We identify three general data sources for forecasting election outcomes: historical, political opinion polls, and prediction markets, as described in a recording by Tom Miller, Editor-in-Chief of Data Science Quarterly:

Accurate election forecasting has been a forte of Data Science Quarterly. Our use of prediction markets and prediction surveys sets us apart from election forecasters who depend on preference surveys (opinion polls).

See reported election forecasts in social media postings on Twitter .

In the 2020 presidential election, we utilized prediction markets and statistical simulations of electoral college voting. Our prediction for the 2020 presidential election of November 2020 was much closer to the actual electoral college outcome than the predictions of other organizations, as shown in this figure and the associated recording:

For the 2020 presidential election, we were able to correctly predict all Electoral College markets except Georgia, as shown in this horizon chart for selected states:

See a tutorial on political horizon charts under Political Horizon Charts.

We returned to Georgia for the senatorial elections in January 2021. Concerned about what we perceived as a slight Republican bias in prediction markets, we introduced the prediction survey, a new form of political survey. We traced preference survey (opinion poll), prediction survey, and prediction market measures in the weeks leading up to the January election, as illustarted in this horizon chart for the two senatorial contests:

Relying on prediction surveys, we correctly predicted results for both Georgia senatorial elections. Our methods gained recognition in Fortune.

Our current thinking regarding preference surveys (opinion polls), prediction surveys, and prediction markets is summarized in this figure:

Return to Research