The Facebook Data Science Team conducted a detailed analysis of candidates for governor, the House of Representatives and the Senate in next month’s midterm elections, examining some 150,000 posts that have generated around 20 million likes, comments and shares.
Solomon Messing, Annie Franco, Arjun Wilkins, Dustin Cable and Matthew Warshauer presented the findings in a note on the Facebook Data Science page, introducing their efforts as follows:
The public posts and comments generated during the campaign offer a new lens to understand candidates’ “campaign style” and key topics in campaign communication in social media. There are a few caveats, such as the level of activity of candidates on their Facebook pages and selection bias for those who use Facebook and visit frequently. However, looking at these statements and associated data across a number of dimensions offers insight into the kind of issues and ideas candidates are trying to communicate to constituents, the types of people who respond and, ultimately, which issues may influence the election.
In terms of campaign style, Democrats focused more on mobilization and fundraising, while Republicans tended to talk about policy-making and endorsements. The most common political issue candidates discussed on Facebook was the economy, while Democrats and Republicans differed in the amount they discussed issues such as immigration, healthcare and job creation.
Findings by the Facebook Data Science team follow.
In terms of frequency of posts, challengers averaged 221, compared with 126 by incumbents, but closer House and Senate races yielded more posts.
On campaign style, they wrote:
We first present our topic model, which estimates topics based on words that frequently occur together in candidate page posts. Our analysis shows that the majority of posts are about campaign events or contain rhetoric related to campaigning — words like “unite,” “believe,” “change,” “work,” “Washington” and “representative” appear prominently in the latter topic.
When it comes to “campaign style” on Facebook, Democrats and Republicans generally look similar. However, Republicans more often shared endorsements and talked about policy-making, while Democrats posted more about campaign events and voter mobilization. Democrats shared more external content in the aggregate, but this was likely driven by a few campaigns, and indeed, on average, Republicans shared more per campaign (not shown). Both sides engaged in campaign rhetoric and posted about the ALS Ice Bucket Challenge at similar frequencies.
In analyzing political issues, the Facebook Data Science Team developed a comprehensive list, tracked by major polling firms and the Wesleyan Media Project and its keyword category analysis found that:
Posts about the economy and economic mobility dominated the discourse, with foreign policy, money in politics, fiscal issues, healthcare, education, immigration and women’s issues appearing in a non-trivial proportion of posts.
Splitting up the data by party, Democrats trended toward economic mobility, education and women’s issues, while dominant topics for Republicans were foreign policy and firearm regulation.
Focusing on the specific issue of immigration, the Facebook Data Science Team found that:
Although both parties discuss immigration, there are important regional patterns — candidates were more likely to discuss the issue of immigration, for example, in the Southwest. This map shows the proportion of candidate page posts about immigration in each state. The colors range from zero in the lightest states to more than 10 percent in the darkest states — 12 percent in Louisiana and 10 percent in Arizona.
Finally, in studying Facebook users who respond to candidates’ posts on the social network, the Facebook Data Science Team found that men comment on more posts, women like more posts, and users older than 35 were more likely to comment or like posts.
And digging further into gender, they wrote:
We also wondered if women were more likely to comment on or like female candidates’ Facebook pages. They are not: The percent of engagement coming from male versus female users is nearly identical for male and female candidates.
Readers: Did any of the findings by the Facebook Data Science Team surprise you?
Collage of election images courtesy of Shutterstock.