Abstract
Collaborative research has been increasingly celebrated by the science community, but the hypothesized positive relationship between research collaboration and research output is more assumed than rigorously tested. In this paper, we identify three methodological gaps in the literature: (a) hierarchical coding based on the ISI Web of Science database causes severe loss of information on local collaboration, (b) the relationship between research collaboration and research output is likely to be confounded by a common latent variable such as a scientist's ability, and (c) the lack of longitudinal analysis prevents causal inferences from being made. To address these methodological concerns, we constructed a longitudinal dataset of 65 biomedical scientists at a New Zealand university and coded collaboration variables by hand checking each of their publications in a period of 14 years. We found that at article level, both within-university collaboration and international collaboration are positively related to an article's quality and that, at scientist-year level, only international collaboration is positively related to a scientist's future research output. (C) 2008 Elsevier B.V. All rights reserved.