A Bechdel Test for #MLA16:
Gendered Acts of Care on Academic Twitter

Introduction

Sitting at #s508 of the Modern Language Association annual conference in Austin in early January of 2016, I realized that the “care and repair” in the title of the panel wasn’t emanating solely from the panelists. Equally watching the room and its Twitter feed, I could easily connect the sounds of typing in the audience to the transcript unspooling on my screen. Conference tweeting is an act of what Lauren Klein has called carework; it cares by recording and annotating the work of our peers up on the dais, and it repairs by weaving in new conversations and supplying citations, thereby ushering into the room our absent colleagues while constructing a spontaneous bibliography.

But who, exactly, is doing this carework? Sitting in panel after panel, following Twitter carefully, I couldn’t shake an impression that #MLA16 was largely a community of women who, donating their attention to this task, may be performing this labor at a certain personal cost. For me, the thrill of keeping up and the desire to transcribe effectively sometimes shaded into a far less fun sense of obligation to type quickly and capture everything. Sometimes, I felt that my labors took away from the mental leisure needed to appreciate and reflect on the fantastic presentations I was transcribing.

Yet, at the same time, women actively shaping the academic record (at least, an academic record) had to be a good thing. Right?

What follows is a snapshot of a moment, rendered in charts—an effort to trace, empirically, the origin of my doubt.

Acknowledgements

Many thanks to Andrew Pilsch for the many hours he spent developing the interactive aspects of this interface and creating the d3 figures from my data. Readers would be stuck with some fairly ugly tables, and would have to scroll quite a bit more, without his labor.

Thank you as well to Eileen Clancy, who generously and helpfully reviewed an early draft of this project.

Theory

Tweeting as a Gendered Act of Care

This case study is not at all an attempt to accuse male tweeters of neglecting female panelists or undervaluing their contributions during conferences. Firstly, such an accusation would be largely unfounded (as the Results section shows). Secondly, making such accusations would posit male tweeting behavior (or female tweeting behavior) as a standard from which the other deviates—which would not only be unhelpful but also essentialist. Isolating some immutable female style and some immutable male style in order to make a judgment call about what supposed style is unilaterally better is not what I am after here. Rather, I want to elucidate the types of labor that female tweeters offered the DH community in the panels I went to at MLA16.

Tweeting is carework when tweets are used to support other scholars, whether that scholar is a panelist, another audience member, or someone not present at the conference or that particular panel. That support may come in the form of preserving and and transmitting their ideas ("signal boosting" so that the idea reaches a larger audience), reviewing their work (whether by giving a solid thumbs-up, constructive critique, or something between the two), offering suggestions for further research or revision, or providing citations (whether indicated by the presenter or proactively identified by the tweeter). Instead of allowing one's thoughts to drift, a tweeter consciously makes an effort to have a device ready and remains on the alert for ways to support the presenter and the audience, both inside and outside the room. In the same way, Clancy's Storifies are also a work of care; as she theorizes in a tweet, "I see conference Storifies as my offsite #DH helpdesk contribution to scholarly communication." Although not all tweeters may not be consciously theorizing why they are doing what they are doing, this carework can contribute powerfully to scholarship.

Small Data

My dataset is of 792 tweets is quite small. This small size resulted more or less incidentally, as I did not go into the conference expecting to come away with a small research project. Nonetheless, I argue that this dataset is useful and appropriate for the type and scale of my research question: Was I right to feel that women in the same panels as me were doing a majority of the Twitter labor being done on behalf of the larger scholarly community? This is offered as a relatively empirical response to a phenomenon subjectively perceived; it is offered as an early step in finding a method to explain what goes on in conference Twitter within the digital humanities community. It is important to remember that these three panels were digitally inflected and do not necessarily represent tweeters covering other #MLA16 panels, as well as to remember that Twitter users are probably more likely than the general MLA community to identify as digital humanists.

Small data has been presented recently as a useful corrective to the digital humanities' perceived overreliance on big data. In "Six Provocations of Big Data," Kate Crawford and danah boyd argue that "in the era of the computational turn, it is increasingly important to recognize the value of 'small data'. Research insights can be found at any level, including at very modest scales" (8). As long as the conclusions claimed do not overreach the scope of their data, these insights can be useful for scholars. Cheryl Ball agrees, pointing out, "In the rush to embrace big data as a funding stream for large, networked humanities projects, researchers in rhetoric and composition and the digital humanities might miss the value in the thousands, perhaps hundreds of thousands, of uncounted data sets we already have" (par. 5). The present case study has adapted one such uncounted set: Storifies based on conference tweeting.

Scott Weingart, in his 2014 DH Forum keynote, "The Moral Role of DH in a Data-Driven World," specifies the ethical value of small data by arguing that using only big data makes it "difficult to see the small but endless inequalities that leave women and minorities systematically underappreciated and exploited" (par. 28) Weingart points out that, by contrast, small data is "better at betraying the features of communities, rather than societies" (par. 29). If I want to learn about my own Twitter community—digital humanists who like to tweet at conferences—then small data may in fact be the best data for the job. In short, though I think the methods I outline can be regarded as scalable, I chiefly want to affirm my results insofar as they taken to reflect only the information that I fed into my statistics and figures.

