Mansfield 1996 - Modern University..

research policy ELSEVIER

Research Policy 25 (1996) 1047-1058

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 Universities play a major role in originating and promoting the diffusion of knowledge and techniques that contribute to industrial innovations. In 1975-85, about 10% of the new products and processes in US high-technology industries were based directly on recent academic research. 2 Policy mak-

* Corresponding author. Part of this paper was presented at the 1995 annual meetings of the American Economic Association. The research on which this paper is based was supported by a grant to Mansfield from the National Science Foundation. Lee's contribution, focused on the survey in Section 4, was part of his work toward a doctoral dissertation at the University of Pennsylvania. We are grateful to Lewis Branscomb, William Eva_n, Diana Hicks, Leonard Lederman, and Jaewoo Ryoo for helpful comments. 2 For example, see Mansfield (1991) and Mansfield (1992).

ers, wanting to increase the economic payoff from academic research, are intensely interested in evidence regarding the characteristics of universities that have contributed most importantly to industrial innovation in various fields. Unfortunately, very little systematic study has been devoted to this important and complex topic. In this paper, we present and analyze for the first time data on this score provided by a carefully selected sample of US firms in seven major industries. Whether or not a particular university makes major contributions to the development o f innovations in a particular industry depends in part on the extent o f industrial support for research at that university. Very little is known about the factors determining which universities firms will support to do R & D o f various types. How important is geographical proximity? 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 analytical and policy viewpoints. This paper presents and analyzes data on this topic too.

2. Universities making significant contributions to US industrial innovations To obtain information regarding which universities have contributed most significantly to US industrial innovation in the electronic, information processing, chemical, petroleum, pharmaceutical, instruments, and metal industries, a random sample of 70 major firms was drawn from these industries. Each firm was asked to cite about five academic researchers whose work in the 1970s and 1980s contributed most importantly to the firm's new products and processes introduced in the 1980s. While the initial requests for data and cooperation were made to the firms' chairmen, the respondents usually were the top R & D executives who based their responses in substantial part on detailed information obtained from people at lower levels of their organizations. Most of the firms went to a considerable amount of trouble to supply these data. Written responses frequently were supplemented with interviews with relevant company personnel. Eventually usable data were obtained from 66 of the 70 firms in the sample. Since these firms account, on the average, for about a third of the R & D expenditures in these industries, the sample seems quite adequate. Taken as a whole, these 66 firms cited 321 academic researchers. 3 Tables 1 - 3 (and Footnote 4) list the universities at

3 An attempt was made to allocate the sample optimallyamong industries(that is, with sample size in each industrybeing proportional to the total number of finns in the industry times the relevant standard deviation). The industrial distribution of the finns in the sample was as follows: electronics, 14; information processing, 16; chemicals, 13; petroleum, 5; pharmaceuticals, 8; instruments,6; and metals, 4. The numberof academic researchers cited by each industrywas as follows: electronics,84; information processing, 64; chemicals, 51; petroleum,28; pharmaceuticals,47; instruments, 22; metals, 25. Eighteen academic researchers were cited by more than one finn, so the number of distinct researchers cited is 303, not 321. In Tables 1-3, each researcher is weighted by the number of firms that cited him or her.

which these researchers worked, and the percentage of citations going to each university. 4 Clearly, these citations are distributed over a considerable number of universities; the top four universities generally receive about 30% of the citations. 5 As would be expected, the most frequently cited universities tend to be w o r d leaders in science and technology. For example, in electronics and information processing, MIT, UC Berkeley, Illinois, Stanford, and CMU head the list. A statistical analysis of these data shows that there generally is a direct relationship between the quality of a university's faculty in the relevant department, as estimated by the National Academy of Sciences (1982), 6 and the percent of citations received by the university. (See Mansfield, 1995a.) However, this does not mean that 'second tier' departments do not play an important role in this regard. On the contrary, the remarkable fact is that about 40% of these citations went to universities with 'adequate-to-good' and 'marginal' faculties in the relevant departments, not to those with 'good-todistinguished' faculties, according to the National

