Britain First’s Facebook page is gone. The data I took from it two days before the ban is not, and it shows what a far-right page’s audience actually rewarded.

On 14 March 2018 Facebook removed the pages of Britain First and its leaders. The page, its posts and the interactions beneath them disappeared from public view. Two days earlier, as part of my PhD research on political parties and Facebook, I had extracted a post-level record of it.

I have now written that data up as a preprint: “What Went Viral on Britain First’s Facebook Page? Outrage, Questions and the Cost of Mobilisation, 2013 to 2018.”

Why this data is rare

Studying a far-right page normally means studying it while it exists. Once a platform deletes a page, the posts can no longer be retrieved from it, and the tools researchers relied on at the time were already limited in how much of a page’s history they could reach. As far as I know, I hold the only post-level record of Britain First’s page that exists in the research community. It covers 3,505 posts from 25 October 2013 to 12 March 2018, which together drew 32.8 million reactions, comments and shares.

What I did

I coded the text of every post for its topic, whether it asked the audience to do something, whether it named the group’s leaders, and its sentiment. I then tested what predicted engagement, comparing posts published in the same month so that audience growth and timing did not drive the results. I also compared the page’s audience with the audiences of the Labour, Conservative, UKIP, Liberal Democrat and Green pages, matching on month and post format.

What the page’s audience rewarded

• A handful of posts did almost everything. The top 1% of posts drew 62% of all engagement.
• Posts about out-groups drew the most engagement. Posts about Islam drew 56% more engagement than comparable posts, posts about immigration 39% more and posts about crime 20% more. The share of posts about Islam rose from 15% in 2013 to 64% in 2016, then fell to 31% in 2018.
• Asking was associated with more engagement. Posts that put a question to the audience drew 137% more engagement, and posts that asked people to share drew 116% more.
• Building the organisation was not. Posts asking supporters to donate or join drew 58% less engagement, and posts containing links off Facebook drew 51% less. The words most tied to low engagement were sign, join, petition and membership. This is the paper’s most suggestive result and also its least stable one: it shrinks when low-engagement posts are trimmed. It points to attention and organisational capacity being different resources, but I would not rest much on it alone.
• Naming the leaders made no difference. The audience engaged with the cause and its enemies, not with Paul Golding or Jayda Fransen.
• This was a page built on anger. On the average post, 28% of reactions were angry, against 2 to 6% on the mainstream party pages. Its audience shared more than the audiences of the other pages I compared, and commented less than the Conservative, Labour and Liberal Democrat audiences.

Figure 1. What lifted and what sank a post

+137% Asks the audience a question

+116% Asks people to share

+78% Animals

+56% Islam and Muslims

+39% Immigration

+38% Women, children, family

+20% Crime

+6% Names its own leaders (not significant)

-51% Contains a link or URL

-58% Asks people to donate or join

Figure 2. Britain First’s audience shared, others commented

Britain First

15%

29%

56%

Conservatives

35%

11%

54%

Labour

24%

24%

52%

Lib Dems

23%

17%

60%

Comments Shares Reactions

What did not matter

I found no significant lift in engagement in the weeks of the Jo Cox murder, the EU referendum, the Westminster attack and the other terrorist attacks in the window, or when Donald Trump re-posted Britain First videos, and no evidence that posting more often on a day diluted the engagement each post received. The event tests have limited power, because the archive holds few posts in some windows, and I say so in the paper.

The limits

The archive is a retrieval-limited subset of the page’s output, not a census. It holds about 600 posts a year, which matches the cap the extraction tool worked under, and it may over-represent high-engagement posts. I treat the levels as descriptions of the archive, compare posts within months, and test how trimming the low-engagement tail changes the results. The hostile-topic and question effects shrink as I trim, which suggests my estimates are conservative. The fundraising penalty is the most fragile result, and the paper says so directly. These are associations in observational data, not causal effects.

Why it matters

Engagement on this page measured how much its audience reacted to provocation. It does not measure how strong the movement was.

If a platform’s ranking favours engagement, pages have a reason to produce more of what this page produced. And if you read engagement as strength, you will overestimate what a page like this could do.

I hold the same kind of data for the mainstream parties and their leaders over the same period, so there is more to come. The coded post-level dataset, without message text, is available on request.

The views in this post and the paper are my own.