Portal de Periódicos da CAPES

The diffusion of IT in higher education: publishing productivity of academic life scientists

2010; Taylor & Francis; Volume: 19; Issue: 5 Linguagem: Inglês

10.1080/10438590903434844

1476-8364

Anne E. Winkler, Sharon G. Levin, Paula E. Stephan,

Digital Platforms and Economics

Abstract This study investigates how the diffusion of Internet access and other advancements in IT across a broad group of institutions of higher education has affected the publishing productivity of life scientists. Several IT indicators are considered: (1) the adoption of BITNET; (2) the registration of domain names (DNS); (3) the availability of the electronic journal database, JSTOR (http://www.jstor.org/), and (4) the availability of electronic library resources. Data on life scientists are from the 1983, 1995, 2001 and 2003 Survey of Doctorate Recipients (SDR). Educational institutions are classified into tiers depending upon research intensity. Three hypotheses are tested: (1) IT enhances the careers of faculty; (2) IT improves the careers of faculty at lower-tiered relative to higher-tiered institutions; and (3) IT increases women's publication rates relative to those of men. The results provide some support for the first two hypotheses but no support for the third hypothesis. Keywords: diffusiontechnologylife sciencesprofessional labour marketsgender JEL Classification : O33J44J16 Acknowledgements This research is funded by a grant from the Andrew W. Mellon Foundation titled 'The Diffusion of Information Technology across Institutions of Higher Education: Effects on Productivity by Type of Institution and Gender'. The use of National Science Foundation (NSF) data does not imply NSF endorsement of the research methods or conclusions contained in this paper. The authors are grateful to Kelly Wilkin, Josh Leesman, and P. Mitchell Downey for research assistance and to JSTOR for the provision of data. We have benefited from the comments of an anonymous referee, Bronwyn Hall, Shiferaw Gurmu, Mary Beth Walker, and Michael Allison. Notes Although the IT revolution can be dated to the creation of ARPANET by the Department of Defense in 1969, restricted access to ARPANET led others to develop their own networks (National Science Foundation Citation2009). While BITNET was not the only one among these, nor the first, it became a leader during this period as discussed in the text. For a review of recent studies and discussion, see Blau, Ferber, and Winkler Citation(2010). All authors have a restricted license to use the SDR data. This network pre-dates what we now call the Internet since it did not use the TCP/IP standard. Part of the explanation for the limited use of an e-mail address prior to 1997 may be that the NBER only explicitly required an address after January 1997. In preliminary work, we also studied the social sciences. The universe of institutions was initially formed by a careful review of years of institutional data available in IPEDS (Integrated Postsecondary Education Data System). For further details, see Levin, Stephan, and Winkler Citation(2010). There were also a number of additional institutions in the SDR sample that were not in the IPEDS data set because of how the institutional database was constructed. Data on the IT measures for these institutions were also collected so that they could be included in the subsequent individual-level analysis of publishing activity. After this date, data were no longer systematically collected because of the availability of competing and superior technology, i.e. the modern-day Internet. In cases where the university had more than one server registered, we examined the dates of all named servers and recorded the earliest date. Because branch campuses may have relied on a system-wide server before obtaining their own domain names, we collected both the earliest date of the domain name registered for the system, along with the earliest date that the branch campus registered its own domain name, and used the former in the study. JSTOR became officially available in January 1997, but some adoptions occurred in late 1996. In surveys after 2000, questions about e-mail reference services and library internet services are combined. See http://nces.ed.gov/surveys/libraries/ While the SDR was conducted for the sciences in other years, information on publication counts, a key measure here, was not collected. The use of e-mail took off around 1994–1995, fueled, in part, by the explosion in lower-cost, more powerful, personal computers and growing access to the Web with 'free' e-mail accounts through providers such as Yahoo. The sample excludes individuals at 2-year institutions; schools of theology; other separate health-related schools (except the Mayo Graduate School); engineering and technology; business; art, music, and design; law; other specialized institutions; tribal schools; and those with missing information on type of school. These schools correspond to the 1994 Carnegie designations of 40, 51, 53, 54, 55, 56, 57, 58, 59, 60, and M, respectively. An earlier version of this paper used the average annual publication flow as the dependent variable. By way of example, if an individual worked at the current institution for 3 of the 5 years, the adjusted count is three-fifths of the total 5-year count; the period of exposure is 3. This adjustment method over-attributes publication counts to the current institution if the current institution's research environment is weaker than the previous institution's; it under-attributes publication counts if the current