4 TO present the data more concisely, pairs of industries are combined in Tables 1-3. Since the electronics and information processing industries are so closely related, they are combined. For the same reason, the chemical and petroleum industries are combined. Pharmaceuticalsand instrumentsare combinedbecause some of the instrument makers are in medical fields. Readers wanting data for these individualindustriescan obtain them from the authors. 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), Virginia (4), Ohio State (8), L'Ecole Central (4), Waterloo (4), Arizona (4), Missouri Rolla (4), Wright State (4), Cincinnati(4), Rhode Island (4). 5 In the metals industry, the percentage is substantiallyhigher, but the data pertain to only four finns. 6 For the electronics industry, we assume that the relevant department is electrical engineering;for informationprocessing, we assume it is computer science; for pharmaceuticals,biochemistry; for chemicals, chemistry;and petroleum, chemical engineering. Of course, this is rough, but the evidence indicatesthat these are key departments for these industries. See Mansfield(1995a). This statistical analysis was conducted only for these five industries because of the relatively small sample of finns in the remaining two industries.For detailed discussionof the National Academy of Sciences ratings, see Foomote 9.

1049

E. Mansfield, J.-Y. Lee/Research Policy 25 (1996) 1047-1058

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 of citations

University

Percentage of citations

University

Percentage of citations

MIT UC Berkeley Illinois Stanford CMU Minnesota Brigham Young Arizona State Caltech Ohio State UCLA Arizona Cambridge Chicago Columbia Connecticut Cornell Harvard Iowa Leuven

12.8 10.8 7.4 6.8 6.1 3.4 2.7 2.0 2.0 2.0 2.0 1.4 1.4 1.4 1.4 1.4 1.4 1.4 1.4 1.4

Pennsylvania Penn State Texas A& M USC Utah Aachen Calstate LA Calstate SLO Central Florida Cincinnati Delaware Drexel Florida Georgia Iowa State lrvine Kansas State Maricopa McGill Miami

1.4 1.4 1.4 1.4 1.4 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7

Michigan Missouri Newcastle New Mexico NY Polytechnic Ottawa Purdue RPI UCSD UCSB UCSC Swansi Tennessee Toronto Wake Forest Washington Univ. of Washington Wisconsin Worcester Polytech Total b

0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 0.7 100.0

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 percentage was higher in the information processing and chemical industries than in the others.) These results strongly suggest that members of 'second tier' departments are a valuable and frequently used source of research findings for industry. Geography too seems to be important. While the proportion of citations to universities outside the United States tends to be higher in the pharmaceutical, chemical, and metals industries (about 20%) than in the others (less than 10%), the bulk of the citations made by these US firms are to US universities. (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 sample tend to be cited relatively often. In all industries, there is a direct relationship between the percentage of firms in the sample that are in the same

state as a particular university and the number of citations received by that university. In electronics and information processing, about 40% of the universities cited are in the same state as the firm making the citations; in chemicals, drugs, and instruments, the figure is about 25%. 7

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

Still another factor that determines the number of citations received by a university is the amount of

7 By industry, the mean percentage of a firm's citations going to universities in the same state was: electronics, 39%; information processing, 38%; chemicals, 27%; pharmaceuticals, 23%; inslruments, 21%; metals, 19%; and petroleum, 5%. Also, see ibid.

E. Mansfield, J.-Y. Lee~Research Policy 25 (1996) 1047-1058

1050

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 of citations

University

Percentage of citations

University

MIT Washington Delaware Florida Penn State Princeton Utah Akron Case Western Res. Houston Imperial Johns Hopkins Lehigh Notre Dame Rutgers VPI

7.8 7.8 6.5 3.9 3.9 3.9 3.9 2.6 2.6 2.6 2.6 2.6 2.6 2.6 2.6 2.6

Aachen Amsterdam UC Berkeley Bowling Green Calgary Caltech Cincinnati CUNY Clarkson Cleveland State Columbia Indiana Iowa State Leuven Louisiana McMasters

1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3

Minnesota Northwestern Oxford Pomona RP1 Rheinische FW Salerno South Carolina Southern Mississippi Stanford Texas Toledo UCLA William and Mary Total b

Percentage of citations 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 1.3 100.0

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 of citations

University

Percentage of citations

University

Percentage of citations

Harvard Yale UCSF Stanford Columbia Cornell Florida Indiana NYU UC Berkeley UC lrvine Wisconsin Bradford Caltech Denver Illinois

8.8 7.4 5.9 4.4 2.9 2.9 2.9 2.9 2.9 2.9 2.9 2.9 1.5 1.5 1.5 1.5

Johns Hopkins Kansas Laval Lehigh MIT U. of M e d . & Dent. Melbourne Miami Michigan Michigan State Minnesota Ohio State Oxford Pennsylvania Purdue Royal Coll. of Med.

