Você está na página 1de 12

research

ELSEVIER Research Policy 25 (1996) 1047-1058


policy

The modem university: contributor to industrial innovation and


1
recipient of industrial R & D support
Edwin Mansfield a,,, Jeong-Yeon Lee b
a Center for Economics and Technology, Department of Economics, 3718 Locust Walk, University of Pennsyloania, Philadelphia, PA
19104-6297, USA
b Korea Institute for International Economic Policy, Seoul, 135-619, South Korea
Final version received March 1996

Abstract

The interface between industry and the universities is of key importance in the promotion of technological change in
many industries. There is intense interest in the characteristics of universities that have contributed most importantly to
industrial innovation in various fields, but unfortunately very little systematic study has been devoted to this important and
complex topic. Also, we know little about the factors determining which universities firms will support to do R & D of
various types. Based on data obtained from a carefully selected sample of major US firms in the electronic, information
processing, chemical, petroleum, pharmaceutical, instruments, and metal industries, this paper sheds new light on these
topics.

1. Introduction ers, wanting to increase the economic payoff from


academic research, are intensely interested in evi-
Universities play a major role in originating and dence regarding the characteristics of universities
promoting the diffusion of knowledge and tech- that have contributed most importantly to industrial
niques that contribute to industrial innovations. In innovation in various fields. Unfortunately, very lit-
1975-85, about 10% of the new products and pro- tle systematic study has been devoted to this impor-
cesses in US high-technology industries were based tant and complex topic. In this paper, we present and
directly on recent academic research. 2 Policy mak- analyze for the first time data on this score provided
by a carefully selected sample of US firms in seven
major industries.
* Corresponding author.
Whether or not a particular university makes ma-
Part of this paper was presented at the 1995 annual meetings
of the American Economic Association. The research on which jor contributions to the development o f innovations
this paper is based was supported by a grant to Mansfield from the in a particular industry depends in part on the extent
National Science Foundation. Lee's contribution, focused on the o f industrial support for research at that university.
survey in Section 4, was part of his work toward a doctoral Very little is known about the factors determining
dissertation at the University of Pennsylvania. We are grateful to
which universities firms will support to do R & D o f
Lewis Branscomb, William Eva_n, Diana Hicks, Leonard Leder-
man, and Jaewoo Ryoo for helpful comments. various types. How important is geographical prox-
2 For example, see Mansfield (1991) and Mansfield (1992). imity? How important is faculty quality? The an-

0048-7333/96/$15.00 Copyright © 1996 Elsevier Science B.V. All rights reserved.


PII S0048-7333(96)00893-1
1048 E. Mansfield, J.-Y. Lee/Research Policy 25 (1996) 1047-1058

swers to these questions are important from both which these researchers worked, and the percentage
analytical and policy viewpoints. This paper presents of citations going to each university. 4
and analyzes data on this topic too. Clearly, these citations are distributed over a con-
siderable number of universities; the top four univer-
sities generally receive about 30% of the citations. 5
2. Universities making significant contributions to As would be expected, the most frequently cited
US industrial innovations universities tend to be w o r d leaders in science and
technology. For example, in electronics and informa-
To obtain information regarding which universi- tion processing, MIT, UC Berkeley, Illinois, Stan-
ties have contributed most significantly to US indus- ford, and CMU head the list. A statistical analysis of
trial innovation in the electronic, information pro- these data shows that there generally is a direct
cessing, chemical, petroleum, pharmaceutical, instru- relationship between the quality of a university's
ments, and metal industries, a random sample of 70 faculty in the relevant department, as estimated by
major firms was drawn from these industries. Each the National Academy of Sciences (1982), 6 and the
firm was asked to cite about five academic re- percent of citations received by the university. (See
searchers whose work in the 1970s and 1980s con- Mansfield, 1995a.)
tributed most importantly to the firm's new products However, this does not mean that 'second tier'
and processes introduced in the 1980s. While the departments do not play an important role in this
initial requests for data and cooperation were made regard. On the contrary, the remarkable fact is that
to the firms' chairmen, the respondents usually were about 40% of these citations went to universities
the top R & D executives who based their responses with 'adequate-to-good' and 'marginal' faculties in
in substantial part on detailed information obtained the relevant departments, not to those with 'good-to-
from people at lower levels of their organizations. distinguished' faculties, according to the National
Most of the firms went to a considerable amount
of trouble to supply these data. Written responses
frequently were supplemented with interviews with 4 TO present the data more concisely, pairs of industries are
relevant company personnel. Eventually usable data combined in Tables 1-3. Since the electronics and information
were obtained from 66 of the 70 firms in the sample. processing industries are so closely related, they are combined.
Since these firms account, on the average, for about For the same reason, the chemical and petroleum industries are
combined. Pharmaceuticalsand instrumentsare combinedbecause
a third of the R & D expenditures in these industries, some of the instrument makers are in medical fields. Readers
the sample seems quite adequate. Taken as a whole, wanting data for these individualindustriescan obtain them from
these 66 firms cited 321 academic researchers. 3 the authors.
Tables 1 - 3 (and Footnote 4) list the universities at The data for the metals industry (which are relatively brief and
come from only four firms) are as follows (percent of citationsin
parenthesis): Utah (16), Col. Sch. of Mines (4), Stanford(4), MIT
(12), Comell (4), UC Berkeley (4), UBC (12), Illinois (4), Vir-
ginia (4), Ohio State (8), L'Ecole Central (4), Waterloo (4),
Arizona (4), Missouri Rolla (4), Wright State (4), Cincinnati(4),
3 An attempt was made to allocate the sample optimallyamong Rhode Island (4).
industries(that is, with sample size in each industrybeing propor- 5 In the metals industry, the percentage is substantiallyhigher,
tional to the total number of finns in the industry times the but the data pertain to only four finns.
relevant standard deviation). The industrial distribution of the 6 For the electronics industry, we assume that the relevant
finns in the sample was as follows: electronics, 14; information department is electrical engineering;for informationprocessing,
processing, 16; chemicals, 13; petroleum, 5; pharmaceuticals, 8; we assume it is computer science; for pharmaceuticals,biochem-
instruments,6; and metals, 4. The numberof academic researchers istry; for chemicals, chemistry;and petroleum, chemical engineer-
cited by each industrywas as follows: electronics,84; information ing. Of course, this is rough, but the evidence indicatesthat these
processing, 64; chemicals, 51; petroleum,28; pharmaceuticals,47; are key departments for these industries. See Mansfield(1995a).
instruments, 22; metals, 25. Eighteen academic researchers were This statistical analysis was conducted only for these five indus-
cited by more than one finn, so the number of distinct researchers tries because of the relatively small sample of finns in the
cited is 303, not 321. In Tables 1-3, each researcher is weighted remaining two industries.For detailed discussionof the National
by the number of firms that cited him or her. Academy of Sciences ratings, see Foomote 9.
E. Mansfield, J.-Y. Lee/Research Policy 25 (1996) 1047-1058 1049

