Article Category: Article
Pubblicato online: 29 ago 2023
Pagine: 50 - 54
DOI: https://doi.org/10.2478/connections-2022-0003
Parole chiave
© 2023 Zachary P. Neal, published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.
The fields of social network analysis and network science have recently undertaken efforts to address issues of gender diversity and representation. For example, in 2018, the Women in Network Science (WiNS) group was formed to “connect and promote network scientists who identify as underrepresented genders.” Similarly, in 2023, the International Network for Social Network Analysis (INSNA) adopted a diversity, equity, and inclusion (DEI) policy to promote diversity by gender and other characteristics across the association's activities. Despite these important efforts, subtle threats to representation are likely to persist.
One such threat is that men and women may judge their network expertise differently. In this paper, I use a 2023 global survey of 260 network researchers to show that men rate themselves as having more expertise with networks than women. This is troubling because, to the extent that women believe they lack the expertise to be considered for awards or collaboration opportunities, it can impede other efforts to diversify the field. I discuss some possible explanations for this finding and some possible strategies for addressing the consequences of a gender gap in self-rated expertise. The data and code necessary to reproduce the analyses reported here are available at
This analysis draws on data collected through two separate web-based surveys designed for a larger study of guidelines for reporting about network data: a probability sample of network journal and association board members, and a convenience sample of network researchers. Both surveys were approved as exempt by Michigan State University's Institutional Review Board (STUDY00008880).
The board survey sought to include all members of the boards of two network associations (INSNA and NetSci) and seven network journals (
The convenience sample sought to solicit responses from anyone with “the expertise necessary to evaluate research about networks where nodes represent something, and that are not solely mathematical abstractions.” Invitations to participate in the survey were circulated via social media (Mastodon, Twitter) and listservs (SOCNET, REDES (in Spanish), cna2023, siam-ns) in May and June 2023. A total of 211 unique non-anonymous individuals provided responses; however, only 176 provided complete data on the variables used in this analysis.
The variable of interest—self-rated expertise in networks—was measured by asking: “On 10-point scale where 1 means ‘novice’ and 10 means ‘expert,’ please rate your level of expertise with network analysis or network science.”
Gender was measured by asking: “With which gender do you most closely identify.” Respondents were provided three response options: Man, Woman, and Non-binary/third gender. Seven respondents did not enter a response, but were personally known to the author; in these cases their gender was imputed. The use of imputed genders does not affect the results reported below. Four respondents reported identifying as non-binary. To preserve respondents’ anonymity, their responses to the gender question are suppressed in the public data and they are excluded from analysis. Grouping these respondents with women, as a minoritized gender, does not affect the results reported below.
Field was measured by asking: “What is the name of the discipline or field with which you most closely identify now?” Responses to this open-ended question were re-coded into three exhaustive and mutually exclusive categories: social, physical, and network. Social science disciplines (e.g., sociology, psychology, political science) were coded as “social,” and physical science disciplines (e.g., biology, neuroscience) were coded as “physical.” Responses that specifically named networks (e.g. network science), and disciplines that deal with network methods rather than substantive topics (e.g. mathematics, statistics), were coded as “network.”
Experience was measured by the logged number of years since the respondent earned their highest degree. A logarithmic transformation is used to account for the diminishing returns to expertise from additional experience.
Because the survey was completed by a global pool of respondents, data on race were not collected. Respondents were asked whether they “identify as a member of a minority racial or ethnic group”; however, many chose not to provide a response. Therefore, these analyses do not consider race, ethnicity, or racial/ethnic minority status.
Table 1 describes the board and web survey samples in terms of gender and field distribution, mean experience, and mean expertise. Although the web sample is a convenience sample, its gender and field distributions closely mirror those of the probability sample of board members, suggesting that it may nonetheless be representative of the population of network researchers. The two samples differ in expected ways by experience and expertise: members of the board sample have more experience (21.3 years vs. 11.7) and higher self-rated expertise (8.8 vs. 7.8). In the analyses reported below, the two samples are pooled to achieve a larger sample size and greater statistical power. The results remain the same if analyses are performed using separate non-pooled samples.