The Bechdel Test

The Bechdel test dates from 1985, when cartoonist Alison Bechdel's comic Dykes To Watch Out For stipulated rules for judging a film's portrayal of women. In the strip "The Rules," Bechdel's character explains, "I only go to a movie if it satisfies three basic requirements. One, it has to have at least two women in it who, two, talk to each other about, three, something besides a man" (panels 4-5). Movies that meet these requirements are considered to portray women as fully rounded characters who are central (not peripheral) to the plot and express interest in issues other than normative heterosexual romance. (Digital humanists may know the test from Apoorv Agarwal et. al's paper on automating the Bechdel test or Scott Selisker's consideration of the test as a character network.)

Applied to academic Twitter, then, I wanted to discern whether women's scholarship was adequately represented. As is suggested by the popularity of the recent "Congrats, you have an all male panel!" Tumblr, setting a standard for each tweet or stream featuring at least more than one woman panelist might be unrealistic. Nevertheless, it is a wise goal to champion: as Jacqueline Wernimont pointed out to me, Monica Rogati sketched out precisely such an adaptation of the Bechdel test in a tweet from November 15, 2015. Rogati explains a "Bechdel test for tech conferences" as requiring "1) two women speaking 2) on the same panel 3) not about women in tech." If we replace "tech conference" with "academic conference" and "women in tech" with "women in DH," a Rogati-Bechdel test would reveal that #s280 (the disrupting panel), #s411 (the pedagogy panel), and #s508 (the repair panel) all passed.

For a panel's Twitter feed to pass such a test, however, more probing is necessary. Does a panel's Twitter stream add more female voices or scholarship beyond the female panelists or female scholars invoked by the panelists? And when panelists are women, or cite female scholars, do women's presentations and everyone's citations of female scholars get taken up equitably in the stream (or even elaborated upon?) Are female tweeters using Twitter to converse with one another? Are female DH scholars cited in other contexts than feminism or feminist DH? Finally, are women's tweets treated with the same kinds of respect and attention as men's tweets? Are they favorited, retweeted, or responded to at the same rate? Are they more or less likely to be neglected or criticized in ways that men's tweets are not?

The DuVernay Test

Manohla Dargis recently devised a race-based analog to the Bechdel test, the "DuVernay test," in her coverage of the Sundance Film Festival for the New York Times. Dargis notes that, despite of the efforts of some film industry denizens to "turn Sundance into a snowy exurb of Hollywood, the festival continues to push against the mainstream tide through some of its selections" (par. 2), particularly through films like The Birth of a Nation, Christine, and Sand Storm. Referencing Selma director Ava DuVernay, Dargis writes that "such films pass what might be called the DuVernay test, in which African-Americans and other minorities have fully realized lives rather than serve as scenery in white stories" (par. 6). Dargis's conclusion—"For women, for minorities, for those seeking something different, it can be hard to love movies, because they don’t always love us back. Sundance makes it easier" (par. 10)—suggests the stakes for tracing tweeting as an activity that is gendered and, indeed, raced. Twitter can be seen as a potential Sundance for women of color (and women more generally), offering an opportunity for keen Twitter users to use the medium to reflect their own research interests and their own scholarly communities.

Although I did not actively track the racial identification of all the tweeters or persons mentioned in tweets, and therefore cannot confirm the representational equity or diversity of this dataset with hard numbers, I will suggest that, in particular, the tweets from the "Disrupting the Digital Humanities" panel provide the kind of "an exhilarating, multifaceted portrait," particularly refreshing for "those seeking something different," that Dargis writes about (par. 9). What I can truly offer, however, is a short (no doubt incomplete) list of guidelines for what Dargis's standard of "fully realized lives" might translate into for academic Twitter: first, that minority presenters are equally represented; second, that minority audience members are represented and listened to; third, that minority tweeters feel free to express any concerns about any race-related issues raised by the presentations; and fourth, that accounts of scholarship by people of color are abundant, nuanced, and regarded as central to the Twitter stream, rather than as a peripheral or "boutique" concern.

Gender Identification

To sort tweets into gendered data, I applied categories (man, woman, or no gender identified) based on the tweeter as well as any person mentioned in the tweet. To determine gender, I relied in most cases on my prior or subsequent acquaintance with the person and his/her/their Twitter persona. For other cases, on what I regarded as publicly available signifiers of gender presentation, such as clothing and pronouns. For those I did not remember from MLA, I sought only readily available information put up online by that person, as any other effort would be invasive. It was a clumsy system, very unsubtle and reliant on my undoubtedly subjective perceptions of gendered categories that themselves are limited and limiting.

Despite these reservations, I hope to limit the negative effects of this initial choice to use a gender binary by not identifying particular tweeters' identities in my data (particularly in the downloadable Excel data) or at any point in this written discussion, not guessing individual tweeters' intentions (and thereby making further gendered assumptions about them), and not tracking the particular set of tweets by an individual. The only exception, as I discuss in the Methods section, was in the case of the two super-tweeters, in which case it was advisable to provide data both including and exclusing them to determine if super-tweeters skew results.

