As a service to journalists, Facebook sent around data on how the 2016 candidates have done at "engagement" over the last month -- how many millions of people have done an engage to a politician's Facebook, and how many engages those people done did.
Here are the numbers.
"Oh ho!" you say. "Look at all that engagement!" And maybe: "Clinton is number one, and she is winning and therefore there is a correlation here! Also, Trump!"
Or maybe you're like, "OK, what does this tell us?"
That is what I am like (as the kids say).
We can compare the data from Facebook to how the candidates have fared in polling over the same time period, which will maybe give us a better sense of the utility of these numbers. We can plot how much the poll numbers have changed versus how much better or worse each candidates' engages are than everyone else's.
Giving us this.
There appears to be a relationship between the number of people talking about a candidate over the last month and how that candidate has fared! Maybe, then, talking about a candidate on Facebook is linked to polling increase! Now, this makes sense, generally, as people get excited about candidates and want to bother their friends and family with their newfound crushes. ("Sam Johnson is in a relationship with Ben Carson.")
If we look at the correlation between the candidates's Facebook engagement and their polling change, it's actually fairly strong. There's a 0.61 r-squared between the number of interactions and the change in the polls, suggesting a link between the two. (Editor's note: Philip is a nerd.) Maybe Facebook can be predictive!
But something else happened over the last month: Joe Biden announced that he wouldn't run for president. That meant that his decent-sized base of support went elsewhere, mostly to Hillary Clinton and Bernie Sanders who, as the only two big-name Democrats in the race, also got a lot of interactions.
If we take the Democrats out of the mix, the correlation between interactions and poll numbers collapses. That decent 0.61 r-squared is now a 0.22 -- and the closer to 1 you get, the stronger the correlation. You can see the problem above. Jeb Bush and Carly Fiorina saw big polling drops but didn't have terrible interaction numbers. So maybe Facebook isn't predictive after all.
We can even loop in another metric: Interactions per person. Maybe if people are inclined to interact over and over, a candidate has more support.
Nope. Pre- and post-excluding Democrats, the numbers here are about the same as for interactions on the whole. Ted Cruz and Ben Carson have more engaged fans, but that's all we learn.
So why does Facebook send these numbers around, if they don't really tell us anything about the candidates? Because Facebook is in the business of getting us to use Facebook, not in the business of predicting political campaigns. They just want media outlets to cover the Facebook numbers, writing about how ...