as the characteristics of the patents issued to all
US universities. We find evidence that both UC and
Stanford University administrators intensified their
efforts to market faculty inventions in the wake of
Bayh–Dole, expanding the pool of university inven-
tions for which patent applications were made and
licensees sought. This enlargement of the pool of mar-
keted inventions appears to have reduced the average
“yield” of this population at both universities. But we
find no decline in the “importance” or “generality” of
the post-1980 patents of these two universities.
The stability in the comparative importance and
generality of the post-Bayh–Dole patents of these two
federally-funded research and license the resulting technology to
businesses and other entities” (Joint Economic Committee US
Congress, 1999, p. 31).
universities with considerable pre-1980 experience
in patenting differs somewhat from the findings of
Henderson et al. (1995, 1998a). These scholars found
that the importance and generality of overall US
academic patents declined after 1980 (relative to a
sample of “control” patents), as did the importance
and generality of the patents assigned to the most
experienced US academic patenters. These differ-
ences motivated the second portion of our analysis,
which compared the characteristics of the post-1980
patents assigned to US universities with considerable
pre-1980 patenting experience and those assigned
to firms with little or no pre-1980 patenting expe-
rience. Our analysis probed the hypothesis that any
post-1980 declines in the importance and generality
of overall US university patents reflected the entry
of inexperienced academic patenters after 1980. Al-
though Henderson et al. (1998a) suggest that declines
after 1980 in the importance and generality of overall
US academic patents may reflect increased patenting
by smaller institutions, they are not able to control
explicitly for the effects of entry by such institutions
into patenting activities after 1980.
Our analysis of the effects of entry by previ-
ously inactive academic patenters on the importance
and generality of US academic patents is intended
to distinguish between two hypothesized effects of
Bayh–Dole on academic research and patenting. Did
the expanded post-1980 efforts by US universities to
promote patenting and licensing of faculty inventions,
especially when faculty share in the financial returns
to these licenses, skew the content and character of
university research to favor more “applied” research
activities? Parallel declines in the importance and gen-
erality of the post-1980 patents of both “incumbent”
and “entrant” universities would indicate that the
Bayh–Dole act affected the incentives of academic
researchers and administrators to disclose and patent
inventions of lower importance and generality in US
academic institutions with and without experience in
patenting.
If changes in the characteristics of the overall popu-
lation of US academic patents after 1980 reflect entry
by less experienced patenters, however, a different in-
terpretation of the effects of Bayh–Dole is plausible.
For example, new entrants that initially patent a broad
cross-section of faculty discoveries may accumulate a
patent portfolio of limited importance and generality
 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
(some anecdotal evidence supports this characteriza-
tion). Over time, as they learn the complexities of
protecting and marketing intellectual property and
become more selective in their patenting, the gaps
between the characteristics of their patents and those
of the “high-intensity” incumbents could narrow
somewhat.
The first interpretation of Bayh–Dole’s effects em-
phasizes lasting changes in incentives and behavior
throughout US universities, while the other views the
1980s as a period of learning and adjustment to a
new incentive environment by organizations inexpe-
rienced in patenting and licensing. Needless to say,
these explanations are not mutually exclusive, and
the development of US academic patenting during
the 1980s may reflect both effects. But this empiri-
cal analysis allows for an assessment of the relative
strength of the two effects.
Immediately below, we examine the characteristics
of the invention disclosures, patents and license agree-
ments of the University of California and Stanford
University before and after the Bayh–Dole act, in an
analysis of the effects of Bayh–Dole on the intensity
and yield of these experienced academic patenters’
technology marketing efforts. This analysis is fol-
lowed by a statistical examination of the “importance”
and “generality” of UC and Stanford patents before
and after Bayh–Dole, an analysis that we extend to
include the patents assigned to Columbia University,
a post-1980 entrant into patenting and licensing ac-
tivities. We also analyze the characteristics of the
patents assigned to a more comprehensive sample of
US universities, in order to assess the effects on the
overall importance and generality of US academic
patents of the entry of less experienced academic
patenters after 1980. The final section considers the
implications of our findings, which indicate that less
experienced academic patenters did indeed receive
less importance or general patents in the aftermath
of Bayh–Dole. The initial effects of the Bayh–Dole
act on entry thus may underpin any observed decline
in the importance and generality of US academic
patents after 1980. Moreover, the passage of time and
the development of greater knowledge of patenting
and increased selectivity in the patenting practices
of these entrants may reduce any gap in importance
and generality between their patents and those of the
experienced pre-1980 academic patenters.
401
2. The effects of Bayh–Dole on the content of UC
and Stanford disclosures, patenting, and licensing
2.1. Introduction
Both the University of California system and
Stanford University established offices to promote
the patenting and licensing of faculty inventions well
before the passage of the Bayh–Dole act. In 1963, the
UC Board of Regents adopted a policy stating that all
“Members of the faculties and employees shall make
appropriate reports of any inventions and licenses
they have conceived or developed to the Board of
Patents”. 2 In 1976, responsibility for patent policy
was transferred from the General Counsel to the Of-
fice of the President of the University of California,
and in 1991 the Patent Office was renamed the Office
of Technology Transfer (OTT).
Stanford University’s Office of Technology Li-
censing (OTL) was established in 1970, and Stanford
was active in patenting and licensing throughout
the 1970s. Disclosure by faculty of inventions and
their management by Stanford’s OTL was optional
for most of OTLs first quarter-century, but in 1994
Stanford changed its policy toward faculty inventions
in two important aspects. First, assignment of title
to the university of inventions “. . . developed using
university resources. . . ” was made mandatory. Sec-
ond, the university established a policy under which
“Copyright to software developed for university pur-
poses in the course of employment, or as part of
either a sponsored project or an unsponsored project
specifically supported by university funds, belongs
to the university” (“Office of Technology Licensing
2 The Board was a committee of UC faculty and administra-
tors charged with oversight of the Patent Office. As revised in
1973, the “University Policy Regarding Patents” states that “An
agreement to assign inventions and patents to The Regents of the
University of California, except those resulting from permissible
consulting activities without use of University facilities, shall be
mandatory for all employees, academic and non-academic”. The
policy statement goes on to emphasize that “The Regents is [sic]
averse to seeking protective patents and will not seek such patents
unless the discoverer or inventor can demonstrate that the secur-
ing of the patent is important to the University”. This latter sen-
timent notwithstanding, UC administrators were actively seeking
patent protection for faculty inventions by the mid-1970s, as the
historical data of the Office of Technology Transfer show.
402 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
Guidelines for Software Distribution”, 11/17/94). 3
The Stanford data contain many more faculty software
inventions than do the UC data, reflecting the fact
that the “mandatory disclosure” policy of the Univer-
sity of California did not cover software, which was
deemed to be copyrightable rather than patentable
intellectual property.
Both UC and Stanford generated significant
numbers of disclosures, patents, and licenses be-
fore the passage of Bayh–Dole. Nevertheless, these
universities’ patents and licenses grew significantly af-
ter 1980. 4 The next section presents comparative data
on the “intensity” and “yield” of these universities’
technology marketing activities before and after 1980.
These data are relevant to an assessment of (1) the ef-
fects of Bayh–Dole on the characteristics of academic
invention disclosures; and (2) the existence and extent
of any “diminishing returns” to these universities’
increased promotional and marketing activities in the
years after Bayh–Dole.
2.2. The “yield” of patenting and licensing at UC
and Stanford, 1975–1979 and 1984–1988
We first use the data on UC and Stanford disclo-
sures, patenting, and licensing to discuss the effects
of Bayh–Dole on the technology marketing efforts
of these universities. The data in Table 1 provide
3
Reflecting faculty sensitivity over assignment to the Univer-
sity of all ownership of all copyrighted material produced under
University sponsorship, Stanford’s OTL explicitly exempted own-
ership of “. . . books, articles, . . . popular nonfiction, novels, po-
ems, musical compositions, or other works of artistic imagination
which are not institutions works” from the policy governing soft-
ware (“Copyrightable Works and Licensing at Stanford”, Stanford
University Office of Technology Licensing, Spring, 1994, p. 1).
4 For the University of California, the average annual number
of “invention disclosures” during 1984–1988, following passage
of the Bayh–Dole act, is almost 237, well above the average
level (140 annual disclosures) for the 1975–1979 period. The
average annual number of disclosures to Stanford’s Office of
Technology Licensing increased from 74 during 1975–1979, prior
to Bayh–Dole, to 149 during 1984–1988. The data in Table 2 show
97 patents issued to UC during 1975–1980 that were based on
disclosures made during the 1970s, slightly more than one-third the
number of patents issued to UC during 1981–1992 that were based
on post-1980 disclosures. The number of “pre-Bayh–Dole” patents
issued to Stanford during 1975–1980 is 110, slightly more than
one-fourth of the 407 patents issued to Stanford during 1981–1992
that were based on post-1980 disclosures.
indicators of change before (1975–1979) and after
(1984–1988) Bayh–Dole in (1) the intensity of these
institutions’ technology marketing efforts, reflected in
their propensities to patent or license faculty inven-
tion disclosures and (2) the yield of these marketing
efforts, based on such indicators as the share of patent
applications that result in issued patents, the share
of issued patents that are licensed, and the share of
licenses that generate positive income. 5
The first row in Table 1 displays “before and after”
data on the share of invention disclosures resulting in
issued patents, which scarcely changes between the
two periods at UC and increases from nearly 15%
to slightly more than 25% (excluding software, from
15.5 to 29.8%) at Stanford. The second row of Table 1
shows that the UC system also increased its patent ap-
plication rates in the wake of Bayh–Dole-the share of
disclosures that generate patent applications rose from
24% to slightly more than 31%. But this increased ten-
dency to file for patents was associated with a decline
in the “yield” of these efforts, as the share of patent
applications resulting in issued patents dropped from
more than 62% in the 1975–1979 period to less than
44% in 1984–1988 (row 5 of Table 1). These changes
in marketing intensity and yield suggest that UC ad-
ministrators intensified their efforts to protect and pro-
mote faculty inventions after the Bayh–Dole act; these
intensified efforts nevertheless appear to have pro-
duced a lower “yield”, as measured by the share of
applications resulting in issued patents. The increase
in the share of Stanford disclosures yielding issued
patents after 1980 is more difficult to interpret, since
we lack comparable data on patent applications. But it
indicates no decline in the “patentability” of Stanford
invention disclosures, and is broadly consistent with
an overall increase in patenting propensity.
Rows 3, 4, 6, and 7 in Table 1 provide additional
evidence of change in the “intensity” and “yield” of
the UC and Stanford technology marketing efforts af-
ter Bayh–Dole. The share of UC disclosures resulting
in licenses (row 3) increased from 5.6% in 1975–1979
to 12.6% in 1984–1988, while the share of UC dis-
closures resulting in licenses that generated positive
royalties (row 4) grew from 4.6% in 1975–1979 to
5 Data on Stanford patent applications are unavailable, and we
therefore are able to compute measures using patent applications
only for the University of California.
 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418 403