I am not aiming to theorize timeless gender categories or behaviors. Rather, I am concerned here with the "optics" of what happened there and then, on the panel's Twitter feed, and I have tried to circumscribe by discussion of what happened within a very specific context. A different presentation of the same data could, for example, slice up by data by comparing casual, active, and super-tweeters (as defined in the Methods section), rather than breaking it down by gender lines. The observations that are here represented as a gendered issue could thus be reframed as a matter of individual tweeting patterns. Still, I stand by this case study as a contribution to the very necessary work being done in the feminist digital humanities, particularly because those who would identify as digital humanists are probably chiefly represented in this dataset (as discussed above). Finding new ways to describe, define, and support the work of women done in the digital humanities is, I think, an important contribution to that work.

Methods

I quantified what happened in the DH panels I attended for which there are Storifies (courtesy of Eileen Clancy): #s280, #s411, and #s508. The first session, #s280, "Disrupting the Digital Humanities" (hereafter "the disrupting panel"), was led by Jesse Stommel and featured Rick Godden, Jonathan Hsy, Spencer Keralis, Eunsong Kim, Angel Nieves, Annemarie Pérez, and Jentery Sayers. (See the Storify here and the panelists' position papers here.) The second session, #s411, "Digital Scholarship in Action: Pedagogy" (hereafter "the pedagogy panel") was led by Marguerite Helmers and Daniel Powell and featured Amy Earhart, Aaron Mauro, Kimberley R. D. McLean-Fiander, Philippa Schwarzkopf, Angel Nieves, and Jacqueline Wernimont. (See the Storify here.) The third session, #s508, "Care and Repair: Designing Digital Scholarship" (hereafter "the repair panel"), was led by Jetery Sayers and featured Lauren Klein, Daniel Anderson, Lisa Marie Rhody, and Susan Brown. (See the Storify here and the panelists' abstracts here.)

No other panels’ feeds were incorporated because it was necessary to have been there, firstly, to sort tweets into categories accurately (for instance, does a tweet record what is being said in a presentation, or does it add some sort of new commentary, idea, or critique?), and secondly, to make the difficult judgment calls required to sort presenters and Tweeters into a gender binary (which I discuss further in the Theory section). This resulted in a set of 792 tweets: a tiny sample size compared to other research projects focusing on Twitter. I maintain that this small sample set minimized the number of mistakes in sorting tweets into topical categories and gendered categories; I double-checked, and in some cases triple- or quadruple-checked, each classificatory decision I made, which took quite a long time. (I even hand-checked every single statistic generated by my spreadsheet to confirm the accuracy of my Excel equations.) More importantly, though, this case study can be seen as taking part in a broader reevaluation of the dominance of "big data" in the digital humanities, as I argue in the Theory section.

When I reached out for permission to use her Storifies, Clancy pointed out that some super-tweeters might skew the results. I classify as casual tweeters those who did not regularly tweet (more specifically, they tweeted at a lower rate than one tweet per presentation), while active tweeters tweet at least once per presentation, and super-tweeters are only those who tweet continuously, producing tweets at such a markedly higher rate than active tweeters that they clearly dominate the panel's Twitter feed. In #s280 (the disrupting panel) and #s508 (the repair panel), each super-tweeter produced 31.50% and 31.46% of that panel's total tweets, respectively.

Given the not-fully-understood impact of super-tweeters (though I do reflect on this issue in the Results section), and given my small sample size, I cannot draw firm conclusions about “all” male and female academics on Twitter. But I am not concerned here with essentializing gendered tweeting habits—for, as I show, they are highly contextualized—beyond the scope of my data. And I am less concerned with individuals than with the record as it stands, which includes both of the super-tweeters in my sample.

I scoured the Storifies and recorded two sets of data: the first being the gender of all people involved in the tweet, including the tweeter and anyone named in the tweet, and the second being the topic or purpose of the tweet. For the first set of data (the gender of those involved), the categories I tracked included tweeters, presenters, non-presenters (typically, audience members and those not present in the room), and “no gender identified.” Tweets from organizations were not counted because their accounts did not emphasize gendered markers in their avatars or profiles. When more than one person is mentioned in a tweet, I defaulted to the first person mentioned because I did not want each tweet to count more than any other tweet. And because this case study is recording the rhetoric embedded in each tweet through structure and diction, to determine the tweet's primary interlocutor, I examined the grammar, punctuation, syntax, and other markers of emphasis in tweets that mention more than one person.

When a presenter and a non-presenter are mentioned, I counted the non-presenter because the primary purpose of such tweets is to emphasize this new person, even though the tweeter is also ensuring that the presenter would still be included in the conversation. Moderators and conference organizers counted as presenters if they made extended remarks to theorize the panel (something far more substantive than author biographies) and were visually present (seated literally at the table, close to and alongside the presenters), but as non-presenters if they did not make remarks beyond presenter biographies and sat with the audience or in a corner, away from the panelist's table. To avoid mislabeling a participant, if a person was invoked but not actually named, the tweet counted as “Person mentioned but not named,” unless the gender of the person invoked was explicitly indicated in that particular tweet through pronoun usage.

I also compared what I call the "expected versus the actual number of tweets" that cover the work of either male or female presenters. In the panels I track, there were 9 female presenters and 8 male presenters, so I assume that we could "expect," in a gender-neutral Twitter stream, 9/17 of the tweets that name presenters to be about female presenters, whereas 8/17 would be about male presenters. I then compared the actual number of tweets generated when male or female scholars presented to see if either gender was overrepresented or underrepresented in comparison with the expected 9:8 ratio.

