Why Demographic Models Are Essential to Understanding and Reaching Key Audiences

Election results come down to two factors — which individuals will vote, and whom they will vote for — resulting in two parallel objectives for political advocacy groups and campaigns. 

First, these organizations want to convince people who are likely to prefer their candidate (or support their ballot measure) to cast a ballot. Demographic modeling, in combination with vote preference polling, estimates the characteristics of their supporters; organizations can then use this support model in combination with a turnout model to initiate door-knocking or phone-banking efforts to contact probable supporters who are nevertheless less likely to cast a vote. Without this turnout model, these groups waste resources reaching out to people who were already sure bets to vote.

The second objective: Persuading likely voters who are undecided about their vote choice to vote for the organization’s preferred candidate or ballot initiative. Here, outreach segments are selected from likely voters with uncertain political preferences, with fewer resources wasted on individuals who are unlikely to vote at all.

Pollsters also need estimates of individual turnout likelihood. If you poll a random sample of Americans about their preferred candidate and count each respondent equally, a person who is unlikely to actually cast a ballot counts the same as someone who surely will vote. Turnout predictions are critical to creating a representative sample of the electorate, rather than of adults who respond to polls.

Read on to discover how Civis Analytics models election turnout by deepening understanding of key populations — and how organizations outside of the political sphere can leverage our approach to hone a richer understanding of the populations that matter most to them. 

Civis kept it simple when building turnout models for the 2022 U.S. midterms. We knew if we could build models that did well at predicting turnout in prior years, the variables that went into those models would likely also be predictive for 2022’s results. If a variable was predictive one election but not the next, we threw it out. If a feature was consistently predictive, we looked for ways to refine it further. 

Key among these variables are Civis’s demographic models, especially race and education. These variables are strong predictors of voter turnout, but aren’t included on most voter files. Civis’s recent work to optimize these models was a key advantage in building 2022 turnout models.

The same strategy applies to all models Civis produces. We spend a great deal of time building and fine-tuning features that are most predictive for a given individual attribute. We have found that throwing the kitchen sink of data into a model can worsen model performance, and make investigating the reasons for the model’s predictions extremely opaque.

Model scores aren’t only useful as predictors of other individual features — they’re also helpful for generating cross-tabs (useful visualizations for comparing aggregates by different splits in demographics). For instance, we used our models to answer the question “What fraction of the electorate will be non-college-educated, white Gen-Xers?” Similar questions can be asked not only about likely voters, but also about individuals likely to be interested in a product or service, for example.

For the turnout model, our predictions were used by our clients to identify likely voters in order to persuade them to vote for a preferred candidate or initiative, and for get-out-the-vote initiatives focused on unlikely voters. Our models were also used in combination with vote choice polling to predict electoral outcomes in advance of the election.

For a much deeper dive into Civis’s 2022 midterm modeling strategy, please download my report Who Turned Out to Vote in 2022? A Post-Mortem of the Civis Midterm Voter Turnout Model. 

Civis’s demographic models are equally effective outside the campaign trail. Check out these two examples.