1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5

SUNY Buffalo Texas Texas A & M Texas Tech Toronto Twente UCLA US Naval Academy Utah Virginia VPI Univ. of Washington Western Connecticut Zurich Total b

1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 100.0

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

New England Middle Atlantic South Atlantic Southeast Southwest Great Lakes Plains Mountain Pacific Mean

100 universities with largest R&D expenditures faculty rating ¢

b _

mean

Good-to-distinguished

Adequate-to-good (mean percen0

Marginal

8.0 7.3 5.9 2.4 d 6.2 5.8 d 5.1 d 4.6 5.7

7.8 5.8 9.8 5.4 5.4 5.6 6.8 6.0 3.2 6.2

4.7 d 8.6 9.8 6.9 4.6 6.9

Universities ranked 101-200 with respect to R&D expenditures b_ mean faculty rating ¢ Good-to-distinguished 6.6 11.8 0.9 ~ 6.4

Adequate-to-good (mean percent)

Marginal

6.5 16.0 11.6 9.7 23.2 a 5.5 d 14.3 d 1.8 d 11.1

6.1 14.5 7.7 14.8 18.2 11.4 7.4 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 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 support it r e c e i v e s f r o m industry. In the U S , a notable d e v e l o p m e n t o f the past decade has b e e n the growth in industrial support o f a c a d e m i c research. In 1993, industry financed about $1.5 billion o f research 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 universities, w h i c h was m o r e than triple the 1984 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 quarters that the universities m u s t l o o k to industry for a s larger share o f their R & D support. In this and subsequent sections, we study the factors influencing the extent o f industrial support o f 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 obtained by the National S c i e n c e F o u n d a t i o n (1993a) and the National A c a d e m y o f S c i e n c e s (1982) to investigate the effects o f region, size o f R & D e x p e n diture, and overall quality o f faculty (as m e a s u r e d by 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 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-

8 For example, see "Walker tells universities to look for help to industry", Science, March 17, 1995, p. 1590.

neering, and electrical e n g i n e e r i n g ) 9 on the percentage o f a u n i v e r s i t y ' s R & D expenditures supported by industry. The results, s h o w n in T a b l e 4, indicate that, a m o n g the 100 universities with the largest R & D expenditures ($55 million and o v e r in 1991), region and overall quality o f faculty s e e m to h a v e a limited effect. In no case does the m e a n percentage o f university R & D expenditure supported b y indus-

9 The National Academy of Sciences (1982) provides ratings of chemistry, chemical engineering, biochemistry, and computer science departments. These ratings are based on the following scale: 0 (not sufficient for doctoral education), I (marginal), 2 (adequate), 3 (good), 4 (strong), and 5 (distinguished). We designate universities with mean ratings (in the departments where they are rated) of 3.0 or more as 'good-to-distinguished'; those with mean 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 importance, they are not the only ones that are relevant (or that are rated by the National Academy of Sciences). But given that there 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. Some universities with low ratings on the average may have 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 ranking in the second hundred by R & D expenditures, and with faculties rated adequate-to-good or marginal, this mean percentage tends to be much higher, sometimes exceeding 20%. According to the National Science Foundation (1993a), the universities that depend most heavily on industry tend to be smaller institutions with a single R & D specialty related to local industry. 10

4. D i s t a n c e , faculty quality, a n d the extent o f a firm's support of a particular university

Turning to the effects of geographical location and faculty quality on the proportion of a f i n n ' s support for academic research o f a given type going to a particular university, the following simple model is proposed:

Ilij-= fi( Oj, Qij .... )

(1)

where H i j is the proportion of a firm's expenditures for academic R & D of the ith type that is received by the j t h university, Dj is the distance (in miles) of the j t h university from this finn, 11 and Qij is the quality (as measured b y the National A c a d e m y of Sciences, 1982) of the faculty in this university's department that would carry out this type of R & D . 12 Other things equal, the probability that a finn supports R & D at a particular university would be

