Who You Know vs. What You Know: The Impact of Social Position and Knowledge on Team Performance

ABSTRACT

Organizational behavior theories generally agree that human capital is critical to teams and organizations, but little guidance exists on the extent to which such theories accurately explain the relative contributions of individual actors to overall performance. Using newly created network measures and simulations based on data obtained from a software development firm, we investigate the relative effectiveness of social network theory and resource dependency theory as predictors of individuals' contributions to team performance. Our results indicate that individual impacts on team performance are more closely associated with knowledge and task dimensions than with social network structure. Furthermore, given that knowledge may be assessed a priori, these factors provide useful guidance for structuring teams and predicting team performance.

Keywords:

• group performance
• social networks
• knowledge networks
• resource dependency
• human resource management

Authors express appreciation for the insightful observations of Patrick Doreian and the anonymous JMS reviewers and acknowledge the helpful comments offered by Dan Brass, Rich Burton, David Krackhardt, Bill McKelvey, Michael Prietula, and Marshall Van Alstyne on earlier versions of this paper. Funding for this work was provided by the National Science Foundation IGERT grant no. 9972762, the Carnegie Mellon University (CMU) School of Computer Science, the CMU Center for Computational Analysis of Social and Organizational Systems, the William Larimer Mellon Foundation, and the CMU Tepper School of Business.

Notes

We use the analogy of Type I and Type II errors strictly for convenience in describing the efficacy of measures used to evaluate the hypotheses. We do not imply that existing theories are invalid because they exhibit false negatives. Instead, we are suggesting that while theories of organization science may reflect Popper's falsifiability criterion, those theories which can be shown to be consistently less falsifiable (viz., exhibiting consistently fewer false negatives or false positives) are arguably more robust.

In the multiple QAP regression (Model 4), the coefficients of the proximity and authority/community matrices were significant at p=0.05 and p=0.001, respectively. The task assignment matrix, however, did not have a statistically significant coefficient (p=0.20) in the multiple regression. Since task assignment was significantly correlated with the social network matrix when considered standalone (Table 1, Model 3), the reduced significance in the combined regression indicates non-linearity in the multiple regression model.

The hierarchical clustering technique used throughout the paper is based on average linkage updating of distance between clusters (Sokal and Michener, 1958). The distance between the coordinates of each actor (as determined by actors' x and y values of the metric being clustered, such as x = degree centrality with y = performance index) is calculated as Euclidean distance. Then, the distance D _ck between clusters c and k is computed as

where T _ck equals the sum of all pairwise distances between actors in cluster c and cluster k, and n _c and n _k are the sizes of clusters c and k respectively. At each stage of clustering algorithm, the clusters for which D _ck is the minimum are merged.

df = 494

*p < .001