The second data set collected was the topic—more accurately, the rhetorical function—of each tweet. I devised the following categorical schema: meta-comment, summary/quotation, citation, response/question, and room climate. Meta-comment tweets refer to the objective state of the panel (has it begun, who is speaking, what are the paper titles, who can or cannot attend), whereas room climate tweets comment on the atmosphere of the room (big, small, crowded, loud, empty, cold, hot). Both categories point to the event-ness of the panel, rather than its intellectual content, but meta-commentary tweets provide the same sort of information that the program gives (only given in real time), compared to the room climate tweets' social focus on characterizing the panel's spatial, material, or emotional atmosphere (thereby narrating the affective or physical experience of being present at the panel).

Summary or quotation tweets record what was said by panelists, or by audience members during Q&A, with an aim to reproducing accurately what is going on. Such tweets are reportage and make up the majority of "carework" tweets. By contrast, response or question tweets are evaluative in nature; they judge, reflect, and editorialize on the intellectual content of the panel so that the tweeter participates actively in the conversation. These tweets often ask panelists questions about their ideas, engage in conversation with other audience members, or bringing in an absent colleague into the Twitter stream. Such tweets are transformative or interactive rather than summative or mimetic. Summary/quotation tweets are essentially focused on the panelists' materials, whereas response/question tweets are focused on the audience's responses.

Citation tweets provide citations, giving URLs leading to work by the panelists; to books and articles mentioned by the panelists; or to work that the tweeter considers relevant but had not been invoked by the panelists. These tweets are outward-facing, unlike meta-comment or room climate tweets; they direct the reader to sources, scholars, or moments outside the room. Like summary/quotation tweets, Citation tweets perform important labor for the panel by spontaneously constructing a bibliography. These tweets can be profoundly political or interventionist in that they have a unique power (among these categories) to give credit. At other times, these tweets can be more stenographical in nature by offering further information about scholarship cited by presenters.

A few tweets might technically have counted in more than one category. I judged which category by counting the number of words that reflected the two categories and classifying it in the one with the greatest number. If the balance between the number of words fitting two categories was roughly equal, I classified the tweet according to which came first; for example, a tweet that begin with strong evaluative language and expressive punctuation (such as multiple exclamation marks), but then summarized the presenter’s words, was counted as response/question, not summary/quotation.

Tables for #s280

Table 1: Storified Tweets by Gender #MLA16 #s280

Female presenter named Male presenter named Female non-presenter named Male non-presenter named Person mentioned but not named Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Table 2: Storified Tweets by Gender #MLA16 #s280 Minus Super-tweeter

Female presenter named Male presenter named Female non-presenter named Male non-presenter named Person mentioned but not named Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Table 3: Storified Tweets by Topic #MLA16 #s280

Metacomment Summary or Quotation Citation Response or Question Room Climate Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Table 4: Storified Tweets by Topic #MLA16 #s280 Minus Super-tweeter

Metacomment Summary or Quotation Citation Response or Question Room Climate Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Tables for #s411

Table 5: Storified Tweets by Gender #MLA16 #s411

Female presenter named Male presenter named Female non-presenter named Male non-presenter named Person mentioned but not named Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Table 6: Storified Tweets by Topic #MLA16 #s411

Metacomment Summary or Quotation Citation Response or Question Room Climate Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Tables for #s508

Table 7: Storified Tweets by Gender #MLA16 #s508

Female presenter named Male presenter named Female non-presenter named Male non-presenter named Person mentioned but not named Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Table 8: Storified Tweets by Gender #MLA16 #s508 Minus Super-tweeter

Female presenter named Male presenter named Female non-presenter named Male non-presenter named Person mentioned but not named Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Table 9: Storified Tweets by Topic #MLA16 #s508

Metacomment Summary or Quotation Citation Response or Question Room Climate Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Table 10: Storified Tweets by Topic #MLA16 #s508 Minus Super-tweeter

Metacomment Summary or Quotation Citation Response or Question Room Climate Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Aggregate Tables For All Panels

Table 11: Storified Tweets by Gender #s280, #s411, #s508

Female presenter named Male presenter named Female non-presenter named Male non-presenter named Person mentioned but not named Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Table 12: Storified Tweets by Gender #s280, #s411, #s508 Minus Super-tweeters

Female presenter named Male presenter named Female non-presenter named Male non-presenter named Person mentioned but not named Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Table 13: Storified Tweets by Topic #s280, #s411, #s508

Metacomment Summary or Quotation Citation Response or Question Room Climate Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Table 14: Storified Tweets by Topic #s280, #s411, #s508 Minus Super-Tweeters

Metacomment Summary or Quotation Citation Response or Question Room Climate Total tweets by gender
Female Tweeter
Male Tweeter
Total Tweets

Figure

Figure 1: Expected vs Actual Tweets About Female & Male Presenters

Figure 2: Expected vs Actual Tweets About Female & Male Presenters (w/o Super-Tweeters)

Figure 3: Mentions of Non-Presenters By Gender

Figure 4: Mentions of Non-Presenters By Gender (w/o Super-Tweeters)

Figure 5: Breakdown of Topics by Gender (Women)

Figure 6: Breakdown of Topics by Gender (Women w/o Super-Tweeters)

Figure 7: Breakdown of Topics by Gender (Men)

Figure 8: Breakdown of Topics by Gender (Men w/o Super-Tweeters)

Figure 9: Breakdown of Mentions By Gender (Women)

Figure 10: Breakdown of Mentions By Gender (Women w/o Super-Tweeters)

Figure 11: Breakdown of Mentions By Gender (Men)

Figure 12: Breakdown of Mentions By Gender (Men w/o Super-Tweeters)

Results

Women out-tweeted men—by a lot.

