Balancing the costs and rewards of multidisciplinarity in Science: the role of collaboration network structure
Karine Revet, Mustapha Belkhouja, Barthélémy CholletThe potential of multidisciplinary research has long been established (Hessels & van Lente, 2008; Raasch et al., 2013). Combining distant bodies of knowledge is necessary to address complex problems (Klein & Falk-Krzesinski, 2017). Important breakthrough can happen through importing in a field approaches or techniques from other fields (Koppman & Leahey, 2019). Thus, much effort has been made to support multidisciplinary collaborations (Brown et al., 2015).
Research shows that from an individual standpoint, however, adopting a multidisciplinary focus can have mixed effects, by increasing visibility but penalizing productivity, impact and career (Leahey, 2018; Leahey et al., 2017). First, the institutional environment and promotion systems usually follow a logic of silos with a prevalence of disciplinary norms and logics (Frickel et al., 2016). Second, multidisciplinary work is penalized by biases inherent to peer-review processes (Bromham et al., 2016; Whalen, 2018).
We build on network research to argue that these risks attached to a multidisciplinary research strategy can be offset by having certain types of collaboration network. For instance Lungeanu et al. (2014) found that winning an interdisciplinary grant was more likely among teams of scientists who have a common history of collaboration. In a similar vein, we argue that certain network structures can mitigate the legitimation issues attached to multidisciplinary strategy and reduce the coordination costs inherent to this type of research.
To study how network structure moderates the effects of multidisciplinary research on individual productivity, we study a sample of authors active in either one of three disciplines (business, economics and psychology). From ISI Web of Science, we collected information on 471.045 articles from 825 scientific peer-reviewed journals, published over a 20-year period. This information includes article title, journal and year of publication, name, rank and affiliation of authors, and total number of citations received. We link this data to the Scimago Journal Ranking (based on Scopus data) to access journal metrics and classify them by subject area(s).
To test our hypothesis, we rearranged the data at the individual level by creating yearly records for all authors identified in our dataset, and excluded the least productive ones (less than ten publications over the 20-year period). We study the evolution of every author’s structural position (centralities, brokerage indices) in the co-authorship network graph, as well as diversification in terms of subject areas (based on the classifications appearing in individual publication records). Our final sample is an unbalanced panel, with 147.353 observations for 12.919 authors over the period 1999 to 2013. Since the number of publication is a count variable, we used a Negative Binomial model, which better fits the distribution of our dependent variable and allows for robust over-dispersion than the Poisson model. More specifically, we employed both fixed and random effects model to capture the within individuals’ variations and found qualitatively similar results.
Taken together, our analyses contribute to the network literature by showing that, depending on the multidisciplinary profile of the scientists, different network structures can reduce the costs related to multidisciplinarity and ensure performance in knowledge production activities.