Facebook’s crystal ball forecasts Republican Senate in 2015

With eight weeks left before the 2014 midterm elections, our Facebook Campaign Forecasting Model predicts that Republicans will pickup five Senate seats in November. Three of these are from toss-up races we are presently tracking – see www.hashtagdemocracy.com for more details.

The Facebook model forecasts Republican candidates winning very close elections in Arkansas and Iowa. Our weekly tracking shows, however, that both races had tightened over the last four weeks.  In the open West Virginia Senate contest, Republican Shelly Moore Capito remains comfortably ahead of her Democratic opponent. Two other contests — in South Dakota and Montana – round out the five Republican pickups.

{mosads}Our model also finds Democratic incumbents maintaining steady leads in Alaska, Colorado, New Hampshire and North Carolina, while the lead of Republican incumbents in Georgia and Kentucky has slipped in the last month. Five of these states are rated as toss-ups by other forecasters with New Hampshire the lone state thought to be leaning Democratic. Senate campaigns in Minnesota and Michigan are now too close to call and, because of its so-called jungle primary, the Louisiana Senate campaign is not tracked by our model.

The Facebook forecasting model is based on the theoretical assumption that online engagement is a real-time measure of campaign effectiveness and voter engagement with candidates. In 2012, our Facebook forecasting model accurately predicted the winner in eight of the nine Senate toss-up races studied. Moreover, when the model’s Senate predictions over the last nine weeks of the 2012 election were compared to 212 polls published during that period, the model did a better job of projecting actual election returns in five of the weeks studied.

For this cycle, starting in September of 2013, we began carefully capturing and analyzing the Facebook “likes” and the “PTAT” of candidates running for U.S. Senate. Variations in Facebook “likes” and “PTAT” were used to build three variables that estimate campaign effectiveness. First, we measured the growth of a candidate’s fan base—or the number of people who “like” a candidate’s fan page over time. Second, the growth or contraction of the number of people Facebook reports are “engaged” with the campaign was calculated. In Facebook parlance, “engagement” is those who comment, like or share posts about the candidate with their Facebook friends. It is called ‘PTAT.” And third, the campaign’s potential to mobilize people at a particular time is estimated as a simple equation of those engaged with a campaign at a particular moment in tim divided by a candidate’s fan base.

These variables are calculated regularly, weighted equally and combined with estimations of electoral fundamentals, such as partisan voting index (PVI) reported by the Cook Report and incumbency, to produce a weekly, rolling forecast of the two-way vote percentage for each race studied. Forecasts for tracked races are presented as two-week and three-week, rolling averages and can be found at www.hashtagdemocracy.com. From our work in 2012, we found that averaging predictions over time generated more accurate forecasts with a higher R-Square.

How accurate will the model’s forecasts be in predicting Senate winners this year? While we have confidence in the model and in our ability to track Facebook metrics accurately, the real proof will be in the pudding that comes out of the Election Day blast chiller on the night of November 4. The success of our model in forecasting 2012 Senate outcomes may have been an anomaly. But if our model logs a repeat performance in 2014, we may have discovered another interesting and useful application of the plethora of data now available in our online world.

MacWilliams is a Ph.D. candidate at UMass, Amherst. He is the founder and president of MacWilliams Sanders Erikson, a strategic communications firm. Erikson is a Ph.D. candidate at UMass, Amherst and a partner at MSE.  Berns is a former research assistant at UMass, Amherst and works for Strategy Group, a direct mail firm.