^^0m^ HD28 .M414 noJW ALFRED P. WORKING PAPER SLOAN SCHOOL OF MANAGEMENT Institutional Variations in Problem Choice and Persistence Among Pioneering Researchers Kocnraad Debackere Gent Rijksuniversiteit December 1991 Michael A. Rappa Massachusetts Institute of Technology Sloan WP # 3389-92 MASSACHUSETTS INSTITUTE OF TECHNOLOGY 50 MEMORIAL DRIVE CAMBRIDGE. MASSACHUSETTS 02139 1 Massachusetts Institute of Technology INSTITUTIONAL VARIATIONS IN PROBLEM CHOICE AND PERSISTENCE AMONG PIONEERING RESEARCHERS Michael A. Rappa Koenraad Debackere Gent Massachmnts Rijksuniversiteit Institute of Technology December 1 99 1991 Sloan WP # 3389-92 MASSACHUSETTS INSTITUTE OF TECHNOLOGY Alfred P. Sloan School of Management Massachusetts Institute of Technology 50 Memorial Drive, E52-538 Cambridge, MA 02139-4307 institutional variations in problem choice and Persistence among Pioneering researchers DEBACKERE Kocnraad and Michael A. RAPPA''' Massachusetts Institute of Technology December 1991 abstract This paper examines institutional variations in the factors that influence pioneering scientists in their choice and persistence in an area of research. presented from an international survey of more than sevenhundred researchers working in the field of neural networks. Elaborating Evidence is on previous which find that pioneers results, their motivations to enter and differ from their peers in their persistence to remain in the field, both this academic pioneers who merit special distinction. In general compared with other scientists, academic pioneers are more deeply influenced by the intrinsic intellectual appeal of the field and are less study suggests that it is influenced by the social dynamics of the research community. Pioneers in commercial and government institutions are not found to he dissimilar from their colleagues scientific who enter the field afier it gains legitimacy within the community. INTRODUCTION The question of why to understanding the set scientists choose to pursue the topic of research they do emergence of new fields central is of science and technology. By selecting one of problems and not another, scientists can thereby steer the frontier of knowledge new and different directions. The "problem of problem sociologists of science, involves sorting scientists in relative significance it is Of known may through the myriad of forces that their decisions to pursue an area of research. determining the scientific choice," as in to influence particular concern is of factors that are internal and external to the community (Merton, 1938; Cole and Cole, 1973; Zuckerman, 1978; 1978; and Ziman, 1987). While some have suggested cognitive factors as Giery-n, the principle force guiding scientists' choices, others have pointed to the strength of social processes within the research community, such as the competition for funding, recognition and rewards (Hagstrom, 1965). Other explanations have focused on external faaors, such influence of military and commercial interests, as well as other forces that immediate realm of the ' Koenraad Debickere Rijltsuniversiteit Sloan School, is scientific beyond the community. a Fulbright post-doctoral fellow at Gent, Belgium. Michael Rappa MIT. lie as the is MIT and a research associate at the Vlcrick School, an assistant professor in the behavioral and polic)' sciences area of the One manner in which approach the problem of problem choice to distinction between pioneering scientists who Pioneers are those scientists and the initiate of practicing large majority and continue working of new fields scientists. in a field before perceived as significant or perhaps even legitimate by their peers. unconventional problem choices, pioneering to recognize the is By it is virtue of their provide the impetus for the creation scientists of research. Therefore, deciphering the faaors that contribute to pioneers' problem choices may hold special relevance to understanding progress in science and technology. In an earlier paper we examine enter and persist in a field (Rappa the factors influencing pioneers in their decisions to and Debackere, 1991). Comparing pioneers who networks research with their colleagues the scientific — group even enter after the field becomes legitimate within community, we show empirical evidence when conform more for viewing pioneers as a distinct controlling for age and professional experience. Briefly stated, pioneers problem choice: they closely to traditional explanations of more by cognitive in neural forces and less by social forces than their colleagues. For example, the choice of neural networks as a problem area, pioneers are intrinsic intelleaual appeal of the while field; at are influenced more influenced by the same time pioneers are by the aaions and opinions of other researchers, the less in the influenced availability of funding, and the potential for financial rewards. Similarly, pioneers exhibit a far stronger attachment to the field. They express that might lead less them concern than their colleagues do with respect to a host of factors to leave the field, such as a lack of funding for their research, and rapid progress expect, pioneers espouse a longer term By of progress in neural networks, a lack in alternative fields. Finally, as commitment to the field. taking into consideration pioneers in the problem of problem choice, show how both cognitive and social processes can operate within a research different times and with different individuals. TTie social processes in the are less problem choices of pioneers one might relative influence is we are able to community of cognitive versus a reasonable finding. That pioneers concerned with the external validation of their problem choices and are content follow their own judgment, no matter how misguided it might seem conclusion that supports general perceptions of pioneering behavior. at to colleagues, It is is to a precisely this willingness not to heed the criticisms of colleagues that enables pioneers to remain steadfast in their commitment to their research interests. concerned about colleagues' criticisms of their work are mainstream in order to pioneer new fields. Researchers less likely to who are overly depart from the 9 But what made of be to is community? Although our previous study scientists' problem choices and persistence from outside the originate that factors scientific may hints to the external forces that influence adequate consideration has yet to be in a field, given to this issue. Both indtistry and government are prominent players within the research establishment, in whose influence cannot be ignored. terms of the dollar on billion spent amount of basic More than research performed. and applied research two-thirds of the $45. 