Table 1
Universities containing academic researchers cited by US electronics and informationprocessing firms as contributing most importantly
(during the 1970s and 1980s) to the developmentof their new products and processes introducedin the 1980s a
University Percentage University Percentage of University Percentage
of citations citations of citations
MIT 12.8 Pennsylvania 1.4 Michigan 0.7
UC Berkeley 10.8 Penn State 1.4 Missouri 0.7
Illinois 7.4 Texas A& M 1.4 Newcastle 0.7
Stanford 6.8 USC 1.4 New Mexico 0.7
CMU 6.1 Utah 1.4 NY Polytechnic 0.7
Minnesota 3.4 Aachen 0.7 Ottawa 0.7
Brigham Young 2.7 Calstate LA 0.7 Purdue 0.7
Arizona State 2.0 Calstate SLO 0.7 RPI 0.7
Caltech 2.0 Central Florida 0.7 UCSD 0.7
Ohio State 2.0 Cincinnati 0.7 UCSB 0.7
UCLA 2.0 Delaware 0.7 UCSC 0.7
Arizona 1.4 Drexel 0.7 Swansi 0.7
Cambridge 1.4 Florida 0.7 Tennessee 0.7
Chicago 1.4 Georgia 0.7 Toronto 0.7
Columbia 1.4 Iowa State 0.7 Wake Forest 0.7
Connecticut 1.4 lrvine 0.7 Washington 0.7
Cornell 1.4 Kansas State 0.7 Univ. of Washington 0.7
Harvard 1.4 Maricopa 0.7 Wisconsin 0.7
Iowa 1.4 McGill 0.7 Worcester Polytech 0.7
Leuven 1.4 Miami 0.7 Total b 100.0

a The followingabbreviationsare used: MIT, Massachusetts Instituteof Technology;CMU, Carnegie Mellon University;UCLA, University
of Californiaat Los Angeles;USC, Universityof Southern California;RPI, RensellaerPolytechnicInstitute;UCSD, Universityof California
at San Diego; UCSB, Universityof Californiaat Santa Barbara; UCSC, Universityof Californiaat Santa Cruz.
b Due to roundingerrors, figures may not sum to total.
Source: see Section 2.

Academy of Sciences (NAS) ratings. (This percent- state as a particular university and the number of
age was higher in the information processing and citations received by that university. In electronics
chemical industries than in the others.) These results and information processing, about 40% of the uni-
strongly suggest that members of 'second tier' de- versities cited are in the same state as the firm
partments are a valuable and frequently used source making the citations; in chemicals, drugs, and instru-
of research findings for industry. ments, the figure is about 25%. 7
Geography too seems to be important. While the
proportion of citations to universities outside the
United States tends to be higher in the pharmaceuti- 3. I n d u s t r i a l s u p p o r t o f a c a d e m i c r e s e a r c h
cal, chemical, and metals industries (about 20%)
than in the others (less than 10%), the bulk of the Still another factor that determines the number of
citations made by these US firms are to US universi- citations received by a university is the amount of
ties. (Most of the non-US citations are to Canada and
the United Kingdom.) Within the United States,
universities located near many of the firms in the 7 By industry, the mean percentage of a firm's citations going
to universities in the same state was: electronics, 39%; informa-
sample tend to be cited relatively often. In all indus- tion processing, 38%; chemicals, 27%; pharmaceuticals, 23%;
tries, there is a direct relationship between the per- inslruments, 21%; metals, 19%; and petroleum, 5%. Also, see
centage of firms in the sample that are in the same ibid.
1050 E. Mansfield, J.-Y. Lee~Research Policy 25 (1996) 1047-1058

Table 2
Universities containing academic researchers cited by US chemical and petroleum firms as contributing most importantly (during the 1970s
and 1980s) to the development of their new products and processes introduced in the 1980s a
University Percentage University Percentage University Percentage
of citations of citations of citations
MIT 7.8 Aachen 1.3 Minnesota 1.3
Washington 7.8 Amsterdam 1.3 Northwestern 1.3
Delaware 6.5 UC Berkeley 1.3 Oxford 1.3
Florida 3.9 Bowling Green 1.3 Pomona 1.3
Penn State 3.9 Calgary 1.3 RP1 1.3
Princeton 3.9 Caltech 1.3 Rheinische FW 1.3
Utah 3.9 Cincinnati 1.3 Salerno 1.3
Akron 2.6 CUNY 1.3 South Carolina 1.3
Case Western Res. 2.6 Clarkson 1.3 Southern Mississippi 1.3
Houston 2.6 Cleveland State 1.3 Stanford 1.3
Imperial 2.6 Columbia 1.3 Texas 1.3
Johns Hopkins 2.6 Indiana 1.3 Toledo 1.3
Lehigh 2.6 Iowa State 1.3 UCLA 1.3
Notre Dame 2.6 Leuven 1.3
Rutgers 2.6 Louisiana 1.3 William and Mary 1.3
VPI 2.6 McMasters 1.3 Total b 100.0

a The following abbreviations (other than those described in Table 1) are used: VPI, Virginia Polytechnic Institute; CUNY, City University
of New York.
b Dee to rounding errors, figures may not sum to total.
Source: see Section 2.