During these three panels, women produced 570 tweets (71.97%), while men tweeted 222 times (28.03%) (see Table 11). In terms of doing the labor of recording what was going on in these panels, women produced 319 tweets that summarized or quoted from the panels, whereas men produced 124. In terms of providing citations to help those on Twitter find proof, articles, books, or websites related to the panel, women provided 86 citations, while men provided 17 (see Table 13).

Women presenters slightly outnumbered male presenters (9 to 8), which might at first seem to suggest that a greater number in women presenters correlates to a greater level of female participation on Twitter. Yet, at #s280 (the disrupting panel), with 2 women and 5 men on the panel, women contributed 87.14% of the total tweets (see Table 1). For #s411 (the pedagogy panel), with the proportion of men and women nearly reversed, with 4 women and 2 men on the panel, women contributed 78.47% of the total tweets (see Table 5). For #s508 (the repair panel), with 3 women and 1 man on the panel, women contributed 46.82% of the total tweets (see Table 7).

Female panelists are not underrepresented on Twitter.

Both men and women were more than happy to tweet about presenters of the other gender, though women were particularly likely to mention women who were not presenting and women who were not present at the panel. Women quote female presenters in 144 tweets and male presenters in 114 tweets, a ratio that corresponds closely to the 9:8 gender ratio of female to male representers, suggesting that female tweeters in these three panels maintained an egalitarian dispensation of tweets. By contrast, if we examine the gender ratio of presenters quoted by male tweeters (see Figure 9). we find that men named female presenters in a whopping 41.89% of their tweets, but named male presenters only 17.12% of the time (see Figure 11). Why did women not name female presenters as frequently as men did? Perhaps it is partly explained because women are more likely than men to tweet without any reference to a particular person (201 tweets by women versus 55 tweets by men), but this latter phenomenon may be due to the super-tweeter in #s280 (compare Figure 9 and Figure 10). Still, considering raw numbers (totals, not percentages), women named female presenters in 144 tweets, but men did so in 93 tweets (see Table 11).

In terms of non-presenters, women tweeters use 19.30% of their tweets to reference non-presenters (see Table 11). These tweeters mention female non-presenters 76 times, devoting 13.33% of their tweets to doing so, suggesting that female tweeters are very keen on bringing in women who are relevant to the conversation who would otherwise be left out (see Figure 3). Female tweeters bring in male non-presenters at a markedly lower rate than they bring in women—34 times (5.96% of their tweets)—but in raw numbers, this is twice the number of male non-presenters brought in by male tweeters. In terms of quantity, then, women are over three times more productive in terms of broadening the conversation to both men and women who are not presenting or who are not in the room.

However, it should be noted that in terms of percentages by gender, men tweeting in these sessions were more than fair in naming female presenters and were slightly more likely to bring in new women (19 times in total) to the conversation than they were new men (17)—again showing that men in these panels were not at all intent on supporting other men over women (see Figure 11). And, regarding both men and women’s tweets about non-presenters, female non-presenters were mentioned in 13.04% of all tweets, whereas male non-presenters clocked in at 6.56% of all tweets, so, in these three panels, we might say that male non-presenters were underrepresented. Controlling for the super-tweeter in #s280 (the disrupting panel) did not appreciably change these numbers (see Table 2). However, controlling for the super-tweeter in #s508 (the repair panel) showed that in that panel, tweeters overrepresented the male presenter, producing 7.38% more tweets than an egalitarian dispensation would (see Figure 2)—but this was by a slimmer margin than female presenters were overrepresented in other panels. The disrupting panel, #s280, produced 9.71% more tweets about female presenters than expected; for #s411, the pedagogy panel, this number was 12.64% (see Figure 1).

Men were more likely to tweet meta-commentary, whereas women were more likely to tweet about room climate.

Regarding the topic, or rhetorical function, of tweets, men were twice as likely (8.56% of their tweets) as women (4.21% of theirs) to provide meta-commentary, which expressed procedural facts about the state of the panel: when the paper started; who was presenting; the full paper title (compare Figure 5 and Figure 7). Women contributed 83.18% of the tweets about the room climate—posting 37 tweets about the size, quality, and comfort of the room or about the emotional atmosphere at the panel (see Table 13). This difference suggests a gendered division of labor regarding what kinds of observations about the event, beyond the substance of the papers, were perceived as interesting or important to record.

Men and women were equally as likely to provide summary or quotations that recorded the panel: nearly identical at 55.96% of women’s tweets and 55.86% of men’s. In terms of raw numbers, the gap is appreciable: 319 by women versus 124 by men. Women inserted themselves into the conversation by responding or asking a question about 5% less frequently than men did (18.25% versus 22.97%), though in terms of raw numbers, again, women responded or asked a question 104 times versus men’s 51 times (compare the numbers of Table 13 with the percentages of Figure 9 and Figure 11).

