Seite - (000234) - in Knowledge and Networks

229 Analyses The dependent variables in our analyses—the number of leadership nominations received by an individual, and the number of times an individual was cited as some- one others turned to for work related advice—are count variables. For these kinds of data, Poisson-based regression models are more appropriate than Ordinary Least Squares (OLS) regression. However, our data showed clear evidence of over- dispersion (e.g., after fitting the ordinary Poisson regression model, the Pearson chi-square goodness-of-fit statistic divided by degrees of freedom was much larger than 1). Therefore, we used the negative binomial variant of Poisson regression that explicitly includes a parameter for over-dispersion (see Hilbe, 2008). In negative binomial regression, the log of the expected values (μ) is a linear function of the independent variables plus the dispersion parameter: log( ) * * .... * .m e= + + + + +intercept b X b X b Xm1 1 2 2 3 We employed the Likelihood Ratio (LR) test to assess the comparative goodness of fit between models (Huelsenbeck & Rannala, 1997). The LR test compares the likelihood scores of two models. The LR statistic, which follows a chi-square distri- bution, assesses whether the addition of an additional parameter (e.g., self- monitoring) leads to a significantly better fitting model than a baseline model (e.g., a model containing just the control variables). To test the interaction Hypothesis 3, we mean-centered measures of self-monitoring and trust brokerage and multiplied them to create a single interaction term. We then included this interaction term in the regression equation containing control variables, self-monitoring, and trust brokerage. Results The descriptive statistics in Table 11.1 show that the typical non-supervisory employee had worked for the firm for four-and-a-half years, was seen as a leader by nine other people, was turned to for advice by 16 other people, and was regarded by the supervisor as a high performer (M = 10.15 on a 15-point scale). Compatible with the first two hypotheses, individuals high in self-monitoring, compared with those low in self-monitoring, tended to receive more leadership nominations (r = .23, p < .05) and more nominations as advice providers (r = .25, p < .05). Further, self- monitoring was related to the tendency to occupy a role as a broker trusted by those who did not trust each other (r = 20, p < .10). Tables 11.2 and 11.3 show the results of tests of hypotheses. All of the binomial regression models in these tables demon- strate goodness-of-fit ratios close to one (the chi-square statistic divided by the degrees of freedom) indicating well-fitting models. 11 Brokering Trust to Enhance Leadership