Table 3
Universities containing academic researchers cited by US pharmaceutical and instrument firms as contributing most importantly (during the
1970s and 1980s) to the development of their new products and processes introduced in the 1980s a
University Percentage University Percentage University Percentage
of citations of citations of citations
Harvard 8.8 Johns Hopkins 1.5 SUNY Buffalo 1.5
Yale 7.4 Kansas 1.5 Texas 1.5
UCSF 5.9 Laval 1.5 Texas A & M 1.5
Stanford 4.4 Lehigh 1.5 Texas Tech 1.5
Columbia 2.9 MIT 1.5 Toronto 1.5
Cornell 2.9 U. of M e d . & Dent. 1.5 Twente 1.5
Florida 2.9 Melbourne 1.5 UCLA 1.5
Indiana 2.9 Miami 1.5 US Naval Academy 1.5
NYU 2.9 Michigan 1.5 Utah 1.5
UC Berkeley 2.9 Michigan State 1.5 Virginia 1.5
UC lrvine 2.9 Minnesota 1.5 VPI 1.5
Wisconsin 2.9 Ohio State 1.5 Univ. of Washington 1.5
Bradford 1.5 Oxford 1.5 Western Connecticut 1.5
Caltech 1.5 Pennsylvania 1.5 Zurich 1.5
Denver 1.5 Purdue 1.5 Total b 100.0
Illinois 1.5 Royal Coll. of Med. 1.5

a The following abbreviations (other than those described in Tables 1 and 2) are used: UCSF, University of California at San Francisco;
NYU, New York University; SUNY Buffalo, State University of New York at Buffalo.
b Dee to rounding errors, figures may not sum to total.
Source: see Section 2.
E. Mansfield, J.-Y. Lee/Research Policy 25 (1996) 1047-1058 1051

Table 4
Mean percent of university R&D expenditures supported by industry, by size of university R&D expenditures, overall quality of faculty,
and region, 200 universities with largest R&D expenditures, 1991
Region a 100 universities with largest R&D expenditures b _ mean Universities ranked 101-200 with respect to R&D
faculty rating ¢ expenditures b_ mean faculty rating ¢
Good-to-distinguished Adequate-to-good Marginal Good-to-distinguished Adequate-to-good Marginal
(mean percen0 (mean percent)
New England 8.0 7.8 - - 6.5 6.1
Middle Atlantic 7.3 5.8 4.7 d 16.0 14.5
South Atlantic 5.9 9.8 8.6 6.6 11.6 7.7
Southeast - 5.4 9.8 - - 14.8
Southwest 2.4 d 5.4 6.9 11.8 9.7 18.2
Great Lakes 6.2 5.6 - - 23.2 a 11.4
Plains 5.8 d 6.8 - - 5.5 d 7.4
Mountain 5.1 d 6.0 4.6 - 14.3 d 11.1
Pacific 4.6 3.2 - 0.9 ~ 1.8 d _
Mean 5.7 6.2 6.9 6.4 11.1 11.4

a For the states included in each region, see National Science Foundation (1993a).
b Universities for which no faculty ratings are available had to be excluded.
c Faculty quality is categorized as good-to-distinguished, adequate-to-good, or marginal. See Footnote 9.
a Only one university.

R & D p e r f o r m e d by the university in the relevant neering, and electrical e n g i n e e r i n g ) 9 on the percent-
area, which in turn is i n f l u e n c e d by h o w m u c h R & D age o f a u n i v e r s i t y ' s R & D expenditures supported
support it r e c e i v e s f r o m industry. In the U S , a no- by industry. The results, s h o w n in T a b l e 4, indicate
table d e v e l o p m e n t o f the past decade has b e e n the that, a m o n g the 100 universities with the largest
growth in industrial support o f a c a d e m i c research. In R & D expenditures ($55 million and o v e r in 1991),
1993, industry financed about $1.5 billion o f re- region and overall quality o f faculty s e e m to h a v e a
search and d e v e l o p m e n t p e r f o r m e d at c o l l e g e s and limited effect. In no case does the m e a n percentage
universities, w h i c h was m o r e than triple the 1984 o f university R & D expenditure supported b y indus-
figure. W h i l e industry still supports only about 7 %
o f university R & D , there is a feeling in m a n y quar-
ters that the universities m u s t l o o k to industry for a
s
larger share o f their R & D support. 9 The National Academy of Sciences (1982) provides ratings of
In this and subsequent sections, we study the chemistry, chemical engineering, biochemistry, and computer sci-
ence departments. These ratings are based on the following scale:
factors influencing the extent o f industrial support o f
0 (not sufficient for doctoral education), I (marginal), 2 (ade-
a c a d e m i c research. T o b e g i n with, w e c o m b i n e data quate), 3 (good), 4 (strong), and 5 (distinguished). We designate
obtained by the National S c i e n c e F o u n d a t i o n (1993a) universities with mean ratings (in the departments where they are
and the National A c a d e m y o f S c i e n c e s (1982) to rated) of 3.0 or more as 'good-to-distinguished'; those with mean
investigate the effects o f region, size o f R & D e x p e n - ratings of 2.0-2.99 as 'adequate-to-good'; and those with ratings
below 2.0 as 'marginal.' While these departments are of obvious
diture, and overall quality o f faculty (as m e a s u r e d by
importance, they are not the only ones that are relevant (or that are
the National A c a d e m y o f S c i e n c e s ratings in c h e m - rated by the National Academy of Sciences). But given that there
istry, c o m p u t e r science, b i o c h e m i s t r y , c h e m i c a l engi- is a high correlation between a university's rating in one field of
science and technology and its rating in another such field, this
measure seems to be an adequate indicator of the perceived
overall quality of a university's faculty. Obviously, however, i t
cannot tell us how much variation exists among departments.
8 For example, see "Walker tells universities to look for help Some universities with low ratings on the average may have
to industry", Science, March 17, 1995, p. 1590. particular departments that are perceived to be quite good.
1052 E. Mansfield, J.-Y. Lee~Research Policy 25 (1996) 1047-1058

