Which Birds Fill the Branches of the AERA Twitter Tree? Twitter Networks Around #AERA19

• In Event: Affordances and Modeling of Intensive Data

Abstract

Twitter is increasingly used in professional communication, including among the education community—there were 25,720 tweets with the #AERA19 hashtag surrounding the AERA 2019 annual meeting. Ongoing communication as through Twitter may be an important part of the Educational Research community of practice (Lave & Wenger, 1991). These communities exist within the larger society and are susceptible to its influences, including power structures inherent in the broader context (Wenger, 2010).

Within academia, recent work has indicated that female scholars may be at a disadvantage in publication and funding (e.g., Kolev et al., 2019; West et al., 2013), including in Educational Psychology (Greenbaum et al., 2018).

Social networks and interactions may provide insight into how gender functions within the Education Research community. The social networking platform Twitter provides ready data to explore such interactions. There is evidence from other fields that men and women have different interaction patterns on Twitter: within political journalism, men amplify voices of other men almost exclusively, reinforcing the gender imbalance in a male-majority field (Usher, Holcomb, & Littman, 2018). We explore these interactions in the female-majority field of Education Research to ask: How does gender predict initiating a conversation or receiving a reply as part of the #AERA19 Twitter hashtag?

We also explore methodological issues relevant beyond the current research question. Namely, we present methods to analyze and interpret the intensive social network data scraped from Twitter, and we examine the accuracy of gender coding specifically.

For two weeks bookending the AERA 2019 conference, we collected tweets that contained the #AERA19 (or #AERA2019) hashtag—a total of 29,635 tweets from 7,923 users. To label tweets by gender, we first relied on pronouns self-identified within Twitter bios, which identified 18.25% of tweeters. Those without profile pronouns were classified using names from a 339,967-name historical census database through the use of the R package gender (Mullen, 2018). Using this system, 4,165 names (63.83%) were classified as men or women (non-binary genders were not coded through the gender package), with 2,360 unable to be classified (e.g., those for organization-related accounts, such as AERA). We had human coders gender code 5% (n = 397) tweeters and obtained satisfactory alignment with the automated coding (k = .72)

Limiting the sample to individuals who sent one or more original tweets (Table 1), we conducted social network analysis to explore the network (Figure 1). We then used a multilevel selection model to predict interactions between individuals. We found men were more likely than women to receive interactions—for every interaction women received, men received around 0.5 more (B = 0.39, SE = 0.18, p = .030, Odds Ratio = .47). We also found that women were somewhat more likely to interact with men than men were to interact with women, although this difference was not statistically significant.

Our analyses provide some evidence for gender imbalance among the Education Research community of practice and provides methodological insights through the application of a multi-level selection model that can be applied to other education-relevant network questions.

Authors

• Teomara Rutherford, University of Delaware

• Joshua Michael Rosenberg, The University of Tennessee - Knoxville

• Bret Staudt Willet, Florida State University