institution is stronger. Sensitivity testing, discussed later, suggests that the results change little if movers are excluded from the sample. Our concerns are further reduced because the lag between paper submission and acceptance is approximately a year in the life sciences (Ellison Citation2002). Work by Stephan and Levin Citation(2001) emphasizes the importance of career stage and not age per se in the life cycle of the academic career. In 1993, doctoral programs were ranked by the NRC as Distinguished, Strong, Good, Adequate, Marginal, or Not Sufficient for Doctoral Education and were assigned scores of 1 to 5, with 'Distingushed' having a score of 1. Institutions with programs ranked lower than 3 (Distinguished or Strong) were placed in Tier 1. In cases in which institutions have multiple doctoral-granting programs in a broad disciplinary area, scores were averaged across programs as was done by Adams et al. Citation(2005). Other institutions that had programs ranked lower by the NRC (scores 3 or higher) were placed in Tier 2. In 1982, fewer programs/disciplines were ranked by NRC than in 1993, but otherwise the same approach was followed. In some instances, an NRC score was available for 1982 or 1993, but not both. In such cases, the score for the available year was used for the missing year. Information was obtained from NIH, Office of Extramural Research (1998). Medical schools are assigned to Tier 1 if they are among the 50 institutions receiving the largest amount of extramural research awards from NIH. The awards ranking for 2001 is matched with the 2003 SDR; 1998 with the 2001 SDR; 1993 with the 1995 SDR; and 1981 with the 1983 SDR. As would be expected, these schools overlap considerably with those institutions identified by the NRC as having Distinguished or Strong programs. For this period only, the criterion that IT must be available for at least 1 year prior to the publication count is relaxed, since BITNET did not begin until 1981. See discussions in Cameron and Trivedi (2008, 560–1) and Wooldridge (Citation2002, 645–56). Specifically, the estimated results presented here were performed using STATA. We employed a quasi-maximum likelihood approach which maximizes the Poisson MLE and uses robust standard errors clustered around institutions. We also invoked the offset option to capture differing periods of exposure for individual researchers. In earlier work, we estimated models of average annual publication flows (publication counts divided by exposure) using ordinary least squares. Notably, signs and statistical significance of the IT variables are quite similar in both specifications, though Poisson is the preferred method given the count nature of the data. An equivalent method of obtaining this 'differential' impact is to estimate a regression of research productivity over two tiers, where tier is interacted with each covariate including IT. In this latter specification, the coefficient on IT*Tier directly provides the 'differential' effect. The method used here was chosen for expository purposes. The SDR sampling frame is drawn from individuals who received their PhD in the USA. Thus, it does not capture immigrant scientists trained outside the USA. Comparisons regarding Tier 4 are also limited because of its very small size, representing just 4–5% of the sample. Coefficient estimates of covariates for some selected models are shown in Table 10. They are not displayed in Tables 5–9 for purposes of brevity. With the exception of Model 1 of Table 5, Tier 4 is excluded from the publishing productivity analysis due to small sample size. In the Poisson regression, coefficients do not directly provide information about the magnitude of a variable's effect. The formula [(exp(B)−1) * 100], where B refers to the Poisson coefficient, provides information about magnitude in percent terms. Thus, the coefficient of 0.46 in Table 6 implies that the publication counts of individuals at institutions with DNS were 58% [(exp(.46)−1)*100] higher than for individuals without this type of IT, controlling for demographic factors. The definition of early adoption builds on Rogers Citation(2003) who classifies 'innovators' as being in the first 2.5% of adopters and 'early adopters' making up the next 13.5%. Here we classify any institution that fell into the first 16% as being an early adopter. The cut-off date for this definition of early adoption is June 1988. We thank the editor for making this suggestion. In 1983, some institutions had access to BITNET as noted earlier, while others did not. In 1995, some institutions had adopted BITNET by 1990 or moved on to the Internet and had registered their domain names, while others had not. Using this variation, we estimated the following model by tier using ordinary least squares: an institution's average publication flow = B0 + B1 IT + B2 1995 + B3 IT*1995 + ϵ. In a separate paper (Ding et al. Citationforthcoming), we are exploring many of these issues by appending our measures of IT 'connectivity' to longitudinal data (1969–1993) on the research productivity of life scientists. Advantages of these data are that they pre-date the IT revolution and span the period of the 1980s when connectivity was evolving. One would, however, expect IT to have become increasingly important in the biomedical sciences as large databases, such as GenBank and the Worldwide Protein Data Bank, have become available through web access. As mentioned earlier, due to the lack of publications data for the critical latter part of the 1980s, SDR data would not be well suited for this purpose.