try exceed 10%. However, among the universities expected to be inversely related to Dj. The smaller
ranking in the second hundred by R & D expendi- the distance b e t w e e n t h e university and the finn, the
tures, and with faculties rated adequate-to-good or easier and cheaper it is for academic and finn per-
marginal, this mean percentage tends to be much sonnel to interact and work together on a face-to-face
higher, sometimes exceeding 20%. According to the basis. (Interview studies suggest that this factor may
National Science Foundation (1993a), the universi- be important, 13 but some R & D executives claim
ties that depend most heavily on industry tend to be that advances in telecommunications have reduced
smaller institutions with a single R & D specialty its significance. 14) Also, if a firm has been estab-
related to local industry. 10 lished near a university by faculty members, perhaps
as a spinoff of academic R & D , it is likely to support
R & D at the nearby university which played a role in
4. D i s t a n c e , faculty quality, a n d the extent o f a its birth, t5 Further, personal ties and community and
firm's support of a particular university regional pride (and politics) may induce a firm to tilt
its decisions in favor of local universities.
Turning to the effects of geographical location Holding distance constant, the probability that a
and faculty quality on the proportion of a f i n n ' s finn supports R & D at a particular university would
support for academic research o f a given type going generally be expected to be directly related to Qij,
to a particular university, the following simple model up to some point. But beyond this point, increases in
is proposed: faculty quality may not be worth the additional costs
they entail. Some types of R & D can be carried out
Ilij-= fi( Oj, Qij .... ) (1)
about as well by a merely good scientist or engineer
where H i j is the proportion of a firm's expenditures as by a Nobel laureate, and whereas some firms may
for academic R & D of the ith type that is received support R & D at leading universities to obtain access
by the j t h university, Dj is the distance (in miles) of to particularly promising students whom they will
the j t h university from this finn, 11 and Qij is the attempt to hire, many other finns do not find this
quality (as measured b y the National A c a d e m y of worthwhile, particularly since the highest rated uni-
Sciences, 1982) of the faculty in this university's versities may impose conditions on industrial support
department that would carry out this type of R & D . 12 that are far more stringent than those imposed by
Other things equal, the probability that a finn less prestigious universities. 16
supports R & D at a particular university would be To study the relationship in Eq. (1), detailed data
were obtained from senior executives at nine major
finns that together account for about 15% of the total
R & D expenditures in the chemical, computer,
Jo National Science Foundation (1993a, p. 137). The percentage petroleum, and pharmaceutical industries. Each firm
of R&D expenditures supported by industry was 7.1 at private
universities and 6.8 at public universities in 1991; thus, there was
little difference, on the average, between them.
J Of course, distance is not the only relevant factor. The entire
geographical distribution of universities in relation to the finn ,3 The R&D executives interviewed by Peters and Fusfeld (1982)
may be of relevance. For example, a finn located near several and by the Government-University-Industry Research Roundtable
good universities may be less likely ~ fund research at a remote (1991) pointed out that this factor was sometimes important.
location than a finn that is in a more isolated location. (Also, in J4 At the beginning of this study, we interviewed a substantial
some instances, the likelihood that a firm supports R&D at a number of executives in the Northeast, some of which expressed
particular university may be influenced by whether any of its this opinion. See Footnote 21 below.
rivals is supporting R&D at this university.) Nonetheless, distance 15For example, see Dorfman (1983).
is frequently alleged to be of primary significance, and it is ~6The academic R&D supported by finns is of a wide variety of
unquestionably important to determine how large its effects seem types, ranging from very fundamental to rather routine work. One
tobe. would expect that the value of Qi2 that maximizes Ilij is higher
,2 For these departments, see Footnote 6. Note that Qo' unlike for fundamental, ambitious types of R&D than for relatively
the overall measures of faculty quality in Table 4, pertains only to routine, mundane types. More will be said on this score below.
the department that would carry out this type of R&D. Also, for some relevant discussion, see Peters and Fusfeid (1982).
E. Mansfield, J.-Y. Lee/Research Policy 25 (1996) 1047-1058 1053

Table 5
Mean proportion of firms' expenditure for academic R&D received by some university with designated location and NAS faculty quality
rating, 20 types of R&D, chemical, petroleum, pharmaceutical, and computer industries, 1980-91
Distance NAS faculty quality rating Total
Good-to-distinguished Adequate-to-good Marginal
Applied R & D ( 18 types combined)
Less than 100 miles 0.27 0.23 0.13 0.63
100 to 1000 miles 0.11 0.09 0.03 0.23
Greater than 1000 miles 0.08 0.05 0.01 0.14
Total 0.46 0.37 0.17 1.00
Basic research (2 types combined)
Less than 100 miles 0.43 0.08 0.04 0.55
100 to 1000 miles 0.20 0.03 0 0.23
Greater than 1000 miles 0.20 0.02 0 0.22
Total 0.83 0.13 0.04 1.00

