Gender variation in the antecedents of task advice network size: Organizational tenure and core self-evaluations

Abstract

Research finds gender differences in the size, quality, and consequences of social networks in the workplace. Building on these studies, we focus on one type of social network: task advice networks, which we define as the networks that act as conduits for information and knowledge directly related to work task completion. Using data on over 1300 employees, we test the relationships between task advice network size and two variables – organizational tenure and core self-evaluations, examining differences by gender. We find a larger positive association between core self-evaluations and task advice network size for men than for women. Additionally, we find that men, but not women, have larger networks when lower in tenure.

Keywords:

• Social networks
• Gender
• Core self-evaluations
• Organizational tenure

Notes

1 Tel.: [phone]; fax: [phone].

2 Other structural variables are also included as controls. In particular, occupation is a key structural characteristic, but the occupational breakdown available in our survey data is so closely tied to the participating organizations that it poses problems for interpretation. To avoid this, we did not select occupation as a key predictor.

3 Despite the advent of anti-discrimination laws, decreasing birth rates, and increasing educational attainment, women remain disproportionately likely to work in service and clerical occupations. Some authors such as Gabriel and Schmitz (2007) attribute these differences largely to personal choice and labor market structural characteristics.

4 We consider other cutoffs, such as 50% and 60%, but as the samples are not substantially different for differences in sample sizes of less than 10 respondents, we select the less restrictive 25% cutoff.

5 The existing literature operationalized task advice networks in two primary ways. A first method entails asking people to rate on a Likert-type scale how often they ask each person in a preexisting coworker pool for advice or for their viewpoint, and then dichotomizing the scale so that lower numbers (represent a coworker not in the task advice network, Goodwin et al., 2009; Wong, 2008b). A second method asks respondents to identify work group members to whom they usually go to for advice with their work (Wong, 2008a), for their advice or viewpoint (Lazega, Mounier, Snijders, & Tubaro, in press), or for sources of job-related advice (Zagenczyk & Murrell, 2009). The analysis reported in this paper uses a measure asking whether network members are identified as having a high or moderate impact. We replicate our analysis using data on the frequency with which respondents received advice from these network members within the past month for “fairly specific or detailed questions at work,” “for general guidance or referrals to other sources of information,” or “to help us think through a problem even when they may not have the specific information that we need,” an approach in line with the first method described above. Only 3.7% of network members are not contacted for any advice type during this time period, indicating a close match between the pool of high-impact and moderate-impact network members and network members asked for advice. We opt to use the slightly broader definition, focused on impact rather than having provided advice within the past month, as previous networking studies often use longer time periods than the time period in our survey. We also replicate our analyses using only high-impact or only moderate-impact individuals, and the coefficients are similar in size and direction.

6 We replicate all analyses in this paper with two modifications. First, we replicate the analysis using the 583 respondents who report task advice networks of zero size. While it is possible to have zero size task advice networks, a zero might also indicate missing value on the dependent variable. The results do not differ substantially, although the standard errors are larger. Second, we replicate the analysis using Hilbe's (2011) survival parameterization of censored Poisson regression in which respondents who report exactly 8 network members are censored at 8, while those who reported exactly 14 network members are censored at 14. The results do not differ substantially and, as there is not a noticeable drop between 8 and 9 in the reverse cumulative distribution (Fig. 1), we opt to retain censoring only at 14.

7 Auxiliary analyses using Blinder–Oaxaca decomposition of non-censored Poisson regression models (Sinning, Hahn, & Bauer, 2008) indicate that only between 21% and 36% of the gender difference in task advice network size is attributable to differences in the characteristics of men and women, while between 64% and 79% is attributable to differences in the coefficients These results are not presented in the results section, because they do not take censoring into account. The coefficients from the uncensored models are similar to those from the censored models, although levels of significance differ. Decomposition methods are not as well-established for censored models as for standard or even truncated models.