1989 was performed by non-academic in As might be expected, the presence of industry and government institutions. pronounced with respea to applied research. But even so, is more more than one-third of the $18.6 1989 on basic research alone was performed outside of university spent in billion In the U.S., this caji be seen very clearly laboratories (National Science Foundation, 1989). The be seen potential influence of industry and government on the direaion of science can also terms of employment. Figures for 1987 show that more than one-fifth of in doctoral scientists and engineers non-academic institutions. who were engaged who were engaged in basic research Furthermore, of the 806,200 doctoral in research and development aaivities all were employed by and engineers scientists in 1987, a staggering 85 percent of them were employed by industry (National Science Board, 1989). Given the tremendous volume of research conduaed laboratories and the examination of military scientists' in industrial from and employed, it is problem choices must be considered and other external internal cognitive when, large percentage of doctorates in industrial and government not surprising that any in light of commercial, influences. Frequently, these interests can conflict with the social influences of the scientific community. Difficulties arise order to preserve commercial advantage or national security, scientists in and military laboratories need their colleagues in the rest to maintain a level of secrecy that divides them of the research commimity (Allen, 1977; Kornhauser, 1962; Pclz and Andrews, 1966). Numerous historical studies (1988) work on Du of industrial laboratories, such Pont, have described clash with scientists' perceptions of Pruitt, 1990). It the scientific is what is how community and scientists, seek to adhere to management. Whereas members of the who The Swann, 1988; Graham and see themselves as members of at odds with community value openness, originality its scientific also: norms, find themselves and professional autonomy, managers seek confidentiality, control over the direction of research. Hounshell and Smith's the commercial concerns of managers can important (see not unusual for industrial as practicality and corporate confusion that can result from this dichotomy was forthrighrly expressed to Du Wallace Carothers, who would Font's scientists. Criticizing the firm's policies for research is young organic chemist, the become one of the later The only guide we have management by firm's most distinguished guiding research, Carothers wrote: own formulating and criticizing our in show the rather desperate feeling that they should a profit at As a result, I think that our problems are being undertaken in a of uncertainty and skepticism without any faith in a successful outcome or even without any clear idea of what would create a successful the end. spirit outcome (Hounsheli and Smith, 1988). The pressure to make problem choices in isolation from their colleagues and that reflect the immediate profit goals of the firm, is as if to leave industrial researchers to explore the fi-ontier of science without a compass. For some, the divergent satisfactorily: the "industrial scientist" Each industrial. is of commerce and science can never be resolved interests a world onto itself, an oxymoron. Science is is academic; technology is bridged together only tenuously by the passage of students from one side to the other and by a rare breed of technologist called "gatekeepers," for which Allen (1977) has provided extensive empirical support. problem of problem choice then is an issue of concern not for the industrial technologist (n^ but for management. scientist), and industry Alternatively for others, the connection between science tenuous enough. Relatively young industries such as some fields, scientists are readily not nearly if not to a fault. aware of the pecuniary value of their work through consulting activities, ownership participation in of their research. Not to be is biotechnology and microelearonics are heavily rooted in science, and scientists' commercial motives are obvious, In The new venture left-out, universities firms, and industrial funding have increasingly sought to exploit the commercial potential of faculty research by patenting inventions and licensing them industry. To the scientific purist, this been led astray and how is a clear demonstration of how modern to science has the problem of problem choice has been transformed into a profit- maximizing funaion. To what constraints influences extent are the problem choices of scientists influenced by the institutional under which they must work? Clearly, further investigation of external on substantially scientists' problem choices is warranted. Having from those researchers who enter a primary concern here will be to field after it shown that pioneers differ becomes legitimated, our examine the influences on problem choice and persistence across different kinds of institutions, homogeneous group and across academic, to ask, in panicular, whether or not pioneers