was asked to pick at r a n d o m two or three important m e a n a category o f R & D project such as a particular
types o f academic R & D it supported during the type of p o l y m e r synthesis or catalysis research. In
1980s and early 1990s, and to estimate the propor- defining such a type, firms were asked to include
tion of its expenditures on this type of academic reasonably similar projects which were o f impor-
R & D going to universities at various locations (less tance to the firm.
than 100 miles away, 1 0 0 - 1 0 0 0 miles away, and
more than 1000 miles away) and with various levels
of faculty quality (good-to-distinguished, adequate- 5. Effects o f distance
to-good, or marginal, according to the National
A c a d e m y o f Sciences, 1982). 17 A v e r a g i n g over all types of R & D (but separating
For all firms c o m b i n e d , data were obtained for 20 basic research from applied R & D ) , we find that, as
types of R & D , two o f which were basic research expected, distance matters. H o l d i n g constant the N A S
(from the f i n n ' s vantage point and based o n the faculty quality rating, the m e a n proportion o f R & D
National Science F o u n d a t i o n ' s definition), the rest supported at some university less than 100 miles
being applied research. 18 By a ' t y p e ' o f R & D , we away is m o r e than double that at some university
located 1 0 0 - 1 0 0 0 miles away, and (with one excep-
tion) 19 m o r e than triple that at some university more
~7Another way to interpret each of these proportions is that it is than 1000 m i l e s a w a y (Table 5). Distance is particu-
the probability that the firm will support a dollar of R&D of this larly i m p o r t a n t for universities with o n l y adequate-
type at universities with the designated location and NAS faculty
to-good or m a r g i n a l faculties; for such universities,
quality rating. Thus, the figures in Table 5 and Table 6 can be
interpreted as mean values of such probabilities. the chances o f support are quite low unless they are
In computer science, 2.7, rather than 3.0, is used as the cutoff w i t h i n 100 miles of the firm. But for all universities,
point between 'good-to-distinguished' and 'adequate-to-good' so taken as a whole, remoteness is far from a bar to
that the number of universities in each distance-quality category substantial support. For applied R & D , the m e a n
is not too small. Otherwise, the definitions are as given in
proportion o f R & D supported at some university
Footnote 9. As pointed out in Footnote 12, the faculty quality
measure pertains to the department that would carry out the R&D.
~s The National Science Foundation defines basic research in
industry as research that advances scientific knowledge but does ~9The exception is basic research in the case where the univer-
not have specific commercial objectives, although such investiga- sity has a good-to-distingnished faculty in the relevant department,
tions may be in fields of present or potential interest to the firm. In this case, the probability that a firm would support a project at
Practically all of the applied R&D projects in this sample were some university less than 100 miles away is 2.15 times the
applied research, not development; however, they often were probability that it would support it at some university more than
related to development projects carried out by the firm. 1000 miles away, according to Table 5.
1054 E. Mansfield, J.-Y. Lee~Research Policy 25 (1996) 1047-1058

Table 6
Mean proportion of finns' expenditure for academic R&D received by a particular university with designated location and NAS faculty
quality rating, 20 types of R&D, chemical, petroleum, pharmaceutical, and computer industries, 1980-91 a
Distance NAS faculty quality rating
Good-to-distinguished Adequate-to-good Marginal
Applied R & D (18 types combined)
Less than 100 miles 0.074 0.066 0.035
100 to 1000 miles 0.006 0.005 0.001
Greater than 1000 miles 0.007 0.005 0.001

Basic research (2 types combined)


Less than 100 miles 0.073 0.016 0.010
100 to 1000 miles 0.009 0.001 0
Greater than 1000 miles 0.013 0.001 0

a In contrast with Table 5, the sum of the figures does not equal 1. See Footnote 20.

m o r e t h a n 100 m i l e s a w a y is o v e r o n e - t h i r d ; f o r a p a r t i c u l a r u n i v e r s i t y o f e a c h kind. 20 V i e w e d this


b a s i c r e s e a r c h , it is a l m o s t o n e - h a l f . B e c a u s e there way, the e f f e c t s o f d i s t a n c e are s h o w n to b e e v e n
are so m a n y m o r e u n i v e r s i t i e s l o c a t e d in r e m o t e t h a n g r e a t e r t h a n i n d i c a t e d in T a b l e 5. F o r a p p l i e d R & D ,
n e a r b y areas, this m e a n p r o p o r t i o n is s u b s t a n t i a l w h e n the N A S f a c u l t y quality r a t i n g is h e l d c o n s t a n t ,
e v e n t h o u g h , o n the a v e r a g e , the c h a n c e t h a t a f i r m the m e a n p r o p o r t i o n o f R & D s u p p o r t e d at a p a r t i c u -
will s u p p o r t R & D at a p a r t i c u l a r r e m o t e u n i v e r s i t y lar u n i v e r s i t y less t h a n 100 m i l e s a w a y is at least ten
is v e r y small. t i m e s as g r e a t as that at a p a r t i c u l a r u n i v e r s i t y m o r e
W h e r e a s T a b l e 5 c o n t a i n s the m e a n p r o p o r t i o n o f t h a n t h a t d i s t a n c e a w a y . Note t o o that, i f a u n i v e r s i t y
R & D s u p p o r t e d at s o m e u n i v e r s i t y w i t h e a c h desig- is m o r e t h a n 100 m i l e s a w a y , it d o e s n o t s e e m to
n a t e d l o c a t i o n a n d N A S f a c u l t y q u a l i t y rating, T a b l e m a t t e r m u c h w h e t h e r the d i s t a n c e is m o r e or less
6 c o n t a i n s the m e a n p r o p o r t i o n o f R & D s u p p o r t e d at t h a n 1000 miles. 21

6. Effects of faculty quality


Table 7
Percent of support by industry and the federal government for F a c u l t y q u a l i t y , as m e a s u r e d b y the N A S ratings,
Academic R&D a going to universities with faculties rated good-
to-distinguished, adequate-to-good, and marginal, 1991 also s e e m s to b e i m p o r t a n t . H o l d i n g d i s t a n c e c o n -
stant, the m e a n p r o p o r t i o n o f R & D s u p p o r t e d at
NAS faculty Industry Federal
quality rating b Sample c Total government