Table 1
University of California and Stanford University invention disclosures, patents, and licenses, 1975–1979 and 1984–1988 (8-year “trailing
window”)a
Indicator University of Stanford (including Stanford (excluding
California software) software)
1975–1979 1984–1988 1975–1979 1984–1988 1975–1979 1984–1988

Marketing intensity indicators

(Disclosures resulting in issued patents)/ 20.2 21.9 14.9 25.1 15.5 29.8
(invention disclosures)
(Disclosures generating patent applications)/ 24.0 31.2 NA NA NA NA
(invention disclosures)
(Disclosures licensed)/(invention disclosures) 5.6 12.6 13.0 36.6 12.8 34.3
(Disclosures generating licenses with income > 0)/ 4.6 5.0 9.8 22.0 9.5 17.9
(invention disclosures)
Marketing yield indicators
(Patents issued)/(patent applications) 62.1 43.6 NA NA NA NA
(Patents licensed)/(patents issued) 25.1 35.5 62.7 63.7 62.7 63.7
(Licenses with income > 0)/(licenses) 84.1 47.1 91.4 90.7 90.0 76.2
a All values in percentage.

5% in 1984–1988. The share of Stanford disclosures
that were licensed, however, nearly tripled between
1975–1979 and 1984–1988. The share of Stanford dis-
closures yielding positive licensing income more than
doubled (growing from 9.8 to 22%) when software in-
vention disclosures are included and nearly doubled
when these are excluded (increasing from 9.5 to almost
18%). 6 At both UC and Stanford, the proportionate
increase in the share of disclosures that were licensed
after 1980 (a measure of the intensity of institutional
technology marketing efforts) exceeded the propor-
tionate increase in the share of disclosures yielding
positive licensing income (an indicator of the yield of
these marketing efforts).
Two additional measures of the yield of these
institutions’ technology marketing efforts, the share
of issued patents that were licensed (row 6 of Table
1) and the share of these licenses yielding positive
income (row 7 of Table 1) also suggest some decline
in the yield of these efforts after the Bayh–Dole act.
The share of UC patents that were licensed grew from
slightly more than 25 to more than 35%, but the share
of these licenses yielding positive income dropped
by almost half, from roughly 84 to 47%. The “before
6 Since the UC disclosure and licensing data largely exclude
software inventions, the most nearly comparable Stanford data are
those excluding software inventions and licenses.
and after” trends at Stanford are similarly mixed. The
share of issued Stanford patents that were licensed
(an indicator that is comparable to the UC data only
for the far right columns that exclude software in-
ventions) is essentially unchanged, increasing from
almost 63 to almost 64%. The share of Stanford li-
censes yielding positive income 7 dropped from 90%
in 1975–1979 to roughly 76% in 1984–1988 when
software inventions are excluded (a more meaningful
basis for comparison with the UC licensing effort). 8
Overall, these data do not reveal significant shifts
in the character of the inventions disclosed to UC or
Stanford administrators for possible patenting after
Bayh–Dole’s passage, which (subject to the caveats
noted earlier) suggests little change in the character
of faculty research disclosures after 1980. At both
universities, however, these data suggest an intensi-
fication in technology marketing activities after the
7 The data on Stanford licenses are incomplete, as was noted
earlier. Moreover, the licensing income data are reported on a
fiscal year basis, and license agreements on a calendar year basis,
making it difficult to reconcile these data. We have attempted to
adjust the licensing income and agreements data to address this
incompatibility, but the data for Stanford in the last line of Table 1
should be treated very cautiously.
8 This share is essentially stable, however, when software inven-
tions are included (the Stanford software licensing activities in-
cluded a large of academic “site licenses”, each of which produced
a small but positive licensing income).
404 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
passage of Bayh–Dole. The post-Bayh–Dole period
of invention disclosures also is associated with more
extensive licensing of the patents resulting from these
universities’ disclosures.
The more extensive licensing efforts of the
post-1980 period also appear to have produced a de-
cline in the “yield” of these licenses at both UC and
Stanford, measured as the share of licenses yielding
positive royalties. But these indicators of decline in
the average productivity of the marketing efforts of
the UC and Stanford technology licensing offices
need not imply inefficient or economically irrational
behavior-after all, it is the marginal, rather than the av-
erage, returns to licensing activities that are most im-
portant in evaluating the returns to these institutions’
licensing activities. Moreover, our measure of the
“yield” of these universities’ licensing activities does
not capture the size of the revenue streams associated
with the average or the marginal license before and
after Bayh–Dole. 9
3. The “importance” and “generality” of
academic patents from “incumbents” and
“entrants”, 1975–1992
3.1. Introduction
These data on the intensity and yield of the technol-
ogy marketing activities at UC and Stanford suggest
that in the first decade after the passage of the act, these
“incumbent” universities’ technology transfer efforts
intensified, with a concomitant decline in their yield.
But this discussion does not address an important issue
raised in the work of Henderson et al. (1995, 1998a),
who suggest that these intensified post-Bayh–Dole ef-
forts to market faculty inventions were associated with
the issue to US universities of patents that were less
“important” and less “general”, based on the patterns
of citations to these patents. As Henderson et al. (1995)
argue in an earlier paper on this topic, these changes
9 For example, average income per license may have increased
in the second period, although the skewed distribution of the
licensing income of both the Stanford and UC technology transfer
offices means that any such changes are likely to be small. See
Sampat and Ziedonis (2000) for an analysis of the relationship
between the level of licensing revenues at UC and Columbia and
the characteristics of these universities’ licensed patents.
in the characteristics of US universities’ post-1980
patents “. . . may reflect changing motives within the
university research community or shifts in the focus of
university research from basic to more applied work”
(p. 1).
In order to address this issue, we analyzed the
“importance” and “generality” of UC and Stanford
patents before and after Bayh–Dole and compared
these characteristics of UC, Stanford, and Columbia
patents after 1980. We also analyzed the character-
istics of all US academic patents from the pre- and
post-Bayh–Dole periods in order to test the effects
on academic patenting of entry into patenting activi-
ties after 1980 by universities with limited patenting
experience. Our analysis follows that of Henderson
et al. (1995, 1998a) and Trajtenberg et al. (1997)
closely, but we employ a slightly different “control
population” of patents (see ahead).
3.2. Comparing UC, Stanford, and Columbia
University patents
Our “before and after” analysis of UC and Stanford
patents uses the year in which the invention was first
disclosed as the key datum in categorizing faculty dis-
closures and any associated patents as falling into the
“pre-Bayh–Dole” or “post-Bayh–Dole” eras. We fur-
ther categorize patented disclosures by the date of the
patent application. Our data for UC and Stanford thus
contain (1) disclosures and patents for which the dis-
closures and patent application occurred during and
after 1970 and the issue of the patent occurred before
1981 (the year during which the Bayh–Dole act went
into effect); (2) disclosures and patent applications that
occurred during or after 1981 and before 1993, and
(3) inventions disclosed after 1974 and prior to 1981
for which patent applications were filed during or af-
ter 1981 and before 1993. 10
We also include Columbia University patents in
our analysis of the post-Bayh–Dole era. For much
10 This third subset of our data is intended to separate those fac-
ulty disclosures that occurred before Bayh–Dole could affect the
content of academic research but whose commercial promotion by
universities (i.e. through filing patent applications and seeking li-
censees) occurred after the effective date of the act. Unfortunately,
this sample of patents is quite small, reducing the power and sig-
nificance of our tests of statistical significance for any differences
among the characteristics of the three subsets of patents.
 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
of the pre-1980 period, Columbia University had no
formal policy for the patenting or administration of
faculty inventions, beyond a statement (rescinded in
1975) prohibiting the patenting of medical inventions.
Columbia’s patent policy was significantly altered af-
ter the passage of Bayh–Dole. The new policy, which
took effect on July 1, 1981 (the effective date of
Bayh–Dole), reserved patent rights for Columbia and
shared royalties with the inventor and department. In
1984, a new policy statement mandated that faculty
members disclose to the university any potentially
patentable inventions developed with university re-
sources. In 1989, Columbia’s policy on asserting its
rights to faculty inventions created with university re-
sources was extended to cover software. Although we
define Columbia as an “entrant” academic patenter,
reflecting the fact that this university developed an
active patenting and licensing policy only after 1980,
in fact Columbia did accumulate a modest portfolio of
fewer than 10 patents during the 1975–1980 period.
Despite its status as an “entrant” into patenting and li-
censing, by the 1990s Columbia University was (along
with Stanford and the UC system) among the leading
US academic recipients of licensing revenues. 11
Henderson and colleagues use patent-citation mea-
sures in their analysis of university patents. When the
US Patent and Trademark Office grants a patent, the
granting officer includes a list of all previous patents
on which the granted patent is based. This list is made
public as part of the publication of the patent at the
time it issues. The patent officer is aided in compil-
ing a list of previous patents by the patent applicant,
who is legally bound to provide with the application a
list of all patents that constitute relevant “prior art”. 12
Citations of prior patents thus serve as an indicator of
the technological lineage of new patents, much as bib-
liographic citations indicate the intellectual lineage of
academic research.
11 According to the annual report of the Association of Univer-
sity Technology Managers (AUTM), fiscal 1993 gross licensing
revenues for the UC system, Stanford University and Columbia
University amounted to US$ 45.4 million, US$ 31.2 million, and
US$ 21.1 million, respectively (AUTM, 1994, p. 8).
12 In addition to the legal requirement, it is in the applicant’s
interest to be forthcoming in this list because a more complete
description of prior art is likely to reduce the prospects of an inter-
ference being declared during processing of a patent application.
405
Our analysis of UC, Stanford, and Columbia patents
focuses on “forward citations” to these patents, de-
fined as the number of citations received by each patent