commercial and government are a institutions. PIONEERS IN NEURAL NETWORKS RESEARCH We which selected as the basis for this examination the field of neural networks research, is one of nearly a dozen concurrently. The of science and technology that fields we arc studying decision to examine neural networks holds no special significance other than the opportunity to do so presented we concluded neural network field, it itself first. would be After a preliminary investigation of the interesting to condua a comprehensive study with a primary focus on pioneering researchers. A of the neural network human brain. certain features that is a type of information processing system that By using make it example, a neural network a biological unique is in model in its is inspired by models design, a neural network system has form and function from conventional computers. For not programmed in the usual sense, but rather it is trained with data. This implies that the computational performance of a neural network improves with experience: as processes it more and more information becomes increasingly more accurate in parallelism in processing a task. Unlike a its in performing a response. Another feature is its task, it degree of normal computer with a single or small number of sophisticated central processing units, a neural network has a very large number of simple processing elements that operate simultaneously on a computational problem. These features allow it perform certain tasks that otherwise might be very to difficult using existing computer technology. Neural networks are also referred to as connectionist systems, adaptive systems, or neurocomputers. For further details, refer to the recent repon by DARPA (1988). Neural networks have a long history of development, stretching back to theoretical explanations of the brain and cognitive processes proposed during the 1940s. In the early years, researchers formulated and elaborated upon basic models of neural computing that they then used to explore random networks. By the most phenomena such as adaptive stimulus-response relations the 1960s there were several efforts to implement neural networks, nouble being the single-layer "perceptron." Among neural network researchers the perceptron was considered a watershed, but at the same time for criticism intelligence. from researchers more interested The in in it served as a lightning rod the burgeoning field of artificial idea of neural networks, as exemplified by the perceptron, quickly became seen almost antithetical to the symbolic reasoning principles of as intelligence. Critical analysis to of the perceptron led some highly respeaed AI researchers proclaim that the concept was fundamentally flawed, and researchers to waste neural networks larger much may have casting doubt as to inappropriate for legitimacy, antagonists of its' cffeaively dissuaded other researchers from entering the field in controversy surrounding neural networks notwithstanding, the early 1970s with perhaps no Undeterred in their belief more than light by researchers a few work continued during hundred researchers worldwide in the of the potential of neural networks, their persistence over the next decade eventually paid-off. new By effort on. as such, numbers (Minsky and Papert, 1988). The field. artificial By the 1980s, neural networks began to be viewed in a of disciplines, such that the in a variety community. position of legitimacy within the scientific A field soon achieved a professional society for neural network researchers was formed, specialized journals and books were published, and the first in a series why exactly of international conferences were held. While it is difficult to explain the perception of the field changed so dramatically, at least four important technical events can be discerned: (1) the evolution of the single-layer perceptron into a multi-layer system; (2) the rapid development of related technologies that enabled researchers to develop, simulate, and diagnose neural networks of greater sophistication; (3) significant progress in theoretical understanding of neuro-biological processes; and (4) the contributions of researchers pursuing the idea of parallel distributed processing, the socalled PDP-group. In became widespread, such expanded rapidly. By from a few hundred to The of these developments, light that the number of emerging we have found that several fields in it researchers working on neural networks the end of the decade the size of the field swelled in is membership thousand researchers worldwide. evolution of the neural network research typical of as well as others, interest in the field some of fairly its common community is not unusual and sociological characteristics. for new fields to lack may From our even be research, widespread acceptance for long periods, sometimes attracting controversy, other times simply being ignored by researchers. But when they do catch on, fields tend to grow rapidly. This pattern has occurred, to greater or lesser extent, in each of the dozen fields Given the recent experience within the excellent opportunity to relative to large examine numbers of field of neural networks, in greater detail the researchers who we have examined this case presents us so far. with an behavior of pioneering researchers follow in their footsteps. METHOD Through an we conference proceedings for the two-year period from 1988 to 1989, identified we were Given the scope of the research community, questionnaire was determined to be the most appropriate method of activities, (b) their decision to begin working on neural networks, might lead them to cease their neural network research interaaion with the demographic Additional characteristics. tests among arising rest (a) their in those respondents for A neural faaors that (c) of another problem area, of the neural network research community, and The were conduaed in favor a