Voting with Facebook Likes
As campaigns this year gear up their marketing efforts on all the social platforms, it begs the question, how can campaigns measure the success of their efforts on each of these platforms and translate that success to the state of the race?
Let’s take Facebook as an example. On Facebook, a campaign is limited to a few metrics to track performance. These metrics include total “Likes”, average post likes, average post shares, and total number of people talking about the page. Marketing efforts are best measured by looking at the reach of each post, but it seems that the campaign as a whole tends to race towards getting as many “Likes” as possible. For example, campaigns frequently post about milestones they’ve reached for Facebook likes and promote it as a metric for success for the performance of the campaign. There’s certainly nothing wrong with this. Campaigns should be doing everything they can to increase their reach across their network of constituents by getting more of them to like or follow their page.
However, does it actually translate as a predictor for a winning campaign? We decided to embark on an experiment to find out the effect of voting with Facebook likes.
For our initial experiment we decided to focus on races in the 2012 election cycle at the national level for U.S. Senate and House seats, and at the local level with Gubernatorial races. We didn’t focus on smaller races because the Facebook data tended to be sparse. We also couldn’t analyze races farther back than 2012 because the time series data through Facebook only goes back so far. Initially we gathered information on approximately 106 races for our sample. After eliminating races where Facebook data was sparse or non existent, we were left with 76 cases for our analysis. We also excluded cases where Independent seats were the incumbents, any new seats that opened up for that election (this would be caused by redistricting most likely on the House side), and any cases where less than 100 likes were found on someone’s Facebook page.
Next, we wanted to isolate those who won their race in 2012 and also had the most Facebook likes as the group we were trying to predict. We assume that Facebook likes don’t translate to wins explicitly and that there are other factors or dimensions within races that might also be good predictors. The predictors we decided to test in this experiment were Race Type, Incumbent Status, and spread of Facebook likes between the competing campaigns.
Race Type: Senate, House, or Governor
Race Type indicates the type of race at the national or local level. We wanted to test the different races to see if one type was more predictable than another.
Incumbent Status: Democrat, Democrat OPEN, Republican, Republican OPEN
Incumbent status indicates whether a current party has a seat or if they are leaving the seat because of term limits. Our reason for testing incumbent status was that it would give us another dimension around estimating the impact of an established incumbency or the impact of fresh new candidates running in an open seat race.
Facebook spread is the numeric difference of likes between competing campaigns. Our theory here was that maybe closer numbers of Facebook likes would be more likely to be inconclusive for prediction purposes.
After testing the variables mentioned previously, we found none of them to be significant predictors of winning. So what does that leave was with? Well although we might not have any good predictors for a winning campaign based on Facebook likes, mathematically we’d still estimate that a campaign leading in Facebook likes would have approximately a 63.2% chance of winning the election. With better and more extensive data we estimate the percent change of winning to be closer to 70%.