following its issue. 13 Citations to patents typically
peak 4–5 years after the date of issue of the cited
patent. As a result, data on citations to more recently
issued patents will be “right-truncated”, i.e. more re-
cent patents will be underrepresented in the citations
data. In order to address this problem, our dataset in-
cludes only citations to patents that occurred within 6
years of the year of issue of the patent, and our sam-
ple includes only patents issued between 1975 and
1992. Our dataset also includes a “control sample” of
non-academic patents for each of these three universi-
ties, spanning the same time period and replicating the
distribution of the UC, Stanford, and Columbia patents
among patent classes. We matched one non-university
patent with each university patent, based on the patent
class and application date of the relevant university
patent. 14 Citations to patents in the control sample
also are restricted to those occurring within 6 years
after the year of issue.
We used the number of citations to a patent dur-
ing the 6 years following its issue as a measure of
the “importance” of the patent, based on the assump-
tion that citations form an index of sorts of the in-
fluence over subsequent inventive activity of the cited
patent. The number of observations for our UC, Stan-
ford, and Columbia samples of patents is displayed
in Table 2 (because Columbia has very few pre-1981
patents, we omit its pre-1981 data). We also divide the
Columbia, UC and Stanford patents into “biomedical”
and “non-biomedical” categories, in order to test for
13 Henderson et al. (1998a) propose a “two-generation” measure
of the number of citations to a given patent, consisting of the
number of citations received by a patent in subsequent issued
patents as well as the number of citations to these citing patents.
Given the limited period of time following the issue of the original
university patent, Henderson, Jaffe, and Trajtenberg were unable
to incorporate the “second-generation” citations in their measure,
and we also exclude such follow-on citations.
14 Our control samples are constructed differently from those of
Henderson et al. (1995, 1998a). Henderson, Jaffe, and Trajtenberg
used as their control sample a one-percent random sample of all
US patents granted during the time period covered by their univer-
sity patent sample. In contrast, our control sample matches each
university patent with a non-university patent from the same patent
class with an application date at or near the date of application
for the university patent.
406 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
Table 2
Number of patents in the UC, Stanford, and Columbia datasets
Overall
UC Stanford
Patent application 1970≤ and <1981; patent 97 110
issued 1975≤ and <1981
Disclosed <1981; patent application 32 30
≥1981 and ≤1992
Disclosed and patent application 245 377
≥1981 and ≤1992
differences in the effects of Bayh–Dole between these
broad technology classes. 15 Clearly, for some cells
of the dataset (e.g. Stanford biomedical patents in the
“disclosed before 1981 and patented during or after
1981” subset), the number of observations is small (far
smaller, for example, than the university patent sam-
ples for individual years in Trajtenberg et al. (1997)).
As such, the standard errors for this subsample are
large, reducing the statistical significance of our tests.
Nevertheless, we believe that measures of the impor-
tance and generality of this category of our patents are
worth computing and analyzing separately, and there-
fore have retained these small subsamples in the re-
sults that follow.
Table 3 displays means and standard errors of the
means for the UC, Stanford, and Columbia patent
samples and their disaggregated subsamples. We also
present means and standard errors for the control
patent samples and subsamples corresponding to each
of the UC, Stanford, and Columbia categories. With-
out exception, the number of citations received in the
6 years following issue by the patents of all three uni-
versities is larger than those received during the iden-
tical period by the “control samples” of patents, for
all time periods and disaggregated categories. There
is no apparent time trend in the absolute magnitude of
differences between the number of citations received
by the Stanford and UC patents and those in the con-
trol sample. With the exception of biomedical patents
in the second and third of our three subperiods, the
15 Our definition of biomedical patent classes is based on Tech-
nology Assessment and Forecast reports published by the USPTO
that identify 3-digit patent classes and sub-classes related to med-
ical and biomedical technologies. For a more detailed description
of this taxonomy, see Ziedonis (2001).
Biomedical Non-biomedical
Columbia UC Stanford Columbia UC Stanford Columbia
– 43 23 – 54 87 –
– 23 10 – 9 20 –
112 154 93 63 91 284 49
mean number of citations received by the Stanford
patents exceeds the mean number of citations received
by UC patents, although these differences frequently
are small. With one exception (Stanford biomedical
patents), UC and Stanford patents also are more heav-
ily cited than Columbia patents during the post-1980
period.
The results of our tests for the significance of the
differences between the mean values of forward cita-
tions to the UC, Stanford, and Columbia patents and
citations to patents in our control samples are dis-
played in the upper panel of Table 4. 16 We find no
evidence of a statistically significant decline in the
importance of Stanford and UC patents, relative to
the control samples, after 1980 (columns 2 and 3 of
Table 4). 17 Tests for the significance of the difference
in mean citations between the Columbia patents and
the Columbia control sample, however, indicate that
the difference is significant only for the overall patent
sample at the 10% level. The absence of significantly
16 In another analysis not reported here, we subdivided the
post-1981 period into three subperiods of equal length, in order to
test for more significant reductions in the importance of Stanford
and UC patents during the late 1980s. Henderson et al. (1998a)
find that the greatest reduction in importance in their sample of
academic patents, especially those assigned to the leading aca-
demic patenters, occurs late in the 1980s). The results of tests
for differences in the means of the university and control patent
samples reveal no significant differences.
17 Interestingly, the corrected results reported in the errata ap-
pendix to Trajtenberg et al. (1997) find that the importance (mea-
sured in terms of the number of forward citations) of academic
patents, relative to those from corporate inventors, declines be-
tween 1975 and 1980, before the Bayh–Dole act. This finding is
broadly consistent with our conclusion that Bayh–Dole per se has
little measurable effect on the relative importance of academic and
non-academic patents.
D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418 407
408 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
Table 4
Tests for significance of differences in means (t-statistics): forward citationsa
Academic vs. non-academic control sample Overall
UC Stanford Columbia UC
Patent application 1970≤ and <1981; 1.15 3.83∗∗
patent issued 1975≤ and <1981
Disclosed <1981; 2.30∗∗ 3.14∗∗
patent application ≥1981 and ≤1992
Disclosed and patent application 4.47∗∗ 6.45∗∗
≥1981 and ≤1992
a Unequal variances assumed.
∗∗ P > 0.05.
∗ P > 0.10.
greater citation rates for Columbia patents could reflect
a less selective approach to patenting during the early
years of its licensing activities by Columbia Univer-
sity, an institution with little experience in patenting,
than UC or Stanford. 18 We return to this issue ahead.
The results of our tests for the significance of dif-
ferences between the mean citation rates of the UC
and Stanford biomedical patents and their respective
non-academic patent control samples indicate that
the higher citation rates for these academic patents
are statistically significant (at the 5% level) only for
the post-1980 period for UC only (the higher number
of citations to UC biomedical patents for the “mid-
dle period” is weakly significant, at the 10% level).
The mean number of forward citations for Stanford
non-biomedical patents significantly exceeds those for
their control samples for all three periods in Table 4,
while the citation frequencies for UC non-biomedical
patents are significantly greater (at the 5% level)
than those in the control sample only once, for the
post-1980 period. 19 The results for Columbia during
18 Any such effect was significant during only the early years of
Columbia’s patenting and licensing activities, since by 1986–1990,
the share of disclosures resulting in issued patents and the share of
disclosures that result in licenses yielding positive royalty income
are fairly similar at Columbia, UC and Stanford (Mowery et al.,
1999, 2001).
19 We also examined the number of UC and Stanford patents
from all three subperiods that yielded zero citations during the 6
years following their issue, in an effort to replicate the analysis
in Henderson et al. (1998a) of “losers”. Here too, we find no
consistent trend. “Zero-cite patents” account for 22 and 18%,
respectively, of the UC and Stanford patents for period 1 in
Tables 2–4; 16 and 0% of the UC and Stanford patents in period
2; and 11 and 8% of the UC and Stanford patents in period 3.
Biomedical Non-Biomedical
Stanford Columbia UC Stanford Columbia
– 0.32 1.56 – 1.12 3.70∗∗ –
– 1.77∗ 1.24 – 1.45 3.02∗∗ –
1.66∗ 2.59∗∗ 1.47 1.40 4.06∗∗ 6.43∗∗ 0.89
the post-1980 period indicate no significant differ-
ences in citation frequencies from the patents in the
Columbia control sample, but there is no evidence that
these Columbia patents are significantly less heavily
cited than those in its control sample.
These results must be interpreted with considerable
caution, in view of the small number of observations
for some time periods. The relatively infrequent signif-
icant differences in “importance” between the univer-
sity and control sample biomedical patents’ citations
is interesting, in view of the importance of biomedi-
cal patents within the patenting and licensing activi-
ties of Stanford and UC before and after Bayh–Dole.
But these data provide no indication of any decline
in the importance of these universities’ patents, rela-
tive to our control samples of non-academic patents,
after the Bayh–Dole act. If anything, these data sug-
gest that the UC and Stanford patents’ relative im-
portance increased, rather than declined, during the
post-Bayh–Dole period. Although these results do in-
dicate that the patents applied for during the 1980s by
Columbia, a university that did not patent significantly
prior to Bayh–Dole, were less “important”, relative
to all non-academic patents, than those of Stanford
and UC during this period, they do not suggest that
Columbia’s patents were significantly less important
than those in its control sample. 20
20 We also tested, but do not report results, for the significance of
year-specific effects on the relative importance of these academic
and non-academic patents, using a negative binomial specification.
We find no significant decline in the importance of these three
universities’ patents relative to non-academic patents during the
post-1980 period.
D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418 409
410 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418