survey investigation. twelve-page questionnaire in English was sent to researchers inquiring about (d) their this able to determine the exact address for each of 2,037 researchers in thirty-five different countries. network more From than 3,000 researchers worldwide working on the subject of neural networks. material, and analysis of published sources, including books, journal anicles, (e) their questionnaire was pretested in the United States. Europe to reduce potential interpretation^ whom English Since there were thirty-seven researchers with is a difficulties second language. more than one address during the time period considered, a total of 2,074 questionnaires were mailed in February 1990. After the third week of data collection, on computer bulletin boards we mailed a follow-up letter to alert neural and posted e-mail messages network researchers of the survey. questionnaires, 162 were returned as undelivered by the post office. seven researchers with more than one address were represented Of the None of in 2,074 the thirty- the undelivered questionnaires. At the completion of the data colleaion period approximately ninety days later, 720 of the 1,875 questionnaires presumed to be delivered returned, yielding a final response rate of 38.4 percent. Some of the were completed and faaors that may have affected the response rate include: the length of the questionnaire, the global scope of the survey, and the institutional mobility of researchers. DATA Validity checks In order to rule out apparent self-seleaion biases, made to determine whether the respondent sample depans significantly from the survey population. First, a geographic comparison was and the survey population into four Middle demographic comparisons were East. Of the made by clustering the respondent sample regions: the Americas, Europe, the Far East, 720 respondents, 63 percent reside in the Americas (all and the but a few percent in the U.S.), 25 percent in Europe, ten percent in the Far East, and about two p>erccnt in the A second Middle East (x^=5.24, test d.f.=3, n.s.). compared the respondent sample and survey population with respect to the type of institutional affiliation. Respondents were classified into three categories: universities, commercial firms, or other types of institutions (mostly government funded Among laboratories that are not university-based). 720 respondents, the percent) are affiliated with academic laboratories, no reveals that statistically significant sample and the survey population (x^=5.6l, d.f =2, A compared the final, albeit less precise, test 452 (63 177 (25 percent) are employed commercial firms, while 91 (12 percent) are engaged comparison a total of in other types of institutions. in The departure exists between the respondent n.s.). disciplinary background of the sample respondents with those of the survey population. Although respondents indicated their disciplinary backgrounds, for the survey population disciplines fi-om their postal address when a we were only departmental affiliation careful inspection of the survey population, the disciplinary researchers were found. Using this data, when comparing respondents with among respondents percent), we were unable the survey population. for about 1,500 to find a significant difference The disciplines most represented include electrical engineering (36 percent), physical sciences (19 computer science (18 percent), biological sciences and engineering (7 percent), (5 percent). characteristics The 720 respondents purpose of this analysis, are we employed classified universities at in 383 different institutions worldwide. For the respondents by their institutional affiliation into three categories: academic, commercial, based was provided. Upon background mathematics (7 percent), and psychology and cognitive science Sample able to infer researchers' and government. Academic institutions include and federally-funded research and development laboratories (FFRDCs) that universities. Commercial operate for-profit, ranging from category, government, of the survey. It is institutions include public new and private firms that ventures to multinational corporations. somewhat more are The last diverse in charaaer given the international scope includes laboratories that are run as an arm of government agencies (such as the military), government funded laboratories unaffiliated with universities, non-profit organizations, and quasi-public firms associate with military research. classifying such a The difficulties in wide variety of institutions should not be minimized; and though it is — certainly possible to devise a more numbers elaborate classification scheme, the small cell that result will ultimately render statistical procedures inappropriate. A number of steps were taken in order to ensure the comparability of the three types of institutions in the present analysis. First, given their special status experience at the time of the survey, wc omincd doaoral of 1990) from the analysis. Since our focus make allow us to scientists. who member, faculty title a5 a who academic typically omission this who report their formal position few respondents report position appointments in scientists who hold secondary number of respondents more than one type of institution positions as managers or industrial affiliated with organizations that do not within the domain of the proposed topology were dropped from the analysis. As a result of these steps, the institutional distribution: (11 The academic in employed by a single firm. industrial respondent in is represent is laboratories, with Tukey-HSD and Schefifif tests in |ii=|lj (for all ANOVA an i, a=.05). j, find any significant differences across institutions with respect to (ANOVA of years of professional experience, which is elapsed since the respondent last graduated, for 32 39.4 years old (s.d.=8.9 years), the average 39.3 years old (s.d.