Jo National Science Foundation (1993a, p. 137). The percentage 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 may be of relevance. For example, a finn located near several good universities may be less likely ~ fund research at a remote location than a finn that is in a more isolated location. (Also, in some instances, the likelihood that a firm supports R&D at a particular university may be influenced by whether any of its rivals is supporting R&D at this university.) Nonetheless, distance is frequently alleged to be of primary significance, and it is unquestionably important to determine how large its effects seem tobe. ,2 For these departments, see Footnote 6. Note that Qo' unlike the overall measures of faculty quality in Table 4, pertains only to the department that would carry out this type of R&D.

expected to be inversely related to Dj. The smaller the distance b e t w e e n t h e university and the finn, the easier and cheaper it is for academic and finn personnel to interact and work together on a face-to-face basis. (Interview studies suggest that this factor may be important, 13 but some R & D executives claim that advances in telecommunications have reduced its significance. 14) Also, if a firm has been established 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 its birth, t5 Further, personal ties and community and regional pride (and politics) may induce a firm to tilt its decisions in favor of local universities. Holding distance constant, the probability that a finn supports R & D at a particular university would generally be expected to be directly related to Qij, up to some point. But beyond this point, increases in faculty quality may not be worth the additional costs they entail. Some types of R & D can be carried out about as well by a merely good scientist or engineer as by a Nobel laureate, and whereas some firms may support R & D at leading universities to obtain access to particularly promising students whom they will attempt to hire, many other finns do not find this worthwhile, particularly since the highest rated universities may impose conditions on industrial support that are far more stringent than those imposed by less prestigious universities. 16 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, petroleum, and pharmaceutical industries. Each firm

,3 The R&D executives interviewed by Peters and Fusfeld (1982) and by the Government-University-Industry Research Roundtable (1991) pointed out that this factor was sometimes important. J4 At the beginning of this study, we interviewed a substantial number of executives in the Northeast, some of which expressed this opinion. See Footnote 21 below. 15For example, see Dorfman (1983). ~6The academic R&D supported by finns is of a wide variety of types, ranging from very fundamental to rather routine work. One would expect that the value of Qi2 that maximizes Ilij is higher for fundamental, ambitious types of R&D than for relatively routine, mundane types. More will be said on this score below. Also, for some relevant discussion, see Peters and Fusfeid (1982).

1053

E. Mansfield, J.-Y. Lee/Research Policy 25 (1996) 1047-1058

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

0.27 0.11 0.08 0.46

0.23 0.09 0.05 0.37

0.13 0.03 0.01 0.17

0.63 0.23 0.14 1.00

0.43 0.20 0.20 0.83

0.08 0.03 0.02 0.13

0.04 0 0 0.04

0.55 0.23 0.22 1.00

Applied R & D ( 18 types combined)

Less than 100 miles 100 to 1000 miles Greater than 1000 miles Total Basic research (2 types combined)

Less than 100 miles 100 to 1000 miles Greater than 1000 miles Total

was asked to pick at r a n d o m two or three important types o f academic R & D it supported during the 1980s and early 1990s, and to estimate the proportion of its expenditures on this type of academic R & D going to universities at various locations (less 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, adequateto-good, or marginal, according to the National A c a d e m y o f Sciences, 1982). 17 For all firms c o m b i n e d , data were obtained for 20 types of R & D , two o f which were basic research (from the f i n n ' s vantage point and based o n the National Science F o u n d a t i o n ' s definition), the rest being applied research. 18 By a ' t y p e ' o f R & D , we

~7Another way to interpret each of these proportions is that it is the probability that the firm will support a dollar of R&D of this type at universities with the designated location and NAS faculty quality rating. Thus, the figures in Table 5 and Table 6 can be interpreted as mean values of such probabilities. In computer science, 2.7, rather than 3.0, is used as the cutoff point between 'good-to-distinguished' and 'adequate-to-good' so that the number of universities in each distance-quality category is not too small. Otherwise, the definitions are as given in 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 not have specific commercial objectives, although such investigations may be in fields of present or potential interest to the firm. Practically all of the applied R&D projects in this sample were applied research, not development; however, they often were related to development projects carried out by the firm.

m e a n a category o f R & D project such as a particular type of p o l y m e r synthesis or catalysis research. In defining such a type, firms were asked to include reasonably similar projects which were o f importance to the firm.

