Modeling/Scores
The scores section of VAN is where we include modeling that predicts an individual voters’ likelihood of supporting a given candidate, caring about a certain issue, or belonging to a particular demographic group.
What do they show?
Most of our models represent the likelihood of a voter meeting a set criteria. For instance, if a group of voters has a Democratic Support score of 50, we would expect that 50% of the voters in that group would support a Democratic candidate, 50% would not. Similarly, we would expect 70% of voters with a college graduate score of 70 to have a 4 year college degree.
How do you use them?
You can use scores to make your voter outreach more efficient. Say that you live in a district that is 50% Democratic, 50% Republican and your campaign has the volunteer capacity to talk to 100 voters during GOTV. If you pick 100 voters to talk to at random, only 50 of them are likely to be Democrats willing to vote for you, while the other 50 won’t.
If, on the other hand, you use a score cutoff of Democratic Support higher than 70, your volunteers will spend more than 70% of their time talking to Democratic voters during GOTV.
If you have the time and volunteer capacity to engage with every single voter in your district, you don’t need to worry about targeting in the same way. But the two big limits on campaigns are time, and resources. Scores help us maximize the use of both.
So someone with a Democratic Party Support score of 90 is more liberal than someone with 50?
No. These models are not a measure of strength or ideology, just likelihood. You would expect 90% of voters with a score of 90 to support Democrats, but they could all be incredibly moderate or even conservative Democrats. Similarly, while only 50% of voters with a score of 50 are expected to support Democrats, that group could be 50% left wing activists, 50% Tea Party Republicans.
These models are set up as binary predictors--what is the likelihood of someone meeting a binary designation (supports Democrats)--not their relative strength of ideology or belief.
How are they built?
Most of our models are built on training data from surveys or field canvassing, though a few models are aggregates of census and consumer data.
Respondents are matched to the voter file, consumer datasets, ebase historical election results, census tracts, party primary history, political donor databases, and any other field data we have about them. Machine learning algorithms identify feature relationships and extrapolate those respondents out to the entire voter file. Scores are then retrained regularly as new field and survey data comes into the pipeline. All DNC scores update nightly.
We source our voter modeling from a number of different places--modeling firms, campaign analytics departments, and in-house models built by our data/analytics team.
Can I trust political models?
Yes, but use them as a tool, not a sacrosanct law. Modeling can help improve voter contact efficiency and inform campaign decision making. However, your knowledge of your own district and voters is more important, and should be the main driver of your campaign’s strategic decisions.
We spend a lot of time validating and evaluating the support scores that are available to us across different types of campaigns (local vs. state house vs. senate) and geography to make sure the models we make available to campaigns will work well for you. We monitor this on an ongoing basis throughout the cycle, so if you’re entering your data in VAN you should be accounted for in your targets.
With that said, every campaign and candidate is different. Most modeling is built around the notion of a “generic Democrat,” but candidates and campaigns can over or underperform their modeling based on their connection with voters and constituencies. If something feels super off, don’t hesitate to reach out (we expected 70% of this list to support our campaign but only 30% did, any idea what happened?), but also don’t obsess over modeling too much. Politics is a diffuse and ever changing thing and we should be flexible and use heuristics from time to time as well.
2022 Scores
- 2022 DNC Dem Party Support Score - Predicts the likelihood that a voter supports Democrats over Republicans in head-to-head settings. 0-100
- 2022 DNC Malaise Score - Identifies voters who are most affected by the negative national mood towards Democrats by modeling individuals whose mood towards President Biden has decreased since 2020. Higher values indicate a voter has become more unfavorable towards Biden since 2020. 0-100
- 2022 DNC/Clarity National Turnout - Predicts the likelihood that an individual will vote in the 2022 General election. 0-100