=9.3 years). we could not 9.9 The 37.2 years old (s.d.=8.4 years), and the average government the respondent's professional experience years), ten. any single organization being seven. (F=2.58, p=.08) do not allow us to rejea the null-hypothesis that Similarly, 174 different any single university being The government respondents average academic respondent is in in industry are employed in 56 different firms, with fourteen of them most respondents employed respondent employed researchers are with the most respondents employed respondents The sample was reduced to 452 cases with the following 286 (63 percent) academic, 115 (25 percent) commercial, and percent) government. universities, the those respondents scientist or engineer. TTierefore, the relatively consultants. TTiird, a small 51 on understanding the significance of report their primary positions as managers or consultants were omitted, as well as those respondents fall (as between academic and non-academic a clearer comparison we included only Second, students and recent graduates on the problem choice and persistence of pioneers, institutional affiliation will is and lack of professional F=1.16, p=.31, defined as the is The number of average years), and 11.9 number years that have 10.8 for academic respondents commercial respondents (s.d.=7.4 respondents (s.d.=9.0 years). n.s.). for (s.d. = 8.7 government 10 The number of years that respondents have been involved in the field of neural networks Academic respondents have on average 7.5 differs significantly across institutions. (s.d.=6.9 years) in the field, commercial respondents have 4.5 years (s.d.=4.5 government respondents have Tukey-HSD and Scheff(f tests and commercial instit-utions. 6.1 years (s.d.=5.8 years): ANOVA years years), and F=9.l4, p<.001. Both with a=.05 reveal a significant difference between academic Idmtifying pioneers Since the focal point of this study is on understanding problem choice motives of pioneering researchers and identifying exaaly who the pioneers are respondent sample should be a critical issue. is classified as a pioneer is institutional variations in the their persistence in the field, Our determination of who when based upon in the a respondent enters the field of neural networks. As previously discussed, the evolution of the neural network community is marked by a period of rapid by the advent of a professional growth in membership during the 1980s and society, specialized journals and books, and an international conference. These events serve as indicators of the establishment of neural networks as a legitimate field of research within the broader scientific community. Thus, demarcation for separating pioneering researchers from the somewhere a point of of the sample should rest lie in this period. The cumulative that the field distribution of the entry year for each of the grew most rapidly entering the field each year was ft-om much of the respondents entered the about 1984 onwards. 720 respondents The number of indicates respondents lower during the 1960s and 1970s. About 25 percent field prior to 1984; whereas about 75 percent of them entered from 1984 to 1990. Both the historical overview of neural network research and the entry pattern suggest that the point of demarcation for identifying pioneers pioneers, while placing the remaining control group (x2=18.15; p<.001). contrasting the responses of pioneers. seCTor is shown in Table 1. who we entered the field by 1983 as 324 (73 percent) who entered The The after 1983 into the control group provides us with a distribution of pioneers As we have previously reported, a the soundness of using 1983 as a cut-off year (see somewhere in this analysis, between 1980 and 1985. As a consequence, for 441 (of 452) cases used decided to classify 117 respondents (27 percent) is means for and controls within each sensitivity analysis confirms Rappa and Debackere, 1991). 11 12 13 14 15 16 number of difference in the lowest N cell cases per cell in the rwo-faaor about 17:1). Nonetheless, the is supported by those previously reported in ANOVA of the two-factor results Tables 2 and A 4. statistically significant interaction effcCTs further illustrates this items that are meant to capture the influence of (the ratio highest-to- communal ANOVA are closer inspection of the On argument. interaction two of the on the respondent's entry decision (the positive opinions of leading researchers and the successes of other researchers), commercial pioneers more closely resemble the controls than they resemble their fellow pioneers in other institutions (refer to the 3 Tabic 2 for details of the means cell in X 2 ANOVA). Given the time-dependent nature of this analysis, create a potential bias in the results. Specifically, were employed in a different type presently. Since we are making it some complications may in interpreting the results if in fact the respondents Migration among the respondents percent) academic respondents, 31 field may present a problem moved between different types of fairly is common. For example, 129 The difference across institutions we can determine whether p=.001). Although (47 (27 percent) commercial respondents and 17 (34 work on percent) government respondents have changed institutions since starting networks. then they are comparisons of entry decisions based upon present affiliations, the past institutional migration of respondents affiliations. may be the case that some respondents of institution when entering the institutional arise that is statistically significant neural (x^=13.50, d.f=2, or not a respondent has changed institutions since entering the field, unfortunately the survey does not allow us to identify the exact type of institution in which these respondents were previously employed. In order to rule-out the existence of bias in the subsequent analysis migration, who same we performed institution over the entire period. who to institutional