Women were more likely to use their tweets to provide citations.

Women contributed 84.29% of the 103 total citation tweets—tweets that explicitly linked to sources that were either quoted by the presenters or were related to the topics at hand (see Table 13). Women tweeters collectively contributed an average of 28.26 citations per panel, compared with men’s collective citation rate of 5.67. Panel #s280 (the disrupting panel) featured 50 by women and 11 by men (see Table 3), #s411 (the pedagogy panel) featured 20 by women and 2 by men (see Table 6), and #s508 (the repair panel) had 16 by women and 4 by men (see Table 9). Proportionally, tweets by men were citations 7.66% of the time, whereas women’s tweets were citations 15.09% of the time (compare Figure 5 and Figure 7).

Because they often name people not in the room, these citation tweets are works of care. On behalf of the presenters, they perform the bibliographic labor of providing proof or materials for further reading, and on behalf of people not present at the panel, they recover the names of scholars who contributed to the results, project, or argument being discussed. Combining the citation results with the non-presenter results, women are far more likely to bring in people not physically at the conference. (What were men tweeting instead? Men provided more meta-comments and responses/questions.)

Tweeting was highly contextual and responsive.

These tweeting patterns varied according to their context. In other words, the proportions of each rhetorical function of tweet varied shifted in response to the type of presentation, the composition of the audience, and the individual habits of tweeters who were more active than others during a given panel. In #s280 (the disrupting panel), an unusual audience—one that tweeters felt had a low number of white men—resulted twice as frequent room climate tweets (compare Table 3 to Table 6). Once an active tweeter (e.g., one who tweeted at least one time per presenter, but not as frequently as a super-tweeter) tweeted a response/question, the overall rate of question/response tweets would increase appreciably.

In #s508 (the repair panel), Dan Anderson’s remarkable multimedia performance evoked tweets that were harder to classify in terms of their rhetorical function: were they evaluative (response/question) or descriptive (room climate, meta-comment) or reportage (summary/quotation)? Tweets during Anderson's presentation were more likely to participate in two or more categories, mixing summary with response or citation with room climate. A creative presentation, it seems, generated more creative tweets. During #s508, which featured 1 male panelist and 3 female panelists, the panel yielded almost precisely the proportion of tweets about panelists that we might expect: 74.86% of tweets naming presenters were about female panelists (see Figure 1). Broken down, when women named panelists, they named women 67.12% of the time while men did so 80.39% of the time. Even though I had expected the novelty of Anderson’s presentation to have resulted in a higher rate of tweeting about male presenters than female, it did not. That is, unless the super-tweeter’s tweets are taken out (in which case, the proportion of tweets about male presenters increases from 25.14% to 32.38%, see Figure 2 or compare Table 7 and Table 8).

Super-tweeters may influence others’ tweeting behaviors.

The two super-tweeters in the three panels under consideration produced 204 tweets (25.76% of the total tweets). The super-tweeter in #s280 (the disrupting panel) contributed 120 tweets (31.50% of that panel’s tweets), while the super-tweeter of #s508 (the repair panel) contributed 84 (31.46% of that panel’s tweets). (See Table 12 and Table 14 to see the aggregate figures without super-tweeters.) The super-tweeter of #s508 (the repair panel) single-handedly negated other gender discrepancies in addition to the one mentioned above regarding Anderson’s presentation. For example, without the super-tweets, male tweeters are less likely to provide summaries/quotations (down to 43.48% of men’s tweets from 55.86%) and more likely to provide responses/questions (up to 29.71% of their tweets from 22.97%) (compare Table 9 and Table 10). And it was in the presence of a male super-tweeter than women tweeted summary/quotation at their lowest rate (only 22.85% of the total tweets, compared with 54.33% in #s280, the disrupting panel), in addition to their lowest rate of citation tweets (only 5.99% of total tweets, compared with 13.12% in #s280, the disrupting panel (see Table 9).

Women’s tweeting behaviors were more similar to those of the super-tweeters than men’s tweeting behaviors were: compare the similarities between Figure 5 and Figure 6 with the differences between Figure 7 and Figure 8. But with the super-tweeters taken out, women were less likely to provide the stenographical carework (summary/quotation, citation tweets) and more likely to inject their own perspectives (response/question tweets). In addition, taking out super-tweeters, the proportion of tweets by women that named presenters and non-presenters rose, suggesting that even one super-tweeter can free up women to do other kinds of Twitter labor than pure recording. Because the two super-tweeters, one male and one female, behaved quite similarly in terms of their dominance of the panel's feed (around 30% of the panel's feed for both of them) and in terms of the content of their tweets, it appears that male and female super-tweeters can equally encourage other tweeters to produce a wider range of tweets.

Perhaps most significantly, as I noticed while moving chronologically through each session’s twitter stream in order to gather my data, during phases in the panel when a super-tweeter was especially active, everyone else’s tweets seemed to become more complex and active. A greater number of people are brought into the conversation, more citations are provided, and more tweets cross rhetorical categories, suggesting that the labor of super-tweeters plays an important role in determining the kinds of labor that others (both male and female tweeters) perform.