Table 6
Tests for significance of differences in means (t-statistics): “generality”a
Academic vs. non-academic control sample Overall Biomedical Non-Biomedical

UC Stanford Columbia UC Stanford Columbia UC Stanford Columbia

Patent application 1970≤ and <1981; −0.35 2.31∗∗ – 0.198 1.56 – −0.70 1.80∗ –
patent issued 1975≤ and <1981
Disclosed <1981; patent 0.08 1.25 – −0.32 −1.53 – 0.91 2.69∗∗ –
application ≥1981 and ≤1992
Disclosed and patent application 3.23∗∗ 4.83∗∗ 2.49∗∗ 1.61 1.76∗ 1.97∗∗ 3.19∗∗ 4.58∗∗ 1.58
≥1981 and ≤1992
aUnequal variances assumed.
∗ P > 0.10.
∗∗ P > 0.05.

Tables 5 and 6 present data on the “generality” of
UC, Stanford, and Columbia patents before and after
Bayh–Dole. The more widely cited a patent outside
of its “home” patent class, the greater its general-
ity, and arguably, the more significant the advance
in knowledge represented by the patent. Following
Henderson et al. (1995, 1998a), we compute general-
ity as follows:
Ni   applied for and issued before 1981. 21 There is no ev-

Ncitingik 2
Generali = 1 − (1)
Ncitingi
k=1
where for patent i, k is the index of patent classes
and Ni the number of different classes to which the
citing patents belong. This measure can be computed
for only those patents with at least one citation, and
the sample in Tables 5 and 6 accordingly differs
slightly from that in Tables 2–4. As in our mea-
sure of importance, we restrict forward citations to
those occurring during the first 6 years following
the issue of the patent. Higher values of Generali
indicate that a patent is relevant to subsequent in-
ventive activities in a broader range of technological
fields.
As a measure of “dispersion” of patent citations,
Generali is inherently biased. The more citations a
patent receives, the greater the likelihood that these ci-
tations will be spread across a larger number of patent
classes. Moreover, this bias is greater for patents with
a small number of citations. Therefore, we employ the
where Ni is the number of forward citations to patent
i during the first 6 years following the issue of the
patent. All reported results for generality incorporate
this correction for bias.
The data in Table 5 indicate that the mean general-
ity measures for overall UC, Stanford, and Columbia
patents are higher than those for their respective con-
trol sample patents, with the exception of UC patents
idence in Table 5 of any decline in generality (relative
to the control sample of non-academic patents) in the
UC and Stanford patents in the post-Bayh–Dole era.
The differences in mean generality between the overall
UC and Stanford patents and their respective control
samples are statistically significant (at the 5% level)
for the post-1981 period (Table 6). The mean gener-
ality score of the post-1981 Columbia patents also is
significantly higher (at the 5% level of significance)
than the patents in its control sample.
Overall, the results of this analysis of the importance
and generality of Stanford, UC, and Columbia patents
yield conclusions that differ from those of Henderson
et al. (1995, 1998a), who analyzed a larger sample
of US university patents. Importantly, these scholars
also found a decline in the importance and generality
of the patents assigned to the universities that oc-
cupied the top decile of the post-1980 patent distri-
bution (using patents from 1988), a group that included
21 Stanford’s “middle period” biomedical patents also display a

following correction (Hall, 2000): lower mean generality score than their control patents, and UC
  non-biomedical patents applied for and issued before 1981 also
Ni exhibit slightly lower mean “generality scores” than their respective
Ĝi = Generali (2)
Ni − 1 control samples.
 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418 411

