The Impact of Research Collaboration on Scientific Productivity
Scientific collaboration often is viewed as a virtue, so much so that several public policies actively encourage scientific collaboration at both the individual and institutional levels. But few studies have actually examined the impacts of collaboration and fewer still have related collaboration patterns to publishing productivity. Based on data from 443 academic scientists, our research examines the effects of collaboration on scientists’ productivity, measured in terms of publications. We examine publications productivity by two measures, numbers of scientific articles and books published and “fractional count,” the number adjusted by number of co-authors. We first examine descriptive data for publishing productivity and find that mean number of publications (by both normal and fractional counts) grows substantially during the past three decade among all groups to 3.6 (normal) and 1.28 (fractional) in 1996, actually declining somewhat between 1996 and 1999. Fractional and normal count productivity is related to being male, tenured, married and, interestingly, being nonnative. Those in the chemistry discipline have an especially high productivity rate and those in computer science publish at a much lower rate than other fields of science and engineering. The descriptive data for collaboration show that the mean number of (self-defined) collaborators for one year’s research work is 13.8. On average, the collaborators are 40.9% graduate students and 27.1% female (females are about 14% of the sample). On average, scientists spend about 16% of their time working alone and the remainder collaborating and more than half of the collaboration time is with those in one’s own research group. We developed a “collaboration cosmopolitanism” scale to measure collaboration outside one’s own work group (e.g. persons in other universities, other nations) and found that physicists tend to be the most cosmopolitan in collaboration patterns. More cosmopolitan collaboration patterns are associated with being male, being a tenured faculty member, and total number of publications since 1996 (normal count, not fractional count). In examining the relation between publishing productivity and collaboration we focused on publications since 1996 because our survey data are cross-sectional and our publications data, which is based on curriculum vitae records, is longitudinal. Our approach was to test a number of regression models, alternative explanations of productivity, to determine if the explanatory power of collaboration was diminished by such factors as job satisfaction, rank and age, gender, discrimination, nationality and collaboration strategy. In each of the models, number of collaborators remains the strongest predictor of productivity, measured by both fractional and normal count.