New Municipal Turnout Score in VAN

04/21/2021
Hi folks,

We’ve added a new score to VAN today - 2021: TargetSmart General Election Turnout. This model represents a given voter’s likelihood of voting in an off year, municipal election.

Odd year elections tend to have lower turnout than Federal or Statewide elections, which makes targeting more difficult and important. We typically recommend using vote history for turnout targets, and that recommendation has not changed. However, we wanted to provide this score as an additional tool you can use on top of vote history targets.

What Do These Numbers Mean? 

Each voter has a score from 0 to 100. This number represents that voter's estimated likelihood of voting in an odd year general election. Of all voters with a score of 50, you would expect 50% to vote, and 50% to not vote.

How Do I Use Scores? 

Scores are designed to improve the efficiency of your outreach efforts. If someone already has a 100% chance of voting in a given election, you’re not going to net many votes by hitting them with a GOTV message. Similarly, in a low turnout election, you might not be able to convince every registered voter of the importance of a given election. In a perfect world, we’d have the time and resources to speak to every single voter, but given constraints we have to target our efforts.

This data team believes that vote history is the best indicator of future likelihood to vote and, philosophically, that campaigns should engage the broadest swath of voters. We tend to use turnout scores to increase the electorate (If someone has no vote history but a higher turnout score, we might add them to a universe they wouldn’t otherwise qualify for). In VAN terms, we would use it in “Add Step” instead of “Narrow List.”

How Was This Score Built? 

This turnout score represents two separate models: one for voters who have previous vote history and one for voters who don’t (new registrants). The first model is built almost entirely using vote history and registration information (elections voted in, number of years registered, etc.). The second model gives the most weight to registration information, and includes more demographic and consumer data.

I’m Getting My Masters in Machine Learning, How Was It Really Built? 

TargetSmart built several sub-models using a rules-based classifier to consider over 100 political, consumer, and demographic variables. The build process identified key variables, and the final model was built using a boosted decision tree classifier.

All the best,
The MDP Data Team