BBC Panorama research: Misogyny and abuse on Twitter before and after Elon Musk’s takeover

6 March 2023

Research by ISD and CASM Technology for BBC Panorama shows how Elon Musk’s Twitter takeover fostered a permissive environment for the creation of tens of thousands of new accounts, which proceeded to immediately follow known abusive and misogynistic channels on the platform.

 _________________________________________________________________________________   

Summary

• ISD and CASM analysts collected data from 7M followers of a seed list of 132 Twitter accounts known to have spread abusive and misogynistic content.  We also created a smaller subset of ~600,000 followers of 10 of these seed accounts, who suggested a specific misogynistic focus in their username or biographical description, for a deeper dive analysis.
• There were clear spikes in numbers of these follower accounts being created – around 5-6 times normal levels – on days related to (i) Elon Musk’s takeover of Twitter and (ii) the return to Twitter, and later, the arrest of Andrew Tate.
• Even filtering out these spikes, the average number of newly created accounts following our list of known misogynistic and abusive channels jumped to an average of ~7,900 per day after Musk’s Twitter takeover, an increase of 69% on pre-takeover baselines.
• These newly created accounts exhibited largely pro-Trump, pro-free speech, and anti-liberal sentiment, as well as conspiracist themes.  Misogynistic agendas and/or ideologies were not a dominant narrative, though some accounts used misogynistic behaviour as part of these broader themes (e.g. to attack liberal women).
• These findings suggest Musk’s takeover of Twitter created the perception of a permissive environment for the creation of tens of thousands of new accounts that immediately followed known abusive and misogynistic channels.
• This research took place under the threat of imminent closure of the free Twitter API, which forced analysts to make up-front decisions around data collection rather than being able to rely on usual practices of iteration and experimentation. This reiterates the threat to public interest research of increasingly restricted access to social media data.

Approach

This study aimed to examine the impacts of Elon Musk’s Twitter takeover (October 2022) on misogynistic behaviour on the platform. Online misogyny is a challenging topic to study using many established approaches to social media research; in particular, tracking uses of keywords is often ineffective as misogynistic language is used in a wide variety of contexts, which might not signal misogynistic intent. Additionally, if the set of keywords is too precise or too broad, the dataset can be incomplete or irrelevant and full of spam (which was found to be especially the case for misogynistic language on Twitter).

Our approach therefore began with developing a list of 132 Twitter accounts that were qualitatively evaluated to have displayed misogynistic or abusive online behaviour, or which self-identified as being part of the wider ‘manosphere’. This list mixed high-follower influencer accounts and low-follower accounts involved in more targeted hate; it also mixed accounts focused on pro-masculinity and anti-women content and accounts found to deploy weaponised misogyny as part of a broader range of hateful content.

Next, we collected the 7,318,438 unique followers of these accounts, on the basis that followers generally— though not always— sympathise with the views and/or support the behaviour of people they follow. As such, this dataset would give us an insight into the behaviours of a much larger number of potentially relevant accounts, to compare against the timeline of Musk’s Twitter takeover.

Given the full list of 132 seed accounts also included accounts with a range of outputs, not necessarily focussed solely on misogynist abuse, we also selected a smaller list of 10 accounts whose usernames and descriptions suggested they were more likely to focus their content on discussing women, and then proceeded to collect 596,312 followers of these accounts.  The trends from this analysis corroborated the findings from the larger list, allowing for further qualitative assessment.

In addition to the above, we also experimented with various keyword searches, which might demonstrate trends in misogynistic behaviour across all of Twitter, not simply limited to these collected accounts. However, for the reasons outlined above (widespread use of misogynistic language without clear misogynistic intent and/or excessive spam) this did not reveal clear trends.

Findings

In this investigation, the clearest trend found were large spikes in newly created accounts following known misogynistic and abusive accounts on Twitter, correlating with notable events in Elon Musk’s takeover of Twitter, as well as around misogynist influencer Andrew Tate.

Specifically, we saw extremely high peaks of accounts created around 26 April (when Musk announced he was buying Twitter), 28 Oct (when the Twitter deal was completed), 19 Nov (when Andrew Tate’s Twitter account was reinstated), and 30 Dec (when Andrew Tate was arrested).

During these peaks, between 22k and 29k accounts were created per day, which would go on to follow these known abusive and misogynistic channels. This was around 5-6 times the average number created on other days among this set in 2022 (4,700 accounts per day).

Even after filtering out these spikes, the average number of newly created accounts following our list of known misogynistic and abusive channels jumped to an average of  ~7,900 per day after Musk’s Twitter takeover, an increase of 69% on pre-takeover baselines.

Figure 1: Number of accounts following our 132 seed list accounts created per day.  The graph only shows numbers from January 2022 to January 2023, but there are no peaks of similar size in earlier years.

We also investigated words, topics, hashtags, and @-mentions used in tweets and biographies amongst these newly created accounts. Following our focus on the impacts of the Musk takeover, we focussed on the accounts created in the 26-30 Oct peak (when Musk took control of Twitter).  Dominant themes included:

• Support for freedom of speech
• Opposition to liberal or ‘woke’ views
• Some conspiracist content, particularly against COVID-19 vaccines and lockdowns
• Support for Trump and MAGA Republicanism

See, for example, the below visualisations of the top 10 most-retweeted tweets by these new accounts, most commonly used hashtags, and most @-mentioned accounts. Many of these newly created profiles following known misogynistic accounts were highly engaged with Elon Musk’s content, with the Twitter owner the most @-ed account in the dataset. This is not a huge surprise – both because these accounts were likely created as a response to Musk’s takeover, but also reflecting broader trends in Musk’s interests; analysts at ISD have previously highlighted how Musk’s interactions with right-wing Twitter accounts (ranging from users who broadcast strong right-wing or far-right views to accounts known to peddle conspiracy theories and disinformation narratives typically associated with the right-wing) increased by a staggering 1,690% after he became CEO.

Figure 2: Top 10 most retweeted messages among newly created accounts following known misogynistic and abusive channels, after Musk’s takeover of Twitter in October 2022.

We did not see clear trends related to misogyny, or feminism/gender as a theme (the exception was opposition to transgenderism and abortion). However, we saw a number of accounts using misogynistic behaviours as part of the themes outlined above, for example to attack prominent liberal women (such as Nancy Pelosi or Jacinda Ardern) or feminism, transgenderism, or abortion.

Conclusions

These findings do not necessarily speak to changes in the overall level of misogynistic abuse on the platform in the wake of Musk’s takeover, a phenomenon which as noted above is exceptionally difficult to analyse systematically. However it does suggest a clear relationship between key events in Musk’s purchase of Twitter – as well as the reinstatement of banned accounts like Andrew Tate – in creating an environment which attracted users to create tens of thousands of new accounts which immediately followed known abusive and misogynistic channels.

We should also note that this research drew heavily on use of the free Twitter API (application programming interface) to collect a large number of accounts in order to conduct a range of analysis. This research was conducted in February 2023, the same month Twitter announced that it would be suspending support of the platform’s free API. Our researchers proceeded to carefully select data in order to ensure we would have complete datasets before the threatened shut off. This limited options for iteration and experimentation in which social media research usually benefits. The threatened suspension did not finalise on the planned date, with prolonged uncertainty around when and if this will happen. This illustrates the ongoing issues of data access to social media platforms that are already affecting public interest research.