Can the engagement a party gets on social media tell us how it will do at the polls? Studies of Twitter have reached mixed and often optimistic answers. Parties’ own Facebook pages, where supporters respond to what the party posts, have rarely been tested across many elections. So I tested them.
The paper
Do Likes Predict Votes? Facebook Engagement and Election Results in Europe, 2010 to 2020
Read the paper (open access) · Abstract and PDF on my site
What I did
For 207 party-elections in 34 national elections in 15 European countries between 2010 and 2020, I compared the share of Facebook interactions each party’s page received in the four weeks before the election with its share of the vote. The posts come from CrowdTangle, Meta’s former tool for tracking public pages.
What I found
Among the 181 party-elections where the party had a previous result, engagement shares were related to vote shares (a correlation of 0.60). But the previous election result was far closer to the new one (0.94). Once the previous result is known, the campaign engagement share added little, and what it added could not be told apart from the size of the party’s audience before the campaign began.
Changes during the campaign did no better. Among parties whose vote share moved by at least a point, the direction of the change in engagement matched the direction of the change in votes in 52% of cases (62 of 120), which is close to a coin toss. In forecasts made one election at a time, adding engagement did not reduce the average error of just under four points from the previous result alone, and engagement on its own more than doubled it. The party with the most engagement won the most votes in only 12 of the 34 elections.
On Facebook, likes appeared to measure the size of parties’ audiences more than the movement of the electorate.
Caveats
The sample is 15 countries and 34 elections, and it covers party pages only. Engagement is attention, not support. It includes opponents and people abroad. The paper is a preprint and has not been peer reviewed.
The views in this post and the paper are my own.