5. Effects o f distance A v e r a g i n g over all types of R & D (but separating basic research from applied R & D ) , we find that, as expected, distance matters. H o l d i n g constant the N A S faculty quality rating, the m e a n proportion o f R & D supported at some university less than 100 miles 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 exception) 19 m o r e than triple that at some university more than 1000 m i l e s a w a y (Table 5). Distance is particularly i m p o r t a n t for universities with o n l y adequateto-good or m a r g i n a l faculties; for such universities, the chances o f support are quite low unless they are w i t h i n 100 miles of the firm. But for all universities, taken as a whole, remoteness is far from a bar to substantial support. For applied R & D , the m e a n proportion o f R & D supported at some university

~9The exception is basic research in the case where the university has a good-to-distingnished faculty in the relevant department, In this case, the probability that a firm would support a project at some university less than 100 miles away is 2.15 times the probability that it would support it at some university more than 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 NAS faculty quality rating

Distance

Good-to-distinguished

Adequate-to-good

Marginal

Applied R & D (18 types combined) Less than 100 miles 100 to 1000 miles Greater than 1000 miles

0.074 0.006 0.007

0.066 0.005 0.005

0.035 0.001 0.001

Basic research (2 types combined) Less than 100 miles 100 to 1000 miles Greater than 1000 miles

0.073 0.009 0.013

0.016 0.001 0.001

0.010 0 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 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 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 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 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 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 is v e r y small. 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 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 design 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 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

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 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 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 , 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 , 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 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 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 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 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 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 t h a n 1000 miles. 21

6. Effects of faculty quality Table 7 Percent of support by industry and the federal government for Academic R&D a going to universities with faculties rated goodto-distinguished, adequate-to-good, and marginal, 1991 NAS faculty quality rating b

Industry Sample c

Total

Federal government

Good-tn-distinguished Adequate-to-good Marginal Total a

50 35 17 100

38 50 12 100

50 43 7 100

a Data pertain to the 200 universities with largest R&D expenditures. Universities for which no faculty ratings are available had to be excluded. b For total industry and the federal government, the overall faculty ratings described in Footnote 9 are used; for the industry sample, faculty ratings in the relevant department are used. c All 20 types of R&D in Table 5 and Table 6 are included. a Because of rounding errors, the total sometimes differs from the sum of the figures.

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, 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

20 Each figure in Table 6 equals the corresponding figure in Table 5 divided by the number of universities in this distance-quality category. If the universities in a particular category had the same mean proportion, this figure would equal this mean proportion; otherwise, it equals the average of the mean proportions that universities in this distance-quality category received. 2~ As pointed out in Footnote 14, we interviewed a number of executives, some of whom said that distance did not matter. Many of the rest expressed a preference for universities that could be reached by automobile in a couple of hours or less. There seemed to be a bimodal distribution of responses in this regard, but no such bimodality showed up in the data on which Table 5 and Table 6 are based. In practically all cases in our sample, distance seemed to matter, at least to some extent.

E. Mansfield, J.-Y. Lee~Research Policy 25 (1996) 1047-1058

some university with a good-to-distinguished department in the relevant field is higher than that at some university with a lower rated department (Table 5). However, the effects of faculty quality are much smaller for applied R & D than for basic research. Among universities less than 100 miles away, the mean proportion of basic research supported at a particular university with a good-to-distinguished faculty is about five times as large as that at a particular university with an adequate-to-good faculty. But for applied R&D, so long as a university is less than 100 miles away and the faculty is rated at least adequate, faculty quality seems to have only a moderate effect on the probability that a firm would support R & D there. Indeed, the remarkable fact seems to be that the firms in our sample tend to support more applied R & D at a university less than 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 goodto-distinguished faculty in the relevant department (Table 6). Due partly to the fact that universities with faculties rated adequate-to-good or marginal out-number those with faculties rated good-to-distinguished, about half of the academic R & D supported by our sample of firms seems to go to universities where the relevant department is rated only adequate-to-good or below (Table 7). For industry as a whole, the proportion going to universities with such overall ratings seems even higher. In 1991, about 62% of industry's support for R & D at the 200 universities with the largest R & D expenditures went to such universities. Given that the firms in our sample are bigger and more technologically advanced than most companies, it is not surprising that they are more inclined than industry as a whole to support universities with good-to-distinguished faculties. Note too that the federal government has been more inclined to support R & D at universities with good-to-distinguished faculties than industry as a whole, a point that will be discussed further in Section 8. 22

22 Columns 2 and 3 of Table 7 are based on data provided by the National Science Foundation (1993a) and National Science Foundation (1993b) regarding the extent of industry and government support for R&D expenditures at each of the 200 universities with largest R&D expenditures.