Good-tn-distinguished 50 38 50 20 Each figure in Table 6 equals the corresponding figure in


Adequate-to-good 35 50 43 Table 5 divided by the number of universities in this distance-qu-
Marginal 17 12 7 ality category. If the universities in a particular category had the
Total a 100 100 100 same mean proportion, this figure would equal this mean propor-
tion; otherwise, it equals the average of the mean proportions that
a Data pertain to the 200 universities with largest R&D expendi- universities in this distance-quality category received.
tures. Universities for which no faculty ratings are available had to 2~ As pointed out in Footnote 14, we interviewed a number of
be excluded. executives, some of whom said that distance did not matter. Many
b For total industry and the federal government, the overall faculty of the rest expressed a preference for universities that could be
ratings described in Footnote 9 are used; for the industry sample, reached by automobile in a couple of hours or less. There seemed
faculty ratings in the relevant department are used. to be a bimodal distribution of responses in this regard, but no
c All 20 types of R&D in Table 5 and Table 6 are included. such bimodality showed up in the data on which Table 5 and
a Because of rounding errors, the total sometimes differs from the Table 6 are based. In practically all cases in our sample, distance
sum of the figures. seemed to matter, at least to some extent.
E. Mansfield, J.-Y. Lee~Research Policy 25 (1996) 1047-1058 1055

some university with a good-to-distinguished depart- 7. Further hypotheses and the meaning of dis-
ment in the relevant field is higher than that at some tance
university with a lower rated department (Table 5).
However, the effects of faculty quality are much Only universities with good-to-distinguished fac-
smaller for applied R & D than for basic research. ulties seem to have much chance of obtaining sup-
Among universities less than 100 miles away, the port from firms at least 100 miles away. Among
mean proportion of basic research supported at a universities with such faculties, we hypothesize that,
particular university with a good-to-distinguished the more fundamental the research is, the less dis-
faculty is about five times as large as that at a tance will matter because fewer and less intensive
particular university with an adequate-to-good fac- interactions between firm and university personnel
ulty. But for applied R&D, so long as a university is will be required, and because the technical expertise
less than 100 miles away and the faculty is rated at of the faculty (which is unlikely to be closely corre-
least adequate, faculty quality seems to have only a lated with distance) will be of greater and greater
moderate effect on the probability that a firm would importance. This hypothesis seems to be borne out
support R & D there. Indeed, the remarkable fact by the data. 23 One might also think that smaller and
seems to be that the firms in our sample tend to less R&D-intensive firms would be more inclined to
support more applied R & D at a university less than support R & D at nearby universities than larger and
100 miles away than at one beyond this distance,
even if the nearby university has only a marginal
faculty while the more remote university has a good- 23 Since relatively fundamental projects are likely to be sup-
to-distinguished faculty in the relevant department ported at good-to-distinguished departments rather than at lesser
rated ones, we begin by using the proportion of a firm's R&D of a
(Table 6). particular type supported at a university where the faculty is
Due partly to the fact that universities with facul- good-to-distinguished as a crude surrogate for how fundamental
ties rated adequate-to-good or marginal out-number this type of R&D is. Based on interviews with R&D executives of
those with faculties rated good-to-distinguished, these firms, this surrogate is reasonably trustworthy, at least for
about half of the academic R & D supported by our this sample. To test our hypothesis, we calculated the following
regression:
sample of firms seems to go to universities where the
Pi=O.74-O.29Hi, ( r 2 = 0.25)
relevant department is rated only adequate-to-good
(0.06) (0.12) (2)
or below (Table 7). For industry as a whole, the
proportion going to universities with such overall where Pi is the mean proportion of academic R&D of the ith type
supported at some university less than 100 miles away, given that
ratings seems even higher. In 1991, about 62% of the faculty is good-to-distinguished, and H i is the mean propor-
industry's support for R & D at the 200 universities tion of R&D of this type supported at a university where the
with the largest R & D expenditures went to such faculty is good-to-distinguished. Clearly, the results are in accord
universities. Given that the firms in our sample are with our hypothesis, since the regression coefficient of H i is
bigger and more technologically advanced than most negative and statistically significant. (However, only about a
quarter of the variation in P~ is explained.) Because the dependent
companies, it is not surprising that they are more variable in Eq. (2) must assume values between 0 and I, it may
inclined than industry as a whole to support universi- appear that least-squares regression is inappropriate. However, if a
ties with good-to-distinguished faculties. Note too two-limit tobit model is fit to the data, the results are basically
that the federal government has been more inclined unaffected. Also, if a logistic model is fit instead, the results are
to support R & D at universities with good-to-dis- essentially the same as in Eq. (2).
Of course, Pi must equal G i / H i , where G i is the mean propor-
tinguished faculties than industry as a whole, a point tion o f R & D oftbe ith type supported at a university less than 100
that will be discussed further in Section 8. 22 miles away with a good-to-distinguished faculty. Thus, if G i were
relatively constant (or under other circumstances), this might
account for the inverse relationship between Pi and H i in Eq. (2).
22 Columns 2 and 3 of Table 7 are based on data provided by the But if we replace H i in Eq. (2) with direct evaluations obtained
National Science Foundation (1993a) and National Science Foun- from executives of how fundamental each of these types of
dation (1993b) regarding the extent of industry and government research was, the results are essentially the same as in Eq. (2).
support for R&D expenditures at each of the 200 universities with Consequently, the evidence supporting this hypothesis seems rea-
largest R&D expenditures. sonably unambiguous.
1056 E. Mansfield, J.-Y. Lee~Research Policy 25 (1996) 1047-1058