the following tests comparing the answers of those respondents migrate between institutions after entering the respondents due are pioneers, it is field with those who remain with the Given the disproportionate number of migrant necessary to control for differences that may exist between them and the control group. Thus, within each type of institution, we made comparisons between migrant and non-migrant pioneers and migrant and non-migrant controls. Using discriminant analysis on academic (both pioneers and controls), commercial (controls only) and government (controls only) respondents, we were unable discriminate between migrants and non-migrants for each on the of the four discriminant functions were well number of pioneers in commercial and government basis to of the entry items (p-vaJues in excess institutions, of .10). Due we were unable to the to low perform 17 a similar discriminant analysis for migrants and non-migrants we compared migrant and non-migrant using t-tcsts on each entry item separately, no pioneers. In each case, Second, we significant differences could be found. repeated the analysis reported in the previous tables, only this time using who just those respondents network did not migrate between institutions since initiating their neural research. Again, the analysis largely confirms the results entire sample. Undoubtedly, knowing the they enter the field tests those scaors. Instead, in is most we have reponed institutional affiliation for each for the respondent when preferable. In absence of this information, however, the previous enable us to conclude that, regardless of whether a respondent has migrated between institutions, the results arc basically the same for both pioneers and the control group. Persistence in the field In our previous analysis, we show from their that pioneers differ significantly colleagues with regard to the faaors that might influence them to leave the field. In almost every respect, pioneers express a more passionate commitment to their chosen of research. results When viewed in conjunction with the analysis of problem choice behavior, the provide a coherent picture of pioneers cognitive perceptions of the field and community. Here, we variations that may revisit exist Table 6 shows the among results pronounced as in case less as researchers who more influenced by influenced by the social dynamics of the research pioneers. of a oneway ANOVA field, now of the factors that might influence the taking into consideration institutional observed across institutions for pioneers and controls are not as Many similarities both pioneers and controls. Tukey-HSD and Scheffd tests reveal of the respondent's decision to enter the exist across institutions for are the question of persistence in light of the institutional respondent's decision to leave the affiliations. TTie differences field significant differences for pairwise field. comparisons of means with respea to only two items: a lack of financial rewards and the discontinuance of neural networks research at the respondent's organization (Table 7). Financial rewards are more important to commercial respondents in the control group than they are to their control group peers in academic and government institutions. As might be expected, the discontinuance of neural networks research at the respondent's organization institutions than it is to those in is more important academic institutions for to those in commercial both pioneers and the control group. Within the control group, government respondents also consider discontinuance of neural networks by their organization more important than academic respondents. 18 19 PIONEER CONTROL GROUP ITEMS McOM lack of financial rewards ^ M^ACAD J^OM* ^^VT discontinuance at organization TABLE 7: ^^OM Pairwise comparisons on * M-ACAD mean l^OM * ^^ACAD mX)VT * M-ACAD scores for factors that might influence the respondent's decision field, The using Tukey-HSD and were further investigated using institutional differences and controls within each scaor (Table to leave the Scheffif tests. t-tests to These comparisons strongly suppon the main 8). finding firom our analysis of a respondent's decision to enter the difference between pioneers and the control group occurs institutions. With respea to respondents in compare pioneers commercial field; namely, the most strongly within academic institutions, except in one instance, there are no statistically significant differences between pioneers and the control group. The picture with government respondents examined interaction muced. Using a two-factor ANOVA, we between institutional type and the pioneer/control effects independent variables (Table is 9). The only "rapid progress in alternative areas." Upon item to attain a significant interaction was closer inspection, we found that pioneers in commercial institutions are more similar to the controls, regardless of institutional type, than to their pioneering colleagues expect, the main effeas again, caution relatively is in academic and government laboratories. As we should are consistent with the results reported in Tables 6 required in interpreting the results of the two-faaor high difference in the Lastly, the results number of cases of our inquiry into how and ANOVA 8. Once due to the p)cr cell. long respondents are willing to work on neural networks, given current progress, are reported in Table 10. Using Kruskal-Wallis non-parametric unable to find tests, we compared statistically significant differences across the three seaors. type using less pioneers and controls across institutions and were We also among the pioneers and the controls compared pioneers and controls within each Mann-Whitney non-parametric tests (Table 1 1). Once support the overall picture that has emerged thus far: institutional again, the results more or academic pioneers are 20 21 22 in the control from their peers significantly different