Conclusion

Passing the Tests

Did #MLA16 pass the Bechdel test? In a word, yes. Women mentioning or citing other women constituted 27.78% of all tweets. Of the 792 total tweets, 220 of them—almost precisely the total number of all men’s tweets (222)—were women writing about women. (By contrast, men tweeting about other men constituted only 6.94% of all tweets, whereas women tweeting about men constituted 18.18% of all tweets.) During these panels, women performed the bulk of the carework, ensuring that panels were summarized, works were cited, links were shared, and audience members and those not present in the conference were brought into the experience remotely. Women made sure that other women’s voices were heard.

Did #MLA16 pass the DuVernay test? I think so—although this may be because the topic of these three panels each invoked some sort of carework or social justice topic. The disrupting panel, #s280, featured issues of race, while the pedagogy panel, #s411, emphasized taking care of students, and the repair panel, #s508, invoked care in the panel's very title. Tweeting diverse voices, whether diversity is measured by race, class, sexuality, able-bodiedness, or other factors, seems to be a priority of female tweeters. In general, the labor of female tweeters pushes each panel's Twitter stream closer toward the finish line to meet the standards of the Bechdel and DuVernay tests. When women or minorities present, female and minority tweeters are there to publicize and respond to their presentations. My favorite tweet of all three streams was one in which one woman-of-color tweeter cheered on the work of a woman-of-color presenter, which made me think of that Field of Dreams adage, "If you build it, they will come:" When we build inclusive panels, we set the stage for inclusive academic social media.

Advice for Academic Tweeting

For those of you who tweet at conferences, I would like to share the following advice. Beyond the continued necessity of making sure we build inclusive panels, my advice for the future is to continue these labors to represent minority and women voices on Twitter. However, do not restrict all of your social media labors to taking care of the panelists, of the audience, and of other academics. Unless you genuinely enjoy pure reportage (in which case, thank you for your work!) and are not interested in other forms of social media participation, make sure you take care of yourself as well: cite yourself more and share more of your own ideas, reactions, and questions.

Such self-carework is not selfish: it will also help the panelists develop their ideas, giving them more a more fully developed "peer review" experience that they can use to improve their future scholarship. It can indeed help you introduce other people to your own work as they read your take on the panel, but this does not have to feel like bald self-aggrandizement if your reactions are organically connected to the presentation at hand and are offered as a constructive part of the Twitter community that arises around the panel. If a super-tweeter is busy working (and perfectly happy about doing so), let them do that labor, and show your appreciation by crafting other kinds of tweets and participating in backchannel exchanges.

At the same time—bearing in mind that I do not want to say, once and for all, that a certain mode of tweeting is always better than another—what is so important about this kind of carework is that you can support the scholars and scholarship you in particular care about. If you truly perceive your role to be pure reportage during a particular panel, this is obviously a laudable goal, and it does not prevent you from taking on a far different role at other panels or other conferences. Perhaps there truly is a community that you want to take care of; indeed, regarding the DuVernay test, I'm sure that men and women of color are not waiting for this paper to be perfectly aware of their sacrifice and labors in ensuring that the voices of scholars of color are brought in.

Above all, do not let a perceived duty to document squeeze out your chance to comment, argue, connect, link, and intervene in the conversation. Panelists occupy a certain position of authority by virtue of being up on stage in front of an audience, so Twitter is an important way to diversify the voices being heard, both during and after the panel. As for male tweeters, you may want to take on more of the labor of sharing citations and bringing in more people to the Twitter feed. In addition to providing support for the scholars and scholarship that you find important, doing so may free up women and scholars of color for contributing in other kinds of ways to academic social media.

New Concepts to Explore in Social Media

Looking back at this case study, I now see opportunities for developing further a few insights I had along the way: 1) the categorization schema I developed to analyze the content or rhetorical function of a tweet; 2) the spontaneous production of bibliographies during conference tweeting; 3) the use of tweeting to bring in people who are not in the room or yet in the conversation; and 4) the influence of super-tweeters on the behavior of casual and active tweeters. These are important issues that I hope can spur further research. For now, I will hazard a few conclusions about these issues. Regarding #1, I am still happy with the categorization schema I devised (metacomment; summary/quotation; citation; response/question; room climate), and I hope it can be useful to others doing research on academic twitter.

Regarding #2, I have already suggested that the purposes of such a spontaneous bibliography differ from those of traditional ones (they allow scholars beyond the presenter to point the audience to sources), but how, precisely, are these resulting bibliographies similar to those that would be (or are) published by the panelist? What kinds of voices do they recuperate? Do they critique or support the paper being read? In my own experience, I react to what other tweeters in the room are doing; if there is a super-tweeter, I can rely on this person to pass along citations actively mentioned by the panelist, thus leaving me free to add in new sources that I think might help the panelist or give the audience background information about the larger intellectual debate around the given topic.

Regarding #3, the "outsiders" brought "inside," how are these people brought in—mostly through direct call-outs through "@" or through the citational networks generated by the tweeters? Are scholars who have active Twitter accounts more likely to be invoked? Are open-access works more likely to be cited? In my own experience, I am more likely to bring in active Twitter users because I expect that I will have a better chance of receiving a response from the scholar being invoked if the person has an "@" to be pinged by. And I am more likely to link to open-access works (or to books that are relatively cheap on Amazon!) because I feel they will better help anyone reading the conference Twitter stream.

