Connections at Work: How Friendship Networks Relate to Job Satisfaction

Abstract

Previous research in the association between network centrality and job satisfaction has not established a consistent relationship between the two. Considering a specific type of network and multiple measures of centrality may clarify this relationship. Thus, the current study examined the association between various types of centrality in workplace friendship networks and job satisfaction in a Korean construction company. Friendship network centrality measured as closeness was positively related to job satisfaction. However, friendship centrality measured as betweenness and degree was not related to job satisfaction. The results suggest that distinguishing among measures of centrality and network type is vital for future research.

Keywords:

• Friendships
• Job Satisfaction
• Network Centrality

The authors wish to thank Editor Wendy Samter and two anonymous reviewers for their helpful comments on this article. An earlier version of this paper was presented at the annual meeting of National Communication Association, San Antonio, TX in November 2006.

Notes

∗p < .05

†p < .01

‡From respondent-only matrix (n = 101)

^§from matrix with non-respondent links (n = 166); df = 99

∗p < .05

†p < .01

‡From respondent-only matrix (n = 101)

Abbreviations: β = standardized regression coefficients, sr = semipartial correlation

Due to the correlations between the independent variables shown in Table 1, some readers may wonder whether the regression results are affected by multicollinearity. Several tests demonstrate that multicollinearity did not affect the results. First, among the predictors included in the regression analysis, the highest correlation was .56 (between in-closeness and betweenness from the respondent-only matrix), which is smaller than the rule of thumb value (r = .70) for raising the possibility of multicollinearity problems. Further, there were two betweenness scores in the correlation matrix and only one was included in the regression because they were essentially the same measure (though a different number of participants was used for calculation) and because these two were too highly correlated (r = .93). Second, none of the predictors had a variance inflation factor (VIF) higher than 2 (the highest one was 1.91, and the lowest one was 1.50). Cohen, Cohen, West, and Aiken (2003) listed 10 as the traditional rule of thumb threshold value and 6 as a more stringent threshold value for VIF statistics, although there is “no good statistical rationale for the choice of any of the traditional rule of thumb threshold values” (p. 424). Third, one of the common remedies is to respecify the model. When the regression analysis was re-run with any combination of the five predictors, the overall pattern of findings did not change: the results still showed that in-closeness and out-closeness were significant and the other three predictors (in-degree, out-degree, and betweenness) were not significant. Finally, the residuals were checked through numerous measures and no abnormality was found.

Additional information

Notes on contributors

Amber N. W. Raile

Amber N. W. Raile (MA, Michigan State University, 2005) is a doctoral student at Michigan State University.

Rachel K. Kim

Rachel K. Kim (MA, University of Hawaii at Manoa, 2004) is Assistant Professor at Michigan State University.

Jounghwa Choi

Jounghwa Choi (PhD, Michigan State University, 2007) is chief consultant at Porter Novelli in Seoul, South Korea.

Kim B. Serota

Kim B. Serota (MA, Michigan State University, 1974)

Hee Sun Park

Hee Sun Park (PhD, University of California, Santa Barbara, 2003) are doctoral students at Michigan State University.

Dong Wook Lee

Dong Wook Lee is a doctoral student in the Department of Civil and Environmental Engineering at Hanyang University in Ansan, South Korea.