1055

7. Further hypotheses and the meaning of distance Only universities with good-to-distinguished faculties seem to have much chance of obtaining support from firms at least 100 miles away. Among universities with such faculties, we hypothesize that, the more fundamental the research is, the less distance will matter because fewer and less intensive interactions between firm and university personnel will be required, and because the technical expertise of the faculty (which is unlikely to be closely correlated with distance) will be of greater and greater importance. This hypothesis seems to be borne out by the data. 23 One might also think that smaller and less R&D-intensive firms would be more inclined to support R & D at nearby universities than larger and

23 Since relatively fundamental projects are likely to be supported 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 particular type supported at a university where the faculty is good-to-distinguished as a crude surrogate for how fundamental this type of R&D is. Based on interviews with R&D executives of these firms, this surrogate is reasonably trustworthy, at least for this sample. To test our hypothesis, we calculated the following regression:

Pi=O.74-O.29Hi, (0.06) (0.12)

( r 2 = 0.25) (2)

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 the faculty is good-to-distinguished, and H i is the mean proportion of R&D of this type supported at a university where the faculty is good-to-distinguished. Clearly, the results are in accord with our hypothesis, since the regression coefficient of H i is negative and statistically significant. (However, only about a quarter of the variation in P~ is explained.) Because the dependent variable in Eq. (2) must assume values between 0 and I, it may appear that least-squares regression is inappropriate. However, if a two-limit tobit model is fit to the data, the results are basically unaffected. Also, if a logistic model is fit instead, the results are essentially the same as in Eq. (2). Of course, Pi must equal G i / H i , where G i is the mean proportion o f R & D oftbe ith type supported at a university less than 100 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). But if we replace H i in Eq. (2) with direct evaluations obtained from executives of how fundamental each of these types of research was, the results are essentially the same as in Eq. (2). Consequently, the evidence supporting this hypothesis seems reasonably unambiguous.

1056

E. Mansfield, J.-Y. Lee~Research Policy 25 (1996) 1047-1058

more R&D-intensive firms, due to the latter's greater resources and technical sophistication. While the data seem to point in this direction, the results are not statistically significant, which may reflect the fact that all of the firms in our sample are quite large and technically sophisticated. 24 The exponential distribution is often used to represent the distribution of time or distance. Katz (1994) has used it to approximate the relationship between distance and rate of collaboration among domestic universities in the United Kingdom, Canada, and Australia. Our data suggest that the probability of university support does not go down as rapidly with increases in distance as the exponential distribution would indicate, perhaps because these firms are more able and willing than the university collaborators to go far afield. 25 Since most major firms have facilities at many locations, it is important to determine, when execu-

24 To test this hypothesis, the firm's sales and ratio of R&D expenditures to sales were used as additional independent variables in Eq. (2). Both regression coefficients have the expected sign, but neither is statistically significant. 25 Given the faculty quality, suppose that the probability distribution of distance between the firm and the university it supports is exponential, which would be the case if the firm considered 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 equal to h/hA, where hi~ is constant for the ith type of R&D and 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) 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 is, for i = l ..... 20), the firms provided us with data regarding Fi~(T) for T = 100 and 1000. Since Eq. (3) implies that A~k = 10Bi~ (4) where Aik = ln[(hi~ - Fik(1000))+ hi~] and Bik = In[(h u Fik(100))+ hi,], one simple way to test whether Eq. (3) holds is to regress Aik on Bik (with the intercept constrained to equal zero) and see whether the regression coefficient equals 10. The 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 equals 10. (This involves a different assumption about the nature of the error terms than in the previous test.) If this test procedure is used instead, the results again indicate that the exponential distribution should be rejected.

