The Facebook Forecasting Model (FFM) is founded on the assumption that past election results and incumbency are fundamentals that play an important role in shaping electoral outcomes. It adds to this foundation a participation variable (quantified through social media statistics generated by Facebook) that captures the effectiveness of each campaign’s effort to enlist and engage voters, as well as the potential of a campaign to mobilize voters on Election Day.

The model is specified as follows:

            Senate Vote = f(partisan voting index+incumbency+participation advantage)

Senate Vote is the forecasted percent of the two-party vote won by either major party candidate. It is a function of partisan vote index (PVI), which measures past election results, incumbency, and the estimated, candidate participation advantage generated from Facebook metrics.

PVI is regularly estimated by the Cook Political Report and has been used to undergird other election forecasts. PVI captures the increasing polarization that presents statistical challenges to presidential forecasting models, but washes out the fundamental advantages enjoyed by some incumbent Senators.

Incumbency is added to the Facebook Forecasting Model to correct the shortcomings of PVI.  In the model, incumbency advantage or disadvantage is determined by calculating how an incumbent performed, compared to the reported PVI, in the last contested election.

The third and final variable in the model is participation. The participation advantage variable is theorized as a real-time measurement of each campaign’s effectiveness in enlisting and engaging Facebook users, as well as the campaign’s potential to mobilize its vote on Election Day. This variable is produced by measuring and analyzing the growth of candidate Facebook “likes” and variations of candidate “PTAT” over time as well as estimating from these measurements each campaign’s potential to mobilize voters. These measurements are designed to capture and quantify Facebook effects identified and studied by political scientists since the 2006 elections.

In the Facebook Forecasting Model, social media measurements of campaign effectiveness are quantified and combined each week to produce the model’s dynamic participation advantage variable (PA).

In 2012, weekly model forecasts in seven contested Senate elections were compared to poll-of-poll averages from 212 polls conducted during the last eight weeks of the election. In five of the eight weeks studied, the Facebook Forecasting Model more accurately predicted the percent of the vote won by major party candidates than the poll-of-poll averages as measured by regressing poll-of-poll average predictions and FFM predictions against the final, two-party candidate results.

Based on the performance of the model in predicting 2012 Senate campaign outcomes, 2014 forecasts report two week and three week averages.