group (p<.001). While government pioneers are also significantly different from their peers in the control group, the difference between commercial pionecn and the control group statistically is weak (p=.048). DISCUSSION A comparative analysis of institutions shows that significant variations factors influencing pioneers' decisions to enter from persist in a field. The results of problem choices indicate that academic pioneers are investigation of pioneers' different and exist in the their peers in that they are our distinctly most influenced by cognitive faaors and least influenced by social faaors in the research community. In contrast, pioneers in commercial and government laboratories more are similar to control their pioneering colleagues in universities. pioneering researchers we group respondents than they are to Thus, we conclude that the charaaeristics of previously found to exist are, in fact, largely associated with pioneers in universities. When we field, we compare pioneers across institutions in terms of the decision to enter the employed find that those in firms are more influenced than those in universities by the opinions of leading researchers, the successes of other researchers, and the community. Thus, beyond the expeaed firm, result of having a greater interest commercial pioneers are more greatly influenced by the internal the research community. Similarly, with respect to the factors that respondent's decision to leave the field, size in building their social dynamics of might influence a commercial pioneers are almost identical attitudes to their peers in the control group, with evidence of only one weakly in their significant difference being the diminished intellectual challenge of neural networks having relevance to pioneers. We Exploring these results express attitudes more their intentions for in like those like are more like is in the field. generally find that commercial pioneers of control respondents show some ability to in firms than they do other contrast with peers in the control group, commercial pioneers academic pioneers. However, the government pioneers remaining more depth, we pioneers. Furthermore, although they government pioneers more only one weak difference between commercial pioneers also find and controls when analyzing of the draw in their attitudes solid conclusions than they are with respect to considerably more difficult due to the relatively small respondents (15 pioneers, 34 controls) with which to make comparisons. A number of larger number of cases might change the overall picture regarding government pioneers in either direction. 23 24 In general, the findings from our analysis point to the special significance of academic institutions in fostering pioneering behavior. To the extent that researchers are provided the freedom to interact with their colleagues and pursue their interests determination, pioneers are perhaps more likely to be found in with single-minded an academic context. Our finding that commercial pioneers (and to lesser degree, government pioneers) are not unlike other researchers who on non-academic constraints placed persistence in the field. Be significant differences can be Having new field. in light this as students academic institutions when group from pioneering new fields interesting to ask fostering pioneers. more in — is it is pioneering a in wonh revisiting them it, interesting to see if pioneering students as is often implied in the literature it problems by faculty? that students play a Or more formidable (e.g. — and role in by leading the way for faculty? in pioneering new fields, whether or not certain institutions have a greater propensity Does the overall research prestige of a university matter? Although there more are they where they might have research agendas? note that very few our previous study, a disproportionate are directed to their research more abundant? Or institutions, of problems and of students this analysis, would be it likely to begin their research at relatively the special being a rather unique group, as Furthermore, given the importance of academic institutions is how entering the field of neural networks. Given the faculty. Is more challenging thought pioneering also interesting to is Although students were omitted from Ziman, 1987), that students a affect their choice to be to investigate the role importance of academic institutions, is it be an indication of found across institutions within the control group. would seem number of pioneers were this may, of the present findings. As we noted differ as a may researchers it identified pioneers within the next logical step may enter after them, is prestigious institutions, more likely to relatively greater latitude in much for Are pioneers where resources emerge from it less are prestigious pursuing non-mainstream anecdotal evidence in support of both views, there has yet to be any systematic examination of the relationship between institutional prestige and pioneering behavior. One caveat in drawing conclusions from our analysis relates to our inability to classify pioneers according to the institution where they the difficulties that can occur when examining with a cross-sectional survey. pioneers, in the who we same one first initiate longitudinal The problem does not so their research. phenomena, such much lie in This is one of as pioneering, evaluating academic are fairly ceruin stan their research in universities (albeit, not necessarily as where they are currently employed), but in evaluating commercial and 25 government pioneers. Nevertheless, the statistical pioneers enable us to be confident that, even with checks of migrant and non-migrant full information, the overall results will not change. Moreover, despite the analytical problems with respea to entry, a consistent picture emerges when studying a respondent's decision to enter together with their decision to persist