tives make decisions regarding which universities to support, from which of these locations distance should be measured. For all 20 types of R&D and all firms in the sample, the answer, according to the relevant executives, was the same: distance should be measured from the firm's R & D laboratories, not its headquarters or its major production facilities or its major marketing centers. If the firm has more than one R & D laboratory, distance should be measured from the laboratory, that is particularly interested in, and responsible for, the academic R&D in question. 26 This is the concept of distance underlying Tables 5 and 6. To see why this is the appropriate concept of distance, it is necessary to understand the nature of academic R & D supported by industry. The bulk of this work is concerned with purely scientific and technological activities that are of little immediate 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 for production and marketing. Recognizing this fact, firms support academic R & D to get up-to-date knowledge of new fields of science and technology, to obtain access to students (potential employees) and faculty (potential consultants), and to get answers to specific problems (as well as special kinds of analyses) that their own R & D laboratories cannot deal with as effectively.

8. Conclusions

Perhaps to a greater extent than at any time in the past 50 years, a debate is going on regarding the proper role of the universities in the process of technological change in the United States. 27 Our findings shed new light on the contributions of universities of various types to industrial innovation, and on the factors influencing industrial support for

26 The above question was put to senior executives at each of the firms in our sample. There was complete agreement among them on this score. 27 See Brooks (1996), Mansfield (1995a), Mansfield (1996) and Mansfield (to be published).

E. Mansfield, J.-Y. Lee/Research Policy 25 (1996) 1047-1058

academic research. Based on our findings, which pertain to seven major industries, the universities cited by firms as having contributed most significantly to their product and process development tend to be the leading generators of new fundamental knowledge - MIT, UC Berkeley, Stanford, Harvard, Yale, and the like. Clearly, these universities have had a major impact on industrial innovation in the short term, as well as over the long run. At the same time, 'second tier' departments too seem to play a very important, and perhaps under-appreciated, role in this regard. About 40% of the academic research findings considered by these industries as most important in product and process development during the 1980s came from universities with adequate-to-good and marginal faculties, not from those with good-to-distinguished faculties. Also, the fact that the large science-based firms in our sample have been almost as likely to support applied R & D at a university with an adequate-togood faculty as at one with a good-to-distinguished faculty indicates that much of the applied R & D supported by industry can be done satisfactorily at less prestigious departments. Here, as in many areas, the diversity of US universities has its advantages. To promote technological change and industrial innovation, a mix of fundamental and applied research, as well as a variety of technical activities aimed at the diffusion and commercialization of new knowledge, must be carfled out. The major research universities have formidable capacities and strengths, but at the stage where finns need to interact with university personnel who are willing to focus on their immediate problems and help them apply new knowledge, less prestigious universities may have a comparative (indeed, an absolute) advantage. According to many firms in our sample, this is the case. Whether industry, itself hard pressed to reduce R & D costs, will continue to increase its support of academic research at the recent rate is by no means obvious. But if so, concern has been expressed in some quarters that a continued increase in the percent of academic R & D supported by industry (and a corresponding decrease in the percent supported by the federal governmen0 may reduce support for leading US research universities. While there may be some redistribution of support in favor of less presti-

1057

gious universities, our results suggest that it would be very small. Even if the percent of academic R&D supported by industry rose 3 percentage points (which would be a considerable increase) and if federal support fell correspondingly, the total amount of R & D supported at universities with overall ratings of good-to-distinguished would be only about 8 / 1 0 of 1% less than if no change occurred in the sources of support (assuming that the percentages in Table 7 remain constant). 28 Faced with reduced government funding for military-related R & D and other budgetary pressures,. many universities have stepped up their efforts to obtain R & D support from industry. Our findings provide what seems to be the first direct evidence of a quantitative nature regarding the effects of location and faculty quality on their chances of success. Particularly for universities with modestly rated faculties, location seems to be of key importance. Holding faculty quality constant, the amount of applied R & D supported by the firms in our sample at a particular university less than 100 miles away tends to be at least ten times as great as at a more remote university. While advances in telecommunications have been important, distance still seems to count. Finally, distance also helps to determine which firms reap the economic benefits from an innovation based on academic research. While economists and others sometime assume that new knowledge is a public good that quickly and cheaply becomes available to all, this is far from true. According to the executives in our sample, firms located in the nation and area where academic research occurs are signifi-

