Team diversity, polarization, and productivity in online peer production

Abstract

We define network-based indicators to characterize diversity of Wikipedia teams and contributing users. A team of Wikipedia users is diverse to the extent that its members edit different articles. An individual user has diverse interests to the extent that she contributes to articles that are not normally co-edited by the same users, i. e., if she contributes to an atypical combination of articles. For both indicators, we propose a model-based normalization by comparing observed and expected values computed on a reference random graph model that preserves expected degrees of users and articles. Using data on all articles of the English-language edition of Wikipedia, we show that diverse teams tend to produce high-quality (or “featured”) articles. In contrast, teams of users that individually have diverse interests tend to produce articles of lower quality. These findings are robust with respect to several alternative explanations for article quality. We also show that the proposed model-based normalization of network indicators outperforms an ad hoc normalization via more conventional cosine similarity measures. Finally, we analyze the interplay between team diversity and polarization sustained by adherence to behavioral norms predicted by balance theory. Results suggest that diversity can mitigate the—otherwise negative—effect of polarization on team productivity.