Short answer
Not well. In my work on Britain, Scandinavia and the United States, Facebook engagement tracked how a party was doing at the time, but it was badly calibrated across parties and gave little or no advance warning of changes in support.
What the evidence shows
- In Britain from 2010 to 2020, UKIP and the Brexit Party each took about a third of party-page interactions in the months they were active while polling under 10%; the Conservatives took a fifth while polling 36%.
- Within a British party, engagement and polls moved together month to month, but I found no evidence that engagement moved first.
- In Sweden and Norway the radical right drew two to three times its poll share in engagement every month. Across 24 Scandinavian parties, a typical three-month change in Facebook share was followed by a poll change of about a tenth of a point, against typical poll movements of one and a half points.
- Facebook gave no clear warning of the Danish People's Party's collapse at the 2019 election.
- In 20 contested 2020 US Senate races, a candidate's share of engagement was not associated with their vote once the state's presidential vote was taken into account.
Caveats
Engagement measures who is active and loud on a platform, not the electorate. It can describe a campaign; it is a poor forecast.
Read the research
Who Carries the Party on Facebook? Leaders, Party Pages and the Polls in Britain, 2010 to 2020
AbstractPDFNot a Forecast: The Radical Right on Facebook in Sweden, Norway and Denmark, 2014 to 2020
AbstractPDFDifferent Campaigns, Similar Totals: Republican and Democratic Senate Candidates on Facebook in 2020
AbstractPDF