Survey descriptive statistics.
| N | 84 | 176 |
| Gender | ||
| Man | 62 | 131 |
| Woman | 22 | 45 |
| Field | ||
| Network | 41 | 85 |
| Social | 38 | 81 |
| Physical | 5 | 10 |
| Experience | 21.3 (11.9) | 11.7 (11) |
| Expertise | 8.8 (1.2) | 7.8 (1.6) |
Figure 1 illustrates the distribution of network expertise self-rated by men (left) and women (right). Men's self-rated expertise is skewed toward higher values (
Figure 1:
Distribution of self-rated network expertise by gender.
This apparent difference could be due to sampling error, if expert men were more likely to complete the survey than expert women. However, this difference persists in the board sample, which is composed solely of respondents who have a sufficiently high level of expertise to have been appointed or elected to journal or association board roles. Within the high-expertise board sample, men report having statistically significantly more expertise (
The apparent difference could also be due to confounding variables. For example, a gender gap in expertise could appear if women entered the field of network analysis more recently and therefore have less experience on average than men. It could also appear if women are more likely to pursue network research in the social or physical science fields where their substantive expertise is primary, while men are more likely to pursue network research in the field of network science where their network expertise is primary. However, this difference persists when controlling for such potentially confounding variables.
Table 2 reports the results of an OLS regression predicting self-rated network expertise as a function of logged experience, board membership, identifying with a network-focused field (as opposed to a substantive discipline), and identifying as a man. As expected, experience, identification with a network-focused field, and board membership are all positively associated with higher self-rated network expertise. But, after controlling for these factors, men still rate their network expertise as 0.75 points higher than women. Indeed, as the standardized estimates indicate, gender has a slightly stronger association with self-rated network expertise (b = 0.21) than does experience (b = 0.17).
Regression predicting self-reported network expertise.
| Intercept | 6.32 | 0.28 | < 0.01 | – |
| log(Experience) | 0.65 | 0.23 | 0.01 | 0.17 |
| Board | 0.8 | 0.2 | < 0.01 | 0.24 |
| Network | 0.54 | 0.18 | < 0.01 | 0.17 |
| Man | 0.75 | 0.2 | < 0.01 | 0.21 |
| R2 | 0.21 |
I find in a global sample of 260 network researchers that men rate themselves as having more expertise in networks than women. Even after controlling for years of experience, membership on a network journal or association board, and identification with a network-focused field, on average men rate their expertise as 0.75 points higher than women on a 10-point scale.
There are several possible explanations for this gender gap in self-rated network expertise. First, there could be true differences in expertise. Decades of research have consistently shown there are no innate differences in men and women's cognitive abilities (Hyde et al., 1990; Spelke, 2005). However, true differences in network expertise could still emerge if, for example, women experience more barriers to building expertise (e.g. bias, harassment) during their training (Hussénius, 2020), or if women are not invited as co-authors in fields dominated by men (Kwiek & Roszka, 2021; Wang et al., 2023). Second, men may overestimate their expertise. Overestimation of one's actual abilities is often described as the
Although these data cannot identify the causes of the observed gender gap in self-rated network expertise, the gap may have significant practical implications for individual network researchers and for the field. Because even the most rigorous and compelling research does not “speak for itself,” advancing one's research and career often requires significant self-promotion. The visibility of one's work, and subsequently its likelihood of being cited, depends in part on sharing work with colleagues (e.g., I thought you might like this paper) and disseminating it broadly via listservs and social media (e.g., check out my new article). Similarly, receiving support and recognition for one's research depends in part on a favorable self-presentation in grant proposals (e.g., I am the best person to do this work), award nominations (e.g., I am the most deserving of this award), and media interviews (e.g., I am confident in this finding). If women are even a little less likely to engage in these forms of self-promotion than men, there is the risk that their work will be less visible, supported, or recognized, even if they are numerically well-represented in the field.
Identifying solutions to the issue is challenging. Frequently proposed solutions including promoting diversity, reducing bias, and equalizing numerical representation may help, but it is unclear whether these specifically address issues associated with inequality in
The data and code necessary to reproduce the analyses reported in this manuscript are available at