UC and Stanford as well as other US universities with
considerable pre-1980 patenting experience. Why do
we find little or no evidence of declines in the impor-
tance or generality of these universities’ patents after
the Bayh–Dole act? First, our sample of university
patents is small, although this should tend to favor
findings of no statistically significant differences
between the university and non-academic patent sam-
ples. Second, these results ignore the effects of entry
by other universities other than Columbia into patent-
ing and licensing after the passage of the Bayh–Dole
act on the characteristics of the overall US academic
patent portfolio. Although Henderson et al. (1998a)
acknowledge that entry into patenting by less experi-
enced institutions could account for their results, they
are not able to test for the effects of entry after 1980 by
inexperienced universities on overall academic patent
quality. In the next section, we analyze the effects on
the importance and generality of US academic patents
of entry by inexperienced academic patenters after
1980.
3.3. Entry, importance, and generality in overall US
academic patenting, 1975–1992
We probed the effects of entry by US universities
into patenting on the importance and generality of aca-
demic patents by constructing a dataset of all patents
assigned to US universities other than Stanford, UC,
and Columbia during the 1975–1992 period. Within
this dataset, we distinguished among three categories:
(1) universities with at least 10 patents that were
applied for after 1970 and issued during 1975–1980
(“high-intensity” incumbents); (2) universities with
at least one but less than 10 patents applied for after
1970 that issued during 1975–1980 (“low-intensity”
incumbents); and (3) universities with no patents is-
sued during the 1975–1980 period and at least one
patent issued during 1980–1992 that was applied for
after 1980 (“entrants”). Our definitions of “entrant”
and “incumbent” universities are somewhat arbitrary,
but we believe that this tripartite distinction enables
us to separate the effects on patent importance and

Table 7
Number of patents by year of issue, 10 or more issued patents applied for prior to 1981 and after 1970 (“high intensity” incumbents), less
than 10 issued patents applied for prior to 1981 and after 1970 (“low intensity” incumbents), and no issued patents applied for prior to
1981 and after 1970 (entrants)
Patent issue year High intensity incumbents Low intensity incumbents Entrants Total

Number Total (%) Number Total (%) Number Total (%)

1975 213 86.9 32 13.1 0 0.0 245
1976 248 89.2 30 10.8 0 0.0 94
1977 243 88.4 32 11.6 0 0.0 239
1978 262 89.7 30 10.3 0 0.0 280
1979 193 86.9 29 13.1 0 0.0 216
1980 295 87.8 41 12.2 0 0.0 332
1981 291 84.6 53 15.4 0 0.0 344
1982 319 88.1 42 11.6 1 0.3 361
1983 259 80.9 58 18.1 3 0.9 320
1984 320 76.2 85 20.2 15 3.6 419
1985 332 71.4 111 23.9 22 4.7 465
1986 384 71.0 130 24.0 27 5.0 541
1987 477 72.4 154 23.4 28 4.2 659
1988 496 74.8 138 20.8 29 4.4 663
1989 743 68.4 281 25.9 62 5.7 1086
1990 738 68.2 282 26.1 62 5.7 1082
1991 819 67.2 307 25.2 92 7.6 1218
1992 932 64.0 431 29.6 93 6.4 1456
Total 7564 2266 434 10264
Number of entities 51 92 81
412 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418

Table 8
Regression coefficients for high intensity incumbent × application yeara
Negative binomial model Tobit model
Dependent variable: importance Dependent variable: generality
Overall Biomedical Non-biomedical Overall Biomedical Non-biomedical
1975 0.36∗∗ (0.11) 0.27 (0.21) 0.39∗∗ (0.13) 0.10 (0.13) 0.14 (0.27) 0.08 (0.16)
1976 0.13 (0.11) 0.29 (0.22) 0.07 (0.12) 0.39∗∗ (0.13) 0.28 (0.27) 0.42∗∗ (0.15)
1977 0.19∗ (0.11) 0.19 (0.19) 0.19 (0.13) 0.15 (0.12) 0.21 (0.23) 0.13 (0.04)
1978 0.19∗ (0.11) 0.41∗∗ (0.19) 0.09 (0.13) 0.30∗∗ (0.13) 0.20 (0.23) 0.35∗∗ (0.15)
1979 0.27∗∗ (0.10) 0.01 (0.17) 0.40∗∗ (0.12) 0.17 (0.11) 0.38∗∗ (0.19) 0.05 (0.13)
1980 0.32∗∗ (0.09) 0.20 (0.15) 0.38∗∗ (0.11) 0.23∗∗ (0.11) 0.19 (0.18) 0.23∗∗ (0.13)
1981 0.38∗∗ (0.09) 0.26∗ (0.15) 0.45∗∗ (0.12) 0.45∗∗ (0.11) 0.63∗∗ (0.18) 0.37∗∗ (0.13)
1982 0.20∗∗ (0.09) −0.12 (0.15) 0.33∗∗ (0.12) 0.12 (0.10) −0.51∗∗ (0.17) 0.52∗∗ (0.13)
1983 0.27∗∗ (0.09) 0.17 (0.14) 0.35∗∗ (0.12) 0.17∗ (0.10) 0.00 (0.15) 0.30∗∗ (0.14)
1984 0.32∗∗ (0.08) 0.04 (0.12) 0.54∗∗ (0.11) 0.15∗ (0.09) 0.22∗ (0.13) 0.09 (0.12)
1985 0.31∗∗ (0.08) 0.02 (0.12) 0.54∗∗ (0.10) 0.35∗∗ (0.09) 0.22∗∗ (0.13) 0.43∗∗ (0.12)
1986 0.37∗∗ (0.08) 0.09 (0.11) 0.57∗∗ (0.10) 0.28∗∗ (0.08) 0.15 (0.13) 0.39∗∗ (0.11)
1987 0.20∗∗ (0.06) 0.03 (0.10) 0.31∗∗ (0.08) 0.22∗∗ (0.07) 0.25∗∗ (0.11) 0.19∗∗ (0.09)
1988 0.15∗∗ (0.06) −0.09 (0.10) 0.29∗∗ (0.08) 0.25∗∗ (0.07) 0.21∗∗ (0.11) 0.27∗∗ (0.08)
1989 0.29∗∗ (0.06) 0.24∗∗ (0.10) 0.32∗∗ (0.08) 0.17∗∗ (0.07) 0.15 (0.11) 0.18∗∗ (0.08)
1990 0.28∗∗ (0.07) 0.19∗ (0.12) 0.32∗∗ (0.08) 0.10 (0.07) −0.01 (0.13) 0.15∗ (0.09)
1991 −0.06 (0.10) −0.23 (0.18) 0.03 (0.12) 0.34∗∗ (0.11) 0.67∗∗ (0.20) 0.20 (0.13)
Constant 2.10 (0.19) 2.54 (0.63) 1.73 (0.46) 0.22 (0.22) −0.09 (0.90) 0.96 (0.51)
Number of 15125 5569 9556 11716 4313 7403
observations
Log likelihood −36170 −13280 −22836 −12468 −4456 −7939
Pseudo R2 0.02 0.01 0.02 0.01 0.01 0.01
a Standard errors in parentheses. Year dummies not reported. Sample excludes patents from University of California, Stanford, and

Columbia.
∗ P > 0.10.
∗∗ P > 0.05.

generality of increased patenting after Bayh–Dole by
active pre-1980 patenters (a group that includes UC
and Stanford) from increased post-1980 patenting by
universities historically inactive in this area and in-
creased patenting by universities (such as Columbia)
that were “low-intensity” patenters before 1980.
Table 7 contains information on the annual number
of patents issuing to these three groups of universi-
ties during the 1975–1992 period that were applied
for after 1970, demonstrating the declining share of
the “high-intensity” pre-1980 academic patenters af-
ter the passage of Bayh–Dole. The “high-intensity”
patenters’ share declines from more than 85% dur-
ing 1975–1980 to less than 65% by 1992. The
“low-intensity” pre-1980 patenters, by contrast, in-
crease their share of all academic patents from 15%
in 1981 to almost 30% in 1992. And entrants, in-
stitutions that display no patenting activity during
1975–1980, increase their share of overall academic
patenting from zero in 1980 to more than 6% by
1992.
Tables 8–10 present the results of separate regres-
sions of importance and generality for the patents
assigned to academic institutions in each of these
three categories, covering 1975–1991 for the “high-”
and “low-intensity” incumbents and 1981–1991 for
the “entrants”. 22 Negative binomial regression re-
sults are reported for models in which importance
is the dependent variable. Our analysis of general-
ity used a tobit specification, since the dependent
variable (generality) is truncated at a lower limit of
zero and an upper limit of one. Each specification
is estimated for a dataset covering the patents of the
relevant academic institutions and those in a control
22 Our construction of the academic patent dataset yields a small
number of observations for 1992, and we accordingly omit this
year from the results reported in Tables 8–10.
 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418 413

Table 9
Regression coefficients for low intensity incumbent × application yeara
Negative binomial model Tobit model

