Abstract
Transactive memory theory suggests that general awareness of expertise location in a group is sufficient to predict expertise seeking. Yet expertise seeking is, at least in part, a social phenomenon between two individuals embedded in a network of social relationships. Taking a multilevel, network perspective, we examined the interplay of affective relationships and awareness on expertise seeking in groups. Hypotheses were tested using network data collected from 693 employees in 53 sales groups. HLM analysis results indicated that awareness of expertise distribution positively influenced the decision to seek expertise at all levels of analysis examined. In addition, both positive and negative affective relationships influenced expertise seeking, although their pattern of influence differed across different levels of analysis. More specifically, having either a positive or a negative affective relationship with another group member affected the decision to seek expertise from that person. Although having many positive relationships had a positive effect on expertise seeking, having many negative affective relationships had no effect. Moreover, having both an awareness and a positive affective relationship with another group member amplified their positive effect on expertise seeking. Last, individuals who had more negative affective ties were less likely to leverage the positive impact of each awareness relationship on expertise seeking.
Acknowledgments
The article is supported by NSF-IOS (Grant No.: 0822784).
Notes
1 To test for possible biases introduced by control variables (Spector & Brannick, 2011), control variables were included before and after entering the hypothesized main and interaction effects. Because results were consistent regardless of the order in which control variables were entered, only models in which controls were entered before research variables are presented.
2 To facilitate understanding, superscripts were added to the end of each variable's name to indicate the level of analysis of each predictors, with 1 for level-1 (dyadic level), 2 for level-2 (individual level), and 3 for level-3 (group level) of analysis.
3 Building Model 6, we have also tried a model in which all the control variables were dropped given the recent debate in the research community about how the choice of control variables may sway the results. Because, in our analysis, excluding the control variables did not change the direction or the significance of the hypotheses tested, we reported in this study the model that contains the control variables.
4 We used results from Model 5 to report hypothesis testing because this model contained results for both supported and nonsupported hypotheses, whereas Model 6 focused on significant results only.
5 Our use of symbols for coefficients is consistent with conventional practices in multilevel modelling (see, for example, Raudenbush & Bryk, Citation2002). More specifically, π is the coefficient for effects at the lowest level in the analysis (level-1), β is the coefficient for level-2 effects, and γ is the coefficient for level-3 effects.