28 To see this, note that the amount of R&D supported at universities with departments rated good-to-distinguished would fall by (0.50-0.38) times 3% divided by the percent of academic R&D at such universities (about 44%) - or about 8 / 1 0 of 1%. Of course, universities without ratings are excluded, but it seems unlikely that many universities with departments that would be rated good-to-distinguished were not rated. If we assume that none of the unrated universities would be rated good-to-distinguished if ratings were available, the results would be essentially the same. Under this assumption, those rated good-to-distinguished receive about 32% of industry support and 42% of federal support. Thus, the amount of R&D supported at universities with departments rated good-to-distinguished would fall by (0.42-0.32) times 3% divided by the percent of academic R&D at such universities (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 opportunity to be among the first to apply the findings of this research. Whether nearby firms seize this opportunity successfully depends on a wide variety of factors, as is well known. But if they are reasonably receptive and capable, the fact that they are more likely than other firms to have this o~portunity can unquestionably be a major advantage.

References Adams, J., 1990, Fundamental stocks of "knowledge and productivity growth, Journal of Political Economy 98, 673-702. Bania, N., R. Eberts and M. Fogarty, 1993, Universities and the startup of new companies, Review of Economics and Statistics (November). Brooks, H., 1996, The Evolution of the U.S. Science Policy, in B. Smith and C. Barfield (eds), Technology, R and D, and the Economy (Brookings Institution and American Enterprises Institute, Washington, D.C.) pp. 15-40. Cohen, L. and R. Noll, 1994, Privatizing public research, Scientific American (September) 72-77. Dorfman, N., 1983, Route 128: the development of a regional high technology economy, Research Policy 12, 299-316. Feldman, M., 1994, The Geography of Innovation (Kluwer, Dordrecht). Feller, I., 1990, Universities as engines of R and D-based economic growth, Research Policy. Government-University-Industry Research Roundtable, 1991, Industrial Perspectives on Innovation and Interactions with Universities (National Academy Press).

29On this topic, and others relevant to this paper, see Adams (1990), Bahia et al. (1993), Cohen and Noll (1994), Feldman (1994), Feller (1990), Jaffe (1989), Jaffe et al. (1993), Mansfield (1995b), Mansfield (1996), Rosenberg and Nelson (1994) and Zucker et al. (1994).

Jaffe, A., 1989, Real effects of academic research, American Economic Review 79, 957-970. Jaffe, A., M. Trajtenberg and R. Henderson, 1993, Geographical localization of "knowledge spillovers as evidenced by patent citations, Quarterly Journal of Economics CVIII, 577-598. Katz, J.S., 1994, Geographical proximity and scientific collaboration, Scientometrics. Mansfield, E., 1991, Academic research and industrial innovation, Research Policy 20, 1- 12. Mansfield, E., 1992, Academic research and industrial innovation: a further note, Research Policy 21,295-296. Mansfield, E., 1995a, Academic research underlying industrial innovations: sources, characteristics, and financing, Review of Economics and Statistics (February) 55-65. Mansfield, E., 1995b, University-Industry R and D Relationships (Harvard University). Mansfield, E., 1996, Contributions of R and D to the Economy, in B. Smith and C. Barfield (eds), Technology, R and D, and the Economy (Brookings Institution and American Enterprise Institute, Washington, D.C.) pp. 114-139. Mansfield, E., to be published, Links Between Academic Research and Industrial Innovations, in P. David and E. Steinmueller (eds), A Production Tension: University-Industry Collaboration in the Era of Knowledge-Based Economic Development (Stanford University Press). National Academy of Sciences, 1982, An Assessment of Research-Doctorate Programs in the United States (National Academy Press, Washington, DC). National Science Foundation, 1993a, Science and Engineering Indicators, 1993 (National Science Foundation, Washington, DC). National Science Foundation, 1993b, Academic Science and Engineering: R and D Expenditures. Fiscal Year 1991 (National 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). Rosenberg, N. and R. Nelson, 1994, American universities and technical change in industry, Research Policy 23, 323-348. Zucker, L., M. Darby and M. Brewer, 1994, Intellectual capital and the birth of U.S. biotechnology enterprises, National Bureau of Economic Research, Working Paper 4653.