Polling

How we improve predictive accuracy

Our techniques allow us to identify the likely voting behavior of "undecideds". We sample millions of voters. We do so over long-periods of time to reduce headline and media cycle fluctuations.

unumAI predicted the most competitive U.S. Senate and U.S. House races twice as accurately (winner) than survey polling leading up to the 2020 election. In U.S. Senate races, we were correct 93% of the time compared with 58% for public, survey polls.

unumAI predicted the right winner in competitive U.S. House races around the country twice as accurately as surveys, and 4.7 times more accurate in suburban U.S House races.

Political Case Studies

Report	Release Date
Phil Weiser 2018, Jason Crow 2020, and Colorado Climate Change – December 2020 How unumAI reimagines political messaging and engagement with the public.	December 12, 2020

Election Predictions

Report	Release Date
2020 Election Recap – November 2020 Analytical recap of election prediction results for competitive races unumAI made predictions on this year.	November 21, 2020
2020 Selected Races – October 2020 Newsletter report of October prediction results for competitive 2020 US elections, one month before election.	October 8, 2020
2020 Selected Races – June 2020 Newsletter report of unumAI June prediction results for competitive 2020 US general and primary elections.	June 26, 2020
2020 Colorado Senate – May 2020 Example of unumAI’s full recommended messaging report covering Colorado’s 2020 Senate election.	May 30, 2020
2020 Selected Senate Races – April 2020 Newsletter report of unumAI April prediction results as compared to surveys for competitive 2020 US elections.	May 1, 2020

Our Philosophy

Unum is Latin for “one” and is taken from the original United States’ motto, “E Pluribus Unum,” which translates to, “Out of many, one.” Aristotle attributes this idea of democracy to Pythagoras who says that the bonds of society, government, and politics are based on those of family and friendship.

Technology

For better or worse, technology changes the ways that we interact with one another. Rather than ask our neighbors if taxation without representation is just, we search online for answers to our questions. Those search queries are logged and accessible in real-time; they are the greatest dataset of human political psychology ever created.

Nobody lies to Google, but we are susceptible to lying to ourselves, to our friends and family, and to pollsters.

People lie to pollsters because they believe that their political views are private or, they don’t feel comfortable telling a stranger which candidate represents their moralistic world-view for fear of being judged.

These deceptions show up in undecided responses in survey polling. Our methodology removes “undecideds” from the equation by probabilistically predicting their voting-day intentions and behaviors by what they search for.

Political Psychology

Our philosophy is that tribalistic competition frames political discourse, debate, democratic politics, and government. Awareness of the evolutionary biological/social psychological motivations for people to behave politically yields empathy towards everyone, allows blame and malice to be placed behind us, values and validates every perspective as historically and presently reasonable and relevant, opens room for debate once again, and allows for us to strive towards a more perfect union.

Results

We rely on extensive back-testing to validate our methodology. See how unumAI’s election predictions compare to traditional polls.

General Elections

unumAI was 1.6 times better than survey polling in predicting the winner and was 2.95 times better than survey polling in predicting the margin between the top-2 finishers when compared to traditional polls across 2020 battleground U.S. Senate general election races.

unumAI was 2.1 times better than survey polling in predicting the winner and was 1.23 times worse than survey polling in predicting the margin between the top-2 finishers when compared to traditional polls across 2020 battleground U.S. House general election races.

For 2020 battleground House elections in suburban areas, unumAI was 4.69 times better than survey polling in predicting the winner and was 1.05 times better than survey polling in predicting the margin between the top-2 finishers when compared to traditional polls. For 2020 battleground House elections in urban areas, unumAI was 1.24 times better than survey polling in predicting the winner and was 1.05 times worse than survey polling in predicting the margin between the top-2 finishers when compared to traditional polls.