Areas of Expertise

Modeling

Who in your district can still be moved, who is a turnout problem, and where the votes you need are actually sitting.

What modeling answers

The three questions modeling answers

Every race has an arithmetic to it. A certain number of votes wins, a certain number of people turn out, and the distance between those two figures is the whole job. Modeling is how that arithmetic gets specific enough to act on.

The work starts with the public record: the voter file for your jurisdiction, registration and vote history, results precinct by precinct for the last three to five comparable elections, census geography, and whatever contact records your campaign has already collected. None of that is expensive to obtain. What costs money is the time to assemble it in the right order and the judgment to read it.

Three questions come out of it.

Persuasion. Which voters have not settled on a choice and can be reached at a cost your budget can carry.
Turnout. Which voters already agree with you and will not vote unless somebody gives them a reason to.
Geography. Which precincts hold both a favorable margin and unused turnout, and which are already giving you everything they have.

How the targets get built

How a target list is built

Set the vote goal. Turnout across the last three comparable cycles, the winning number in each, and a target that still holds if turnout comes in low.
Score the file. Every registered voter gets a likelihood of voting and, where the data supports one, a likelihood of being reachable on your side.
Rank the precincts. Vote share against turnout, so the places with margin and slack separate clearly from the places with neither.
Cut the universe. A named list with a count on it, a cost per contact, and a plan for how many times each person hears from you before election day.
Update it monthly. New registrations, early vote returns and your own contact history change the list, so the list changes.

Down ballot races are harder to model than statewide ones. Turnout is lower and moves more, almost no public numbers exist for a single school board district, and precinct returns are sometimes the only honest signal available. Federal races bring the opposite problem: plenty of published numbers, most of them about the state or the nation rather than about your district. In both cases the modeling leans on the file and the returns rather than on anyone's impression of the place.

Limits

The limits of a model

A score is an estimate of a probability, and it should be read that way in every meeting where it comes up.

A model inherits every error in the file it is built from, including bad addresses and voters who have moved away.
Scores describe groups well and individuals poorly. A voter scored at 70 is not 70 percent persuaded.
No model creates votes that are not in the district. If the arithmetic does not close, the honest finding is that it does not close, and the plan has to change.
Modeling says who to talk to. What to say to them comes from Insights.

Targeting that is never checked against what voters say turns into a list nobody questions. The two services are built to run together, and the monthly cycle in Research Method is where they meet.

Want the arithmetic of your own race?

Send the office you are running for, the election date and what you have set aside for research. We write back with a scope and a price.

Contact Upset Research