Dependent variable: importance Dependent variable: generality

Overall Biomedical Non-biomedical Overall Biomedical Non-biomedical

1975 0.65∗∗(0.32) 0.38 (0.77) 0.71∗∗(0.34) −0.30 (0.35) −1.01 (0.84) −0.15 (0.38)
1976 −0.11 (0.33) −0.49 (0.55) 0.24 (0.42) 0.84∗∗ (0.39) 1.85∗∗ (0.69) 0.23 (0.49)
1977 0.63∗∗ (0.30) 1.69∗∗ (0.67) 0.45 (0.34) 0.36 (0.33) — 0.19 (0.37)
1978 0.01 (0.27) 0.18 (0.47) −0.05 (0.32) −0.29 (0.29) −0.15 (0.57) −0.35 (0.33)
1979 0.66∗∗ (0.25) 0.35 (0.45) 0.79∗∗ (0.30) 0.62∗∗ (0.29) 0.69 (0.54) 0.59∗ (0.34)
1980 0.35 (0.24) 0.13 (0.38) 0.51∗ (0.31) −0.01 (0.25) 0.23 (0.41) 0.14 (0.31)
1981 −0.12 (0.20) −0.15 (0.34) −0.10 (0.26) 0.31 (0.22) 1.01∗∗ (0.40) 0.00 (0.27)
1982 −0.06 (0.18) −0.21 (0.25) 0.10 (0.25) 0.38∗∗ (0.18) 0.35 (0.25) 0.42∗ (0.26)
1983 0.16 (0.15) 0.50∗ (0.27) 0.03 (0.18) 0.18 (0.16) 0.63∗∗ (0.32) 0.02 (0.19)
1984 −0.18 (0.15) −0.25 (0.22) −0.11 (0.20) 0.04 (0.16) −0.13 (0.24) 0.19 (0.22)
1985 −0.05 (0.14) −0.04 (0.23) −0.06 (0.18) −0.07 (0.16) −0.19 (0.25) 0.00 (0.19)
1986 0.40∗∗ (0.13) 0.38∗∗ (0.19) 0.41∗∗ (0.17) 0.41∗∗ (0.14) 0.38∗∗ (0.20) 0.43∗∗ (0.19)
1987 0.11 (0.11) 0.04 (0.19) 0.15 (0.14) 0.28∗∗ (0.12) 0.06 (0.20) 0.40∗∗ (0.15)
1988 0.17∗ (0.10) 0.02 (0.17) 0.24∗∗ (0.12) 0.15 (0.10) 0.15 (0.18) 0.16 (0.12)
1989 0.38∗∗ (0.10) 0.22 (0.17) 0.46∗∗ (0.11) 0.29∗∗ (0.10) 0.26 (0.18) 0.29∗∗ (0.12)
1990 0.15∗ (0.09) 0.17 (0.18) 0.14 (0.11) 0.20∗∗ (0.10) −0.05 (0.18) 0.29∗∗ (0.11)
1991 0.11 (0.16) −0.19 (0.30) 0.28 (0.18) 0.53∗∗ (0.16) 0.38 (0.31) 0.57∗∗ (0.19)
Constant 1.10 (0.38) 2.30 (1.13) −16.41 (3661.49) 0.69 (0.38) 0.67 (1.01) 0.20 (0.52)
Number of 4535 1627 2908 3528 1242 2286
observations
Log likelihood −10980 −3893 −7056 −3745 −1285 −2436
Pseudo R2 0.01 0.01 0.01 0.01 0.02 0.01
a Standard errors in parentheses. Year dummies not reported. Sample excludes patents from University of California, Stanford, and

Columbia.
∗ P > 0.10.
∗∗ P > 0.05.

sample of non-academic patents (matched by applica- “high-intensity” incumbents exhibit consistently

tion year and patent class). We control for year effects
and interact a dummy variable denoting academic
patents with a dummy for the application year-the re-
ported coefficients are those for the interaction terms
for 1975–1991. As in previous sections, we present
results for overall academic patents, as well as for
biomedical and non-biomedical patents.
The results for our analysis of each of these
three samples of academic patents display contrast-
ing patterns of importance and generality. 23 The
23 We also ran probit regressions that examined the fraction of
patents assigned to universities in these three groups that received
no citations in the first 6 years following their issue. Henderson,
Jaffe and Trajtenberg use this “zero-citation” measure as another
measure of change in the characteristics of university patents. The
results of these probit regressions were broadly similar to those
reported in Tables 8–10.
more important and more general patents, relative to
non-academic patents, throughout the 1975–1991 pe-
riod (Table 8). Thirteen of the seventeen interaction
coefficients in column 2 for the overall patent sample
associated with this group of universities are positive
and statistically significant at the 5% level, indicating
that for most of this period (and increasingly so during
the 1980s), these institutions’ patents were cited more
intensively than those in the matched industrial sam-
ple. As was true in the analysis of the UC, Stanford,
and Columbia patents, the differences in importance
between the patents of these experienced academic
patenters and the non-academic patent control sample
are most significant for the non-biomedical patents.
A broadly similar pattern is generated in the results
of our specifications comparing the generality of
the entire sample of patents accounted for by these
414 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418

Table 10
Regression coefficients for entrant university × application yeara
Negative binomial model Tobit model

Dependent variable: importance Dependent variable: generality

Overall Biomedical Non-biomedical Overall Biomedical Non-biomedical

1981 0.39 (0.51) 0.51 (0.81) 0.18 (0.68) 0.38 (0.57) 0.35 (0.78) –
1982 0.80 (0.56) 1.56∗ (0.86) 0.11 (0.72) −0.81 (0.54) −0.42 (0.77) −1.09 (0.74)
1983 0.18 (0.36) 0.53 (0.60) −0.20 (0.44) −0.53 (0.34) 0.18 (0.52) −0.95∗∗ (0.44)
1984 −0.19 (0.32) −0.28 (0.44) −0.03 (0.45) −0.11 (0.32) 0.00 (0.39) −0.22 (0.49)
1985 −0.03 (0.35) −0.63 (0.65) 0.11 (0.40) 0.06 (0.35) −0.72 (0.61) 0.33 (0.45)
1986 0.23 (0.29) 0.12 (0.48) 0.29 (0.35) 0.28 (0.28) 0.49 (0.42) 0.13 (0.37)
1987 −0.16 (0.23) −0.62∗∗ (0.31) 0.50∗ (0.31) 0.35 (0.22) 0.33 (0.27) 0.37 (0.33)
1988 0.19 (0.23) −0.20 (0.34) 0.46∗ (0.29) 0.37∗ (0.21) 0.30 (0.31) 0.43 (0.28)
1989 0.51∗∗ (0.19) 0.11 (0.26) 0.85∗∗ (0.25) 0.51∗∗ (0.18) 0.56∗∗ (0.3) 0.44 (0.27)
1990 0.26 (0.20) 0.10 (0.28) 0.47∗∗ (0.28) 0.19 (0.20) 0.27 (0.26) 0.10 (0.29)
1991 0.32 (0.32) 0.53 (0.55) 0.24 (0.38) 0.38 (0.20) 0.31 (0.50) 0.42 (0.38)
Constant 1.70 (0.83) 1.32 (0.59) 0.25 (0.52) −0.52 (0.42) 1.36 (0.67) 1.46 (0.80)
Number of observations 868 367 501 666 284 382
Log likelihood −2105 −919 −1163 −700 −290 −399
Pseudo R2 0.01 0.02 0.02 0.03 0.04 0.04
a
Standard errors in parentheses. Year dummies not reported. Sample excludes patents from University of California, Stanford, and
Columbia.
∗ P > 0.10.
∗∗ P > 0.05.