more R&D-intensive firms, due to the latter's greater tives make decisions regarding which universities to
resources and technical sophistication. While the data support, from which of these locations distance
seem to point in this direction, the results are not should be measured. For all 20 types of R&D and
statistically significant, which may reflect the fact all firms in the sample, the answer, according to the
that all of the firms in our sample are quite large and relevant executives, was the same: distance should
technically sophisticated. 24 be measured from the firm's R & D laboratories, not
The exponential distribution is often used to rep- its headquarters or its major production facilities or
resent the distribution of time or distance. Katz its major marketing centers. If the firm has more
(1994) has used it to approximate the relationship than one R & D laboratory, distance should be mea-
between distance and rate of collaboration among sured from the laboratory, that is particularly inter-
domestic universities in the United Kingdom, ested in, and responsible for, the academic R&D in
Canada, and Australia. Our data suggest that the question. 26 This is the concept of distance underly-
probability of university support does not go down ing Tables 5 and 6.
as rapidly with increases in distance as the exponen- To see why this is the appropriate concept of
tial distribution would indicate, perhaps because these distance, it is necessary to understand the nature of
firms are more able and willing than the university academic R & D supported by industry. The bulk of
collaborators to go far afield. 25 this work is concerned with purely scientific and
Since most major firms have facilities at many technological activities that are of little immediate
locations, it is important to determine, when execu- interest to the production or marketing segments of
the firm. Academic R&D, while it can provide
important new theoretical and empirical findings,
seldom yields specific inventions or products ready
24 To test this hypothesis, the firm's sales and ratio of R&D
for production and marketing. Recognizing this fact,
expenditures to sales were used as additional independent vari-
ables in Eq. (2). Both regression coefficients have the expected firms support academic R & D to get up-to-date
sign, but neither is statistically significant. knowledge of new fields of science and technology,
25 Given the faculty quality, suppose that the probability distri- to obtain access to students (potential employees)
bution of distance between the firm and the university it supports and faculty (potential consultants), and to get an-
is exponential, which would be the case if the firm considered
swers to specific problems (as well as special kinds
universities in order of their distance and if the probability of its
finding the right one at distances between T and T + A miles was of analyses) that their own R & D laboratories cannot
equal to h/hA, where hi~ is constant for the ith type of R&D and deal with as effectively.
the kth level of faculty quality.
If this probability distribution is exponential,
F~k(T) = h i k [ l - e x p ( - h,kT)] (3) 8. Conclusions
where Fi~(T) is the probability that the firm will support R&D of
the ith type at some university less than T miles away with the
kth level of faculty quality. For each of the 20 types of R&D (that
Perhaps to a greater extent than at any time in the
is, for i = l ..... 20), the firms provided us with data regarding past 50 years, a debate is going on regarding the
Fi~(T) for T = 100 and 1000. Since Eq. (3) implies that proper role of the universities in the process of
A~k = 10Bi~ (4) technological change in the United States. 27 Our
where Aik = ln[(hi~ - Fik(1000))+ hi~] and Bik = In[(h u - findings shed new light on the contributions of uni-
Fik(100))+ hi,], one simple way to test whether Eq. (3) holds is versities of various types to industrial innovation,
to regress Aik on Bik (with the intercept constrained to equal
zero) and see whether the regression coefficient equals 10. The
and on the factors influencing industrial support for
results indicate that the exponential distribution is not the fight
model here. The hypothesis that this regression coefficient equals
10 should be rejected at practically any significance level. Another
way to test this hypothesis is to see whether the mean of Aik/Bib 26 The above question was put to senior executives at each of
equals 10. (This involves a different assumption about the nature the firms in our sample. There was complete agreement among
of the error terms than in the previous test.) If this test procedure them on this score.
is used instead, the results again indicate that the exponential 27 See Brooks (1996), Mansfield (1995a), Mansfield (1996) and
distribution should be rejected. Mansfield (to be published).
E. Mansfield, J.-Y. Lee/Research Policy 25 (1996) 1047-1058 1057

academic research. Based on our findings, which gious universities, our results suggest that it would
pertain to seven major industries, the universities be very small. Even if the percent of academic R&D
cited by firms as having contributed most signifi- supported by industry rose 3 percentage points (which
cantly to their product and process development tend would be a considerable increase) and if federal
to be the leading generators of new fundamental support fell correspondingly, the total amount of
knowledge - MIT, UC Berkeley, Stanford, Harvard, R & D supported at universities with overall ratings
Yale, and the like. Clearly, these universities have of good-to-distinguished would be only about 8 / 1 0
had a major impact on industrial innovation in the of 1% less than if no change occurred in the sources
short term, as well as over the long run. of support (assuming that the percentages in Table 7
At the same time, 'second tier' departments too remain constant). 28
seem to play a very important, and perhaps under-ap- Faced with reduced government funding for mili-
preciated, role in this regard. About 40% of the tary-related R & D and other budgetary pressures,.
academic research findings considered by these in- many universities have stepped up their efforts to
dustries as most important in product and process obtain R & D support from industry. Our findings
development during the 1980s came from universi- provide what seems to be the first direct evidence of
ties with adequate-to-good and marginal faculties, a quantitative nature regarding the effects of location
not from those with good-to-distinguished faculties. and faculty quality on their chances of success.
Also, the fact that the large science-based firms in Particularly for universities with modestly rated fac-
our sample have been almost as likely to support ulties, location seems to be of key importance. Hold-
applied R & D at a university with an adequate-to- ing faculty quality constant, the amount of applied
good faculty as at one with a good-to-distinguished R & D supported by the firms in our sample at a
faculty indicates that much of the applied R & D particular university less than 100 miles away tends
supported by industry can be done satisfactorily at to be at least ten times as great as at a more remote
less prestigious departments. university. While advances in telecommunications
Here, as in many areas, the diversity of US have been important, distance still seems to count.
universities has its advantages. To promote techno- Finally, distance also helps to determine which
logical change and industrial innovation, a mix of firms reap the economic benefits from an innovation
fundamental and applied research, as well as a vari- based on academic research. While economists and
ety of technical activities aimed at the diffusion and others sometime assume that new knowledge is a
commercialization of new knowledge, must be car- public good that quickly and cheaply becomes avail-
fled out. The major research universities have form- able to all, this is far from true. According to the
idable capacities and strengths, but at the stage where executives in our sample, firms located in the nation
finns need to interact with university personnel who and area where academic research occurs are signifi-
are willing to focus on their immediate problems and
help them apply new knowledge, less prestigious
universities may have a comparative (indeed, an
28 To see this, note that the amount of R&D supported at
absolute) advantage. According to many firms in our universities with departments rated good-to-distinguished would
sample, this is the case. fall by (0.50-0.38) times 3% divided by the percent of academic
Whether industry, itself hard pressed to reduce R&D at such universities (about 44%) - or about 8 / 1 0 of 1%. Of
R & D costs, will continue to increase its support of course, universities without ratings are excluded, but it seems
unlikely that many universities with departments that would be
academic research at the recent rate is by no means rated good-to-distinguished were not rated. If we assume that none
obvious. But if so, concern has been expressed in of the unrated universities would be rated good-to-distinguished if
some quarters that a continued increase in the per- ratings were available, the results would be essentially the same.
cent of academic R & D supported by industry (and a Under this assumption, those rated good-to-distinguished receive
corresponding decrease in the percent supported by about 32% of industry support and 42% of federal support. Thus,
the amount of R&D supported at universities with departments
the federal governmen0 may reduce support for lead- rated good-to-distinguished would fall by (0.42-0.32) times 3%
ing US research universities. While there may be divided by the percent of academic R&D at such universities
some redistribution of support in favor of less presti- (about 37%) - or about 8 / 1 0 of 1%.
1058 E. Mansfield, J.-Y. Lee~Research Policy 25 (1996) 1047-1058