in the field. CONCLUSION Elaborating on an earlier study of pioneers in neural networks, in which factors that influence their decisions to enter institutional variations in and persist in the field, this we examine the paper reports on pioneering behavior. Comparing pioneers in academic, commercial and government laboratories, we find that it is only academic researchers who demonstrate the kind of behavior we have previously found to be associated with pioneers. In their choice of a problem area and more influenced by their cognitive perceptions of their persistence in the field, what is academic pioneers are an interesting problem and less influenced by the actions and opinions of others in the research community. In contrast, commercial and colleagues who (to a lesser extent) enter the field after government pioneers are more similar to their has gained legitimacy than they are to fellow it pioneers in universities. Pointing-out a paradox of emerging fields. Carver Mead, a prominent neural networks, has remarked: "In a field like neural networks, one optimistic in the shon run, but one is who seem of persisting long enough in their research for the reality enthusiasm they have for it. laboratory they can fostering the call of a to have mastered the field to an catch-up with the the extent that universities can provide pioneers with a home, academic emergence of new usually too never optimistic enough over the long run" (DARPA, 1988:281). Pioneers are those relatively few individuals To is scientist in fields institutions will continue to play a special role in of science and technology. REFERENCES ALLEN, T.J. BARNES, MIT 1977. Managing the Flow of Technology. Cambridge, Mass.: S.B. 1971. "Making Out Press. in Industrial Research." Science StudiesVo\. 1, pp. 157-75. CHUBIN, D.E.,A.L PORTER, AND M.E. BOECKMAN. 1981. "Career Patterns of Scientists: Case for Complementary Data," Amrrican Sociological Review, Vol. 46. pp. 488-96. A COLE, JONATHAN R. AND STEPHEN COLE. 1973. Social Stratification in Science. Chicago: University of Chicago Press. Darpa. 1988. Neural Network Study. Fairfax, Virginia: APCEA International GIERYN, THOMAS 1978. F. "Problem Retention and Problem Change Sociological Inquiry, Vol. 48, pp. 96-1 GLUECK, W.F. AND 1 in Science." 5. jAUCH. 1975. "Sources of Research Ideas L.R. Press. Among Productive Scholars," Journal of Higher Education, Vol. 46, pp. 103-14. GR\HAM, M.B.W. AND B.H. PRUITT. 1990. Innovation at Alcoa. Cambridge, HagstrOm, Warren HOUNSHELL, DAVID Pont R&D, for Industry: A Century of Technical University Press. O. 1965. The Scientific Community. A. AND JOHN KENLY SMITH. New York: 1988. Science and Basic Books. Corporate Strategy: 1902-1980. Cambridge, UK: Cambridge University KORNHAUSER, W. 1962. MERTON, ROBERT New R&D UK: Cambridge K. Scientists in Industry. Berkeley: 1938. Science, Technobgy York: Harper & Row and University of California Press. Society in Seventeenth Century England. [1970]. MINSKY, MARVIN AND SEYMOUR PAPERT. 1988. Perceptrons: An Introduction Computational Geometry. Expanded Edition. Cambridge, Mass.: MIT Press. NATIONAL SCIENCE BOARD. 1989. Science and Engineering Washington, DC: Government Printing Office. NATIONAL SCIENCE FOUNDATION. 1989. National Report NSF 89-308. Washington, DC: Government AND R&D. New PELZ. D.C. F.M. ANDREWS. 1966. Indicators — 1989. NSB R&D Patterns of Printing Office. Resources. to 89-1 Final Scientists in Organizations: Productive Climates for York: Wiley. A. ANDKOENRAAD DEBACKERE. 1991. "Beyond Pcnistence: Understanding Commitment of Pioneers in Emerging Fields of Science and Technology," MIT RaPPA,MICHAEL the Du Press. Working Paper, December 1991. 27 ZIMAN.JOHN. 1987. "The Problem of 'Problem Choice,'" Minerva, Vol. 25, No 1-2. pp. 92-106. ZUCKERMAN. HARRIET. 1978. "Theory Choice and Problem Choice Sociobgical Inquiry, Vol. 48, pp. 65-95. in Science," . 28 APPENDIX Items inquiring into the respondent's decision to enter the field of neural networks: How important were each of the following factors in influencing your research agenda (respondents were asked important) -scale with midpoint 4 (somewhat important)): networks in to circle your on a initial decision to include neural I (not at all important) to 7 (very compelling nature of neural networks 1. intellectually 2. lack 3. availability 4. potential for financial rewards 5. potcndal for recognition by peers 6. dissatisfaction 7. positive opinions 8. successes of other researchers with neural networks 9. opportunity to build a commercial enterprise 10. opportunity to solve an important societal problem of other promising research topics of funding for neural networks research with a previous research agenda of leading researchers in the field Items inquiring into what might lead the respondent to leave the field of neural networks: How important would each of the following factors be network activities (respondents were asked to circle on a in diminishing your current interest in neural 1 (not at all important) to midpoint 4 (somewhat important)): 1. slow progress in solving technical problems in neural networks 2. lack of funding for your neural 3. diminished interest 4. rapid progress in alternative areas of research 5. opinions of leading researchers unfavorable to neural networks 6. negative opinion of your supervisor 7. discontinuance of neural net activities at your organization 8. lack of financial rewards 9. diminished intellectual challenge of neural network research 10. increased financial cost of conducting neural network research 1 overcrowding in terms of the number of neural network researchers 1 12. difficulty in among network research other researchers in neural networks keeping up with (if any) toward neural networks new developments 55U in neural 037 networks 7 (very important) -scale with IRARIF.S 3 TDflO DIlPl DD7Sbfll3 1 Date Due JAN. P^ 7P >l\ Lib-26-67 lilifini =1060 Q075b6l3 ^