Finally, regarding #4, based on my interaction with this dataset, I would venture that super-tweeters allow other tweeters to vary their Twitter behavior without feeling that they have let down the panelist or any readers of the conference Twitter stream by neglecting to communicate a crucial point. I would also venture that super-tweeters encourage greater participation in the backchannel; their followers on Twitter know they can come to this person's feed to stay in the mix, and those in the audience can sit back and wait for the super-tweeter to say something that provokes them to tweet back to the super-tweeter.

Avenues for Future Research

To explore these questions, as well as improve the accuracy of my results, I would make quite a few changes if I ran such a test again. First, I would attend more panels in order to extend the dataset, although if tweeters knew in advance their tweets would be recorded as gendered behavior, it might alter those behaviors. This larger sample size would increase the accuracy and statistical significance of my results, which at present hover somewhere between anecdote and evidence. Nevertheless, I think this case study usefully exemplifies new approaches to "small data" or "boutique data." I was able to explore a phenomenon I perceived during MLA16—an insistent though vague hunch that something was happening that I did not like—that I may not have perceived had I been purposively amassing large number of data points. And that, of course, is one of my major recommendations: to stay in the moment, whatever that may mean to an individual conference-goer.

Second, I would track my two types of data (the genders of people tweeting and mentioned in tweets; the topic or rhetorical function of each tweet) across time. Doing so would reveal chronological trends, such as when different types of tweets wax or wane during a single panel or conference, or how influential tweets or tweeters change overall patterns in a given panel. Such chronlogical data is, of course, embedded in each tweet's metadata, and I could return to my data and transform my synchronic figures and tables into a timeline or video. However, this problem was neither in the scope of my original research question nor necessary for the theses sketched out in the Results section. Quite simply, this dimension of the data had not seemed relevant until after I had anlayzed my data and realized what questions remained.

Third, I would try to redesign the way I classify tweets to accommodate a more inclusive understanding of gender. I would examine the limits of the simple gender binary I have used (e.g., do tweeters really self-identify their genders, and are two genders really enough here?) and try to find a model that reflects the uses by, and specific affordances of, twitter for the LGBTQ* community. This might involve the circulation of a survey, or a far more closed, purposive experiment carefully designed with the help of an ally organization.

Finally, I would consider whether race could be a useful distinction to make in collecting and organizing this type of data (rather than consider race after-the-fact, as I must admit that I did)? What about women of color? Are they performing more of this carework, or different types of carework, than white women? Considering the undoubted influence of institutional racism in academia, listening to women of color on Twitter, retweeting them, and mentioning their work in your tweets, represents ways we can all perform carework for women of color in academia. Academic twitter, then, may become one place where academia can truly pass the DuVernay Test, as women of color tweet to ensure that digital scholarship by people of color is recognized and engaged with in ways that are more than cursory name-checks or token gestures of inclusion. The effort by women of color to bring in other female scholars of color was palpable especially during #s280, the Disrupting DH panel—during which the diversity of the audience (not only the panel as a whole) was explicitly noted. It seems quite likely, then, that this panel disrupted not just DH, but #DH as well.

Works Cited

Agarwal, Apoorv et al. "Key Female Characters in Film Have More to Talk About Besides Men: Automating the Bechdel Test." Paper presented at Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, Denver, Colorado, June 1, 2015. http://aclweb.org/anthology/N/N15/N15-1084.pdf

Ball, Cheryl. "From Big Data to Boutique Data." Digital Rhetoric Collective. Last updated November 12, 2013. http://www.digitalrhetoriccollaborative.org/2013/11/12/from-big-data-to-boutique-data/

Bechdel, Alison. "The Rules." DTWOF: The Blog. Last updated August 16, 2005. http://alisonbechdel.blogspot.com/2005/08/rule.html

boyd, danah and Kate Crawford. "Six Provocations of Big Data." Paper presented at A Decade in Internet Time: Symposium on the Dynamics of the Internet and Society, Oxford, United Kingdom, September 21, 2011. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1926431

Clancy, Eileen. "I see conference Storifies as my offsite #DH helpdesk contribution to scholarly communication. #mla16 #scholcomm." Tweet, January 9, 2016. https://twitter.com/clancynewyork/statuses/685891707330781184

Dargis, Manohla. "Sundance Fights Tide With Films Like ‘The Birth of a Nation.’" New York Times, January 29, 2016, accessed May 25, 2016. http://www.nytimes.com/2016/01/30/movies/sundance-fights-tide-with-films-like-the-birth-of-a-nation.html?_r=0

Klein, Lauren. "The Carework and Codework of the Digital Humanities," last modified June 8, 2015. http://lklein.com/2015/06/the-carework-and-codework-of-the-digital-humanities/

Rogati, Monica. "The Bechdel test for tech conferences: 1) two women speaking 2) on the same panel 3) not about women in tech." Tweet, November 15, 2015. https://twitter.com/mrogati/status/665962306694615040

Selisker, Scott. "The Bechdel Test and the Social Form of Character Networks." New Literary History 46.3 (2015): 505-523. https://www.researchgate.net/publication/284433910_The_Bechdel_Test_and_the_Social_Form_of_Character_Networks

Weingart, Scott. "The Moral Role of DH in a Data-Driven World." Paper presented at the 2014 DH Forum, Lawrence, Kansas, September 13, 2014. http://www.scottbot.net/HIAL/index.html@p=40944.html