cantly more likely than distant firms to have an Jaffe, A., 1989, Real effects of academic research, American
opportunity to be among the first to apply the find- Economic Review 79, 957-970.
Jaffe, A., M. Trajtenberg and R. Henderson, 1993, Geographical
ings of this research. Whether nearby firms seize this localization of "knowledge spillovers as evidenced by patent
opportunity successfully depends on a wide variety citations, Quarterly Journal of Economics CVIII, 577-598.
of factors, as is well known. But if they are reason- Katz, J.S., 1994, Geographical proximity and scientific collabora-
ably receptive and capable, the fact that they are tion, Scientometrics.
more likely than other firms to have this o~portunity Mansfield, E., 1991, Academic research and industrial innovation,
Research Policy 20, 1- 12.
can unquestionably be a major advantage. Mansfield, E., 1992, Academic research and industrial innovation:
a further note, Research Policy 21,295-296.
Mansfield, E., 1995a, Academic research underlying industrial
References innovations: sources, characteristics, and financing, Review of
Economics and Statistics (February) 55-65.
Adams, J., 1990, Fundamental stocks of "knowledge and productiv- Mansfield, E., 1995b, University-Industry R and D Relationships
ity growth, Journal of Political Economy 98, 673-702. (Harvard University).
Bania, N., R. Eberts and M. Fogarty, 1993, Universities and the Mansfield, E., 1996, Contributions of R and D to the Economy, in
startup of new companies, Review of Economics and Statistics B. Smith and C. Barfield (eds), Technology, R and D, and the
(November). Economy (Brookings Institution and American Enterprise In-
Brooks, H., 1996, The Evolution of the U.S. Science Policy, in B. stitute, Washington, D.C.) pp. 114-139.
Smith and C. Barfield (eds), Technology, R and D, and the Mansfield, E., to be published, Links Between Academic Re-
Economy (Brookings Institution and American Enterprises search and Industrial Innovations, in P. David and E. Stein-
Institute, Washington, D.C.) pp. 15-40. mueller (eds), A Production Tension: University-Industry Col-
Cohen, L. and R. Noll, 1994, Privatizing public research, Scien- laboration in the Era of Knowledge-Based Economic Develop-
tific American (September) 72-77. ment (Stanford University Press).
Dorfman, N., 1983, Route 128: the development of a regional National Academy of Sciences, 1982, An Assessment of Re-
high technology economy, Research Policy 12, 299-316. search-Doctorate Programs in the United States (National
Feldman, M., 1994, The Geography of Innovation (Kluwer, Dor- Academy Press, Washington, DC).
drecht). National Science Foundation, 1993a, Science and Engineering
Feller, I., 1990, Universities as engines of R and D-based eco- Indicators, 1993 (National Science Foundation, Washington,
nomic growth, Research Policy. DC).
Government-University-Industry Research Roundtable, 1991, In- National Science Foundation, 1993b, Academic Science and Engi-
dustrial Perspectives on Innovation and Interactions with Uni- neering: R and D Expenditures. Fiscal Year 1991 (National
versities (National Academy Press). Science Foundation, Washington, DC).
Peters, L. and H. Fusfeld, 1982, Current U.S. university-industry
research connections, in: National Science Board (Editors),
University-lndustry Research Relationships (National Science
Foundation, Washington, DC).
29On this topic, and others relevant to this paper, see Adams Rosenberg, N. and R. Nelson, 1994, American universities and
(1990), Bahia et al. (1993), Cohen and Noll (1994), Feldman technical change in industry, Research Policy 23, 323-348.
(1994), Feller (1990), Jaffe (1989), Jaffe et al. (1993), Mansfield Zucker, L., M. Darby and M. Brewer, 1994, Intellectual capital
(1995b), Mansfield (1996), Rosenberg and Nelson (1994) and and the birth of U.S. biotechnology enterprises, National
Zucker et al. (1994). Bureau of Economic Research, Working Paper 4653.

Você também pode gostar