At what rates and in what capacity do women participate in radical right communities? Gathering precise demographic details – such as gender, age or location – for members of clandestine or extremist groups can be difficult. However, in the wake of the deadly “Unite the Right” event in Charlottesville, Virginia, one expert claimed that 20% of the alt-right might be female. The Anti-Defamation League (ADL) used video evidence to conclude that “alt-right is overwhelmingly white and male” – with the ADL only being able to identify 7% of the Unite the Right attendees as female. Before Charlottesville, a further 2016 psychological study yielded a self-identified sample claiming to be 34% female. Even earlier, a 2010 Quinnipiac University poll of the Tea Party movement (some of which subsequently morphed into the anti-government “patriot” militia movement of today), showed that women make up 55% of self- identified Tea Party members. So what are we to believe when looking at female involvement – especially in the online space – within the radical right?
How many women?
To better understand gender and participation rates in the online radical right, I accessed the public Facebook Graph API to create a large dataset of 700,204 members of 1,870 Facebook groups spanning 10 different far-right ideologies during the time period June 2017 – March 2018. Next I applied two different gender resolution software packages – Gender-guesser & Genderize.io API. (In prior articles about this data set, I described how it can be used to understand participation in anti-Muslim groups and to explain attendance at the “Unite the Right” event.) Table 1 shows the counts of groups and users in each ideology.
Table 1. Radical right Facebook groups, divided by primary ideology
Ideology | # Groups | Total Users | Max Group Size | Mean Group Size |
Neo-Confederate | 453 | 182,621 | 19,447 | 662 |
White Nationalist | 379 | 73,582 | 14,712 | 233 |
Anti-Govt/Militia | 273 | 101,211 | 11,509 | 473 |
Alt-Right | 246 | 99,996 | 36,666 | 587 |
Proud Boys | 157 | 7,920 | 1,348 | 72 |
Anti-Muslim | 136 | 128,467 | 17,824 | 1,270 |
Manosphere | 82 | 36,435 | 8,658 | 643 |
Anti-Immigrant | 51 | 115,511 | 51,117 | 2,823 |
Neo-Nazi | 48 | 6,218 | 1,251 | 139 |
Anti-Semitic | 45 | 16,498 | 9,310 | 400 |
In terms of mechanics, Gender-guesser classifies first names into one of five categories: male, mostly male, female, mostly female, androgynous, or unknown. Genderize.io classifies first names into one of three categories: male, female, or unknown. Table 2 shows the result of applying both these gender guessers to the 62,792 distinct first names in my Facebook data set.
Table 2. Results of gender resolution process
Gender-Guesser | Genderizer.io | |||
# | % | # | % | |
Female/mostly female | 176,829 | 25% | 198,116 | 28% |
Male/mostly male | 435,023 | 62% | 465,851 | 66% |
Androgynous | 6,924 | 1% | – | – |
Unknown | 83,384 | 12% | 38,193 | 5% |
To test the accuracy of the predictions, the software-generated gender guesses were compared to a pre-classified set of 1,855 users whose gender was specified by them in their Facebook public profile. Table 3 summarizes the results. Genderize.io yields fewer unknowns (5% vs 12%) and also has fewer errors (7.8% vs 13.2%).
Table 3. Error estimation for gender guessers
Gender-Guesser | Genderizer.io | |||
# | % | # | % | |
Correct | 1,611 | 86.8% | 1,711 | 92.2% |
Incorrect | 244 | 13.2 | 144 | 7.8% |
Both gender resolution packages yield figures (25% and 28% respectively) that are believable, given the prior estimates reported above of female radical right participation (20%, 7%, 34%, and 55%). Automated gender inference is imprecise, but even taking a maximum error rate into account,we can still say that these figures are still within the bounds given by prior estimates.
What groups do women join?
Now we can begin to study the behavior of these “users inferred to be women” on the Facebook platform. What types of groups do the women join? Do they prefer some ideologies over others?
Figure 1.
Figure 1 shows summary statistics for each inferred gender (male, female, unknown) as resolved by Genderize.io. These are sorted high-to-low by the proportion of female users participating in that ideology. Anti-Immigrant, Neo-Confederate, and Anti-Muslim have the highest participation rates by women, and unsurprisingly, the Proud Boys and Manosphere categories have the lowest participation rates by women.
“Wheat fields”
Are there particular groups within each ideology that attract more women? In the leaked chat logs from the Discord server allegedly used to plan the Unite the Right event held on August 12, 2017 in Charlottesville, Virginia, female-identified users were placed into a special role and chat channel called “Wheatfield Dwellers” [sic]. The term “wheat field” refers to stock photographs of attractive white women cavorting in fields of grass or wheat. These photographs are a popular meme in far-right online communities to refer to an idealized vision of white womanhood. Wheat field imagery has been used as cover art by online groups across multiple radical right ideologies, including the Alt-Right.com web site, and by former Ku Klux Klan grand Wizard David Duke on Twitter.
Within the Facebook extremist ecosystem, I did find several “wheat field” groups that were designed for women and also comprised mostly of women. Figures 2 and 3 show bar graphs of Facebook group participation by ideology. The bars show participation rates by women in each group, sorted high-to-low, left-to-right. Figure 2 shows five ideologies that have “wheat fields” – very tall bars on the left. These groups all were designed for women and have a supermajority of women members. Examples of these groups include “Proud Boys Girls”, “Daughters of Eve” (an Anti-Semitic “Christian Identity” group), and “Tradwives and Wives-to-Be” (promoting a racist, traditionalist lifestyle).
Figure 2. Five ideologies with “wheat fields” for women
Figure 3 shows ideologies without wheat fields. In general, participation in these groups is more equalized between genders, and there are no groups specifically designed for women.
Figure 3. Four ideologies without “wheat fields” for women
The hybrid case is the right-wing anti-government/patriot/militia movement, shown in Figure 4. This movement has no groups specifically designed for women, but nonetheless there are several groups with a supermajority of women members. What is attracting women to these groups? An evaluation of the top twenty groups in this category with the highest percentages of women members reveals that more than half of them are affiliated with a particular militia leader who makes daily Facebook Live videos. He is something of an “e-celebrity” in this particular subculture.
Figure 4. Some radical right anti-government/militia groups attract women but are not designed for women
Conclusion
By inferring gender based on first names in a large data set of radical right Facebook users, we are able to generate reasonable estimates of participation by women in the movement overall. Furthermore, we are able to determine whether different groups or ideologies have women’s auxiliaries, and we can begin to gauge whether those “wheat fields” are actually successful at attracting women.
The complete findings from this project include more details about the groups and ideologies collected, women’s leadership rates in both the wheat field and non-wheat field groups, and a discussion of the similarities between these wheat fields and the women’s auxiliaries of the 1920s Ku Klux Klan. The final paper, Which way to the wheat field? Women of the Radical Right on Facebook has been accepted for publication and will be presented in early January 2019.
Professor Megan Squire is a Senior Fellow at CARR, and Professor of Computer Science at Elon University in North Carolina, USA. See here profile here:
© Megan Squire. Views expressed on this website are individual contributors and do not necessarily reflect that of the Centre for Analysis of the Radical Right (CARR). We are pleased to share previously unpublished materials with the community under creative commons license 4.0 (Attribution-NoDerivatives).