WASHINGTON, August 11, 2020 - A new election prediction model will use Twitter data, which measures the pulse of public conversations about campaign politics, to improve predictions of a model using polling and other fundamental data. Researchers at the George Washington University Graduate School of Political Management (GSPM) will apply the novel technique to forecasts of 12 battleground states and publish projection updates every two weeks on Tuesdays until the election on Nov. 3.
The first edition of the PEORIA Presidential Prediction Project, released today, projects that Arizona, Colorado, Florida, Michigan, Nevada, North Carolina, Pennsylvania and Wisconsin will go to Democratic candidate Joe Biden, creating a nearly insurmountable lead (334 electoral votes) in the electoral college even though Iowa is projected for the Republican candidate, President Donald Trump (adding 6 for a total of 132 electoral votes), and Georgia, Ohio and Texas are listed as toss-up states (totaling 72 electoral votes).
“Trump has changed the way in which social media, but particularly Twitter, has influenced politics. We are interested if his use of Twitter can be quantified and used to better explain polling and election results,” research scientist Meagan O’Neill said. “Throughout the election cycle, we will be using Twitter data to identify events that reflect changes in support.”
“Political tweeters are opinion leaders,” said Associate Professor Michael Cornfield, “and who and what they are talking about provides clues about what’s in voters’ minds.”
The GW model, developed by O’Neill, Cornfield, and GSPM professors Todd Belt and Lara Brown, predicts incumbent (Trump) vote share in 12 battleground states: Arizona, Colorado, Florida, Georgia, Iowa, Michigan, Nevada, North Carolina, Ohio, Pennsylvania, Texas and Wisconsin. The remaining 38 states and D.C. are divided into groups of “base” states for each political party (e.g. Massachusetts for Democrats and Wyoming for Republicans). The base states and projected incumbent vote shares for the battleground states are calculated to produce a total projected electoral vote.
The new battleground state model is part of the Presidential Prediction Project, under GSPM’s Public Echoes of Rhetoric in America (PEORIA) Project, which strives to quantify how voters react to campaign messages and events. Model projections will be published periodically throughout the 2020 campaign. It follows a similar GW effort during the Democratic presidential primary elections in early 2020.
Each state prediction incorporates state-level polling data, information on the state’s partisan lean, data on the change in sentiment toward the incumbent on Twitter, state-level changes in unemployment since January and the national net favorability ratings for the candidates. More information about the new project, including a complete explanation of its methodology, is available on the PEORIA Project website.
The PEORIA Project research joins a prestigious cohort of GW research related to campaigns and elections. GSPM currently partners with GW’s School of Media and Public Affairs and Department of Political Science to conduct the GW Politics Poll, a measure of public opinion on political and policy questions. GSPM formerly housed the GW Battleground Poll, a nationally recognized series of surveys conducted by a bipartisan set of pollsters, which ended in 2018.