“high-intensity incumbents” with those in the control
sample-all of the interaction coefficients are posi-
tive wherever significant, and 10 of the 17 positive
interaction coefficients are statistically significant
at the 5% level. Once again, these differences in
generality are more frequently significant for the
non-biomedical patents of these experienced academic
patenters.
The results for the other two groups of academic
institutions, whose share of total US academic patent-
ing increased substantially after 1980, reveal con-
sistently lower levels of importance and generality
for their patents throughout the 1980s. The results
of our regressions for the “low-intensity” pre-1981
patenters (Table 9) reveal only 5 years of patents
that are more intensively cited (at the 5% signifi-
cance level) than those in the non-academic control
sample. The results of the regressions for generality
of their overall patents for this group of institutions
yield positive and statistically significant (5% signif-
icance level) year-coefficients for 8 of the 17 years.
Finally, the results for 1981–1991 for the “entrant”
population of institutions with no patenting activ-
ity during the 1975–1980 period also indicate lower
levels of importance (only 1 year interaction is pos-
itive and statistically significant at the 5% level in
Table 10) and generality (also one significant interac-
tion coefficient), relative to non-academic patents, than
those accounted for by the “high-intensity” pre-1981
patenters.
These results suggest that the patents of the two
groups of US universities that increased their share of
overall academic patenting during the 1980s were of
lower importance and generality, by comparison with
non-academic patents, than the patents issuing to US
universities with long histories of patenting activity.
The findings broadly corroborate our earlier conclu-
sion concerning the lack of any deterioration in the
importance and generality of post-1980 patents issu-
ing to UC and Stanford University. Taken together,
these results indicate that any “deterioration” in the
importance and quality of post-1980 US academic
patents may have resulted from the Bayh–Dole act’s
encouragement of entry into patenting by academic
 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
institutions with relatively little experience in this
activity. 24
In this view, Bayh–Dole’s effects on the content of
US academic research during 1981–1992 were mod-
est, by comparison with the act’s encouragement of a
new group of universities to expand or begin patent-
ing of faculty inventions. With the passage of time,
learning on the part of these entrants may gradually
improve their management of patenting and licensing
activities, and the apparent differences between their
patent portfolios and those of the institutions long ac-
tive in patenting may decline. 25 We lack a sufficiently
lengthy or rich longitudinal time series to test this pos-
sibility, although it is consistent with some anecdo-
tal evidence on many “entrant” academic institutions’
patenting activities during the 1980s. But this interpre-
tation of the effects of Bayh–Dole has different, and
arguably less worrisome, implications for the future
of the US academic research enterprise than the alter-
native characterization noted above, and the causes of
any change during the 1980s in the characteristics of
US academic patents merits additional study.
4. Conclusion
Since the passage of the Bayh–Dole act, many
US universities have expanded (or begun) programs
to patent and license the results of federally and in-
dustrially funded research. At both UC and Stanford
University, Bayh–Dole resulted in expanded efforts
to market licenses to academic inventions. These
expanded marketing efforts appear to have been as-
sociated with a modest decline in the “yield” of the
invention marketing efforts at both of these institu-
tions. At both universities, Bayh–Dole’s effects on
24 Our findings of difference in significance in importance and
generality between the “high intensity” incumbent, “low intensity”
incumbent, and entrant sub-samples and their respective control
groups could be affected by differences in standard errors due to
the various sizes of the three sub-samples. To check the robustness
of our results, we conducted, but do not report here, similar
regressions combining all three sub-samples and found broadly
similar results to those reported in Tables 8–10.
25 In a recent paper (Mowery et al., 2000), we find that these
characteristics of post-Bayh–Dole entrant universities’ patents do
improve during the 1980s, relative to a matched control sample
of non-academic patents.
415
the content of academic research were modest, espe-
cially given the pre-1980 shift in the composition of
this academic invention portfolio in favor of biomed-
ical inventions that were themselves highly attractive
to commercial licensors. Moreover, the pre- and
post-Bayh–Dole licensing efforts, and the revenue
flows associated with these licensing efforts, tended
to be concentrated in the biomedical area.
Nevertheless, the upsurge in patenting and licens-
ing at these and other US research universities after
1980 was affected by other factors in addition to the
Bayh–Dole act, and it is difficult to separate the ef-
fects of the act from those of other factors. In par-
ticular, by the mid-1970s biomedical technology, es-
pecially biotechnology, had increased significantly in
importance as a productive field of university research
with research findings that were of great interest to
industry. The feasibility of technology licensing in
biotechnology was advanced by the Diamond versus
Chakrabarty Supreme Court decision, which opened
the door to patenting the organisms, molecules, and re-
search techniques emerging from biotechnology. This
judicial decision, as well as the broader shift in US
policy to strengthen intellectual property rights, con-
tributed to the increased post-1980 patenting and li-
censing activities of US research universities.
Our analysis of the effects of Bayh–Dole on the con-
tent of academic research and the importance of the
patents assigned to these two leading research univer-
sities supports some reinterpretation of the findings of
Henderson et al. (1995, 1998a). Both UC and Stanford
appear to have intensified their marketing of faculty in-
ventions after 1980, with some decline in the “yield” of
their activities. But there is no evidence of significant
change in the characteristics of the academic research
disclosures received by these universities’ technology
licensing offices. Nor does our analysis of the pre- and
post-Bayh–Dole patents assigned to these two univer-
sities reveal any decline in the importance or gener-
ality of these universities’ patents after Bayh–Dole,
in contrast to the findings of Henderson et al. (1995,
1998a) for a larger sample of US academic patents and
a subset of intensive academic patenters. The patents
of a major post-1980 “entrant”, Columbia University,
also are no less important or general than those within
a matched sample of non-academic patents.
How do we reconcile a finding that citation-based
measures of UC and Stanford patents reveal no
416 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
decline in importance after Bayh–Dole with our
conclusion that both UC and Stanford’s technology
licensing operations appear to have experienced a de-
cline in “yield”, i.e. a decline in the share of licenses
yielding positive revenues? Fundamentally, these two
sets of indicators measure different characteristics of
the invention and patent portfolios of these universi-
ties. Along with other scholars, we interpret patent
citations as measures of the importance of the con-
tribution to inventive knowledge of a given patent.
But this contribution may or may not be correlated
with the willingness of an industrial firm to pay for a
license for this patent. The extent of correlation be-
tween licensing revenues and patent citations remains
an important question for future research.
Our analysis of overall US academic patenting
before and after 1980 yields conclusions that are
broadly consistent with the results of the analysis of
UC, Stanford, and Columbia patenting. Specifically,
we find no decline in the importance and general-
ity of the post-1980 patents assigned to universities
with substantial pre-1980 patent portfolios. But the
patents of institutions with little or no previous his-
tory were not significantly more important or general
than non-academic patents. This empirical analysis
suggests that any effects of Bayh–Dole on overall aca-
demic patent quality reflect the act’s effects on entry,
rather than on the incentives of academic researchers
and administrators in long-active academic patenters.
The evidence from the Columbia post-1980 patent
sample, which is no less heavily or broadly cited than
those of its non-academic control sample, suggests
that there is considerable heterogeneity within the
population of entrant institutions.
The limitations of our analysis are apparent. We
start with detailed institutional data on patenting and
licensing for only three very unusual universities,
which include institutions that were among the leaders
in patenting and licensing of faculty inventions before
and after the act’s passage. The empirical results for
both our three-university and broader academic sam-
ples are sensitive to the composition and construction
of the patent control samples. In assessing the effects
of Bayh–Dole on university “technology transfer”, we
are analyzing only the formalized technology transfer
activities of these universities, and cannot exclude
the possibility that activities in invention disclosure,
patenting, and licensing may affect the numerous
other channels through which university knowledge
reaches commercial application. Moreover, the small
size of our samples of university patents, especially
those covering the pre-Bayh–Dole period, limits the
robustness and power of our statistical tests. Finally,
this analysis of the post-Bayh–Dole period necessarily
covers only the early years of this new “regime”. As
denizens of any university can attest, change within
these institutions occurs slowly, and it is possible
that the “true” effects of the Bayh–Dole act are only
now being revealed. Nevertheless, we believe that the
results of this analysis underscore the importance of
complementing analyses of aggregate data on aca-
demic and non-academic patenting trends with work
on individual institutions engaged in these pursuits,
be they firms, universities, or public laboratories.
Acknowledgements
Authors’ names appear in alphabetical order. Ear-
lier versions of this paper were presented at the
1999 AEA meetings and the July 1999 meeting of
the NBER’s Science and Technology Policy Group,
and benefited from comments by participants at both
meetings. The paper also benefited from the com-
ments of anonymous reviewers. Portions of this paper
draw on an earlier paper co-authored with Richard
Nelson and Bhaven Sampat of Columbia University
(Mowery, Nelson, Sampat, and Ziedonis, 2001), and
we have benefited from numerous conversations and
comments from them on the issues covered in this
paper. We are indebted to the staff of the technology
licensing offices of Stanford University, the Uni-
versity of California, and Columbia University for
invaluable assistance with the collection and analysis
of these data. Michael Barnes and Lynn Fissell of the
University of California assisted in the collection and
analysis of the University of California data, and the
research on the Stanford data benefited from the assis-
tance of Sandra Bradford. Special thanks to Michael
Barnes for the use of his university patenting data and
to Adam Jaffe of Brandeis University, Bronwyn Hall
of UC Berkeley, and Manuel Trajtenberg of Tel Aviv
University and the NBER for making the NBER-Case
Western Reserve patent data available to us. Support
for this research was provided by the California Pol-
icy Seminar, the UC President’s Industry-University
 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
Cooperative Research Program, the Alfred P. Sloan
Foundation, the Huntsman Center for Global Re-
search of the Wharton School at the University of
Pennsylvania, and the Andrew Mellon Foundation.
Further reading
Arrow, K., 1962. Economic welfare and the allocation of resources
for invention. In: Nelson, R.R. (Ed.), The Rate and Direction
of Inventive Activity. Princeton University Press, Princeton.
Bok, D., 1982. Beyond the Ivory Tower. Harvard University Press,
Cambridge.
Cohen, W., Florida, R., Goe, R., 1994. University-
Industry Research Centers in the United States. Technical
report, Center for Economic Development, Carnegie-Mellon
University.
Cohen, W., Florida, R., Randazzese, L., Walsh, J., 1998. Industry
and the academy: uneasy partners in the cause of technological
advance. In: Noll, R. (Ed.), Challenges to the Research
University. Brookings Institution, Washington, DC.
Caves, R., Crookell, H., Killing, P., 1983. The Imperfect Market
for Technology Licenses. Oxford Bulletin of Economics and
Statistics 249–267.
Crow, M.M., Gelijns, A.C., Nelson, R.R., Raider, H.J., Sampat,
B.N., 1998. Recent Changes in University-Industry Research
Interactions: A Preliminary Analysis of Causes and Effects.
Working Paper, Columbia University.
David, P.A., Foray, D., 1995. Accessing and expanding the science
and technology knowledge base. STI Review 16, 13–68.
David, P.A., Mowery, D.C., Steinmueller, W.E., 1992. Analyzing
the economic payoffs from basic research. Economics of
Innovation and New Technology 73–90.
Eisenberg, R., 1996. Public research and private development:
patents and technology transfer in government-sponsored
research. Virginia Law Review 1663–1727.
Evenson, R.E., 1982. Agriculture. In: Nelson, R.R. (Ed.),
Government and Technical Progress: A Cross-Industry
Analysis. Pergamon Press, New York.
Heilbron, J., Seidel, R., 1989. Lawrence and His Laboratory: A
History of the Lawrence Berkeley Laboratory. University of
California Press, Berkeley.
Henderson, R., Jaffe, A.B., Trajtenberg, M., 1994. Numbers Up,
Quality Down? Trends in University Patenting, 1965–1992.
In: The CEPR Conference on University Goals, Institutional
Mechanisms, and the ‘Industrial Transferability’ of Research,
Stanford University, 18–20 March 1994.
Henderson, R., Jaffe, A.B., Trajtenberg, M., 1998b. University
patenting amid changing incentives for commercialization. In:
Barba Navaretti, G., Dasgupta, P., Mäler, K.G., Siniscalco, D.
(Eds.), Creation and Transfer of Knowledge. Springer, New
York.
Katz, M.L., Ordover, J.A., 1990. R&D competition and
cooperation. Brookings Papers on Economic Activity:
Microeconomics 137–192.
417
Levin, R.C., Klevorick, A., Nelson, R.R., Winter, S., 1987.
Appropriating the returns from industrial research and
development. Brookings Papers on Economic Activity 783–820.
Mowery, D.C., 1983. The relationship between intrafirm
and contractual forms of industrial research in American
manufacturing, 1900–1940. Explorations in Economic History
351–374.
Mowery, D.C., Rosenberg, N., 1993. The US national innovation
system. In: Nelson, R.R. (Ed.), National Innovation Systems:
A Comparative Analysis. Oxford University Press, New York.
Mowery, D.C., Rosenberg, N., 1998. Paths of Innovation:
Technological Change in 20th-Century America. Cambridge
University Press, New York.
Mowery, D.C., Ziedonis, A.A., 1998. Market failure or market
magic? Structural change in the US national innovation system.
STI Review.
National Research Council, 1997. Intellectual Property Rights and
Research Tools in Molecular Biology. National Academy Press,
Washington, DC.
National Science Board, 1996. Science and Engineering Indicators.
US Government Printing Office, Washington, DC.
National Science Foundation, 1994. National Patterns of R&D
Resources. US Government Printing Office, Washington, DC.
National Science Foundation, 1996. National Patterns of R&D
Resources. US Government Printing Office, Washington, DC.
National Science Foundation, 1998. National Patterns of R&D
Resources. US Government Printing Office, Washington, DC.
Nelson, R.R., 1959. The simple economics of basic scientific
research. Journal of Political Economy, 297–306.
Nelson, R.R., 1992. What is ‘commercial’ and what is ‘public’
about technology, and what should be? In: Rosenberg, N.,
Landau, R., Mowery, D.C. (Eds.), Technology and the Wealth
of Nations. Stanford University Press.
Nelson, R.R., Mazzoleni, R., 1997. Economic theories about the
costs and benefits of patents. In: National Research Council,
Intellectual Property Rights and Research Tools in Molecular
Biology. National Academy Press, Washington, DC.
Nelson, R.R., Romer, P., 1997. Science, Economic Growth, and
Public Policy. In: Smith, B., Barfield, C. (Eds.), Technology,
R&D, and the Economy. Brookings Institution, Washington,
DC.
Office of Technology Transfer, University of California, 1997.
Annual Report: University of California Technology Transfer
Program. University of California, Oakland, CA.
Rosenberg, N., 1992. Scientific instrumentation and university
research. Research Policy 21, 381–390.
Rosenberg, N., 1998. Technological change in chemicals: the
role of university-industry relations. In: Arora, A., Landau, R.,
Rosenberg, N. (Eds.), Chemicals and Long-Term Economic
Growth. Wiley, New York.
Rosenberg, N., Nelson, R.R., 1994. American universities and
technical advance in industry. Research Policy 23, 323–348.
Science, 1997. Publishing sensitive data: Who’s calling the shots?
pp. 523–526.
Stata Corp., 1999. Stata Statistical Software: Release 6.0. Stata
Corporation, College Station, TX.
418 D.C. Mowery, A.A. Ziedonis / Research Policy 31 (2002) 399–418
Trajtenberg, M., Jaffe, A., Hall, B.H., 2000. NBER-Case Western
University Patent Data. NBER, Brandeis University, University
of California, Berkeley, and Tel Aviv University.
Williamson, O.E., 1979. Transaction cost economics: the
governance of contractual relations. Journal of Law and
Economics 233–262.
Williamson O.E., 1985. The Economic Institutions of Capitalism.
Free Press, New York.
US General Accounting Office, 1995. University research: effects
of indirect cost revisions and options for future changes. US
Government Printing Office, Washington, DC.
US Office of Management and Budget, Executive Office of the
President, 1995. The Budget of the United States Government
for Fiscal 1996. US Government Printing Office, Washington,
DC.
References
Association of University Technology Managers, 1994. The AUTM
Licensing Survey: Executive Summary and Selected Data,
Fiscal Years 1993, 1992, and 1991. Association of University
Technology Managers, Norwalk, CT.
Hall, B.H., 2000. A Note on the Bias in the Herfindahl Based on
Count Data. Unpublished note, http://emlab.berkeley.edu/users/
bhhall/hhibias.pdf.
Henderson, R., Jaffe, A.B., Trajtenberg, M., 1995. Universities as
a Source of Commercial Technology: A Detailed Analysis of
University Patenting, 1965–1988. National Bureau of Economic
Research Working Paper #5068.
Henderson, R., Jaffe, A.B., Trajtenberg, M., 1998a. Universities
as a source of commercial technology: a detailed analysis
of university patenting, 1965–1988. Review of Economics &
Statistics 119–127.
Joint Economic Committee US Congress, 1999. Entrepreneurial
Dynamism and the Success of US High-Tech: Joint Economic
Committee Staff Report. US Government Printing Office,
Washington, DC.
Mowery, D.C., Nelson, R.R., Sampat, B.N., Ziedonis, A.A.,
1999. The effects of the Bayh–Dole act on US university
research and technology transfer: an analysis of data from
Columbia University, the University of California, and Stanford
University. In: Branscomb, L., Florida, R. (Eds.), Industrializing
Knowledge. MIT Press, Cambridge, MA.
Mowery, D.C., Sampat, B.N., Ziedonis, A.A., 2000. Learning to
patent: institutional experience, learning, and the characteristics
of US university patents after the Bayh–Dole act, 1981–1996.
In: The Conference on University-Industry Technology Transfer
and Entrepreneurship, Georgia Institute of Technology, 8–9
December 2000.
Mowery, D.C., Nelson, R.R., Sampat, B.N., Ziedonis, A.A., 2001.
The growth of patenting and licensing by US universities:
an assessment of the effects of the Bayh–Dole act of 1980.
Research Policy 30, 99–119.
Sampat, B.N., Ziedonis, A.A., 2000. Patent citations and the
value of university patent rights. Wharton School Working
Paper.
Trajtenberg, M., Henderson, R., Jaffe, A.B., 1997. University
versus corporate patents: a window on the basicness of
inventions. Economics of Innovation and New Technology
19–50.
Ziedonis, A.A., 2001. The commercialization of university research
through patenting and licensing: implications for firm strategy
and public policy. Unpublished Dissertation, University of
California, Berkeley.