Definition
The Local Voter Score is a general indicator of propensity to vote in local (i.e. non-statewide) elections. We recommend using this score when the highest office contested on the ballot is not covered by any other existing turnout scores ( Municipal Turnout Score, County Turnout Score, and School Board Turnout Score).
Technical Details
In the summer of 2025, Murmuration released three turnout models specifically built for elections where municipal, county, and school board races are top-of-ticket. These scores cover approximately 70 million voters for municipal races, 27 million for county races, and 68 million for school board races. They are available in geographies with more than 5,000 registered voters where sufficient historical vote data exists, and are the recommended scores when applicable — each is tuned to the election type and geography most similar to your upcoming race.
There are situations, however, where none of these scores will have coverage:
- Election timing changes — e.g., a school board race that historically ran concurrent with midterm or presidential elections shifts to an off-cycle date
- Special elections — e.g., a city council vacancy triggers a January special election in a jurisdiction that normally holds elections on federal election dates
- Consolidated jurisdictions — e.g., two previously separate jurisdictions merge (such as a city-county consolidation), creating a new electoral unit without a clean historical analogue
- Newly incorporated geographies — e.g., an unincorporated area becomes a city; without prior municipal vote history, a score cannot be built with the same confidence
- Small or data-limited geographies — jurisdictions below the 5,000-voter threshold or where vote history data is unavailable
The Local Voter Score is designed to provide coverage in exactly these situations. It was built by taking a sample of registered voters in geographies with municipal, county, or school board vote history, and imputing what their turnout score would be for each of those election types — based on their federal and local vote history, registration information, and demographic data from the Atlas by Murmuration dataset.
Separate CatBoost Regressor models were trained for each of the three election types (municipal, county, and school board), each on a sample of 2 million registered voters. The three component scores are then averaged to produce a single score that serves as a proxy for an individual's propensity to vote in local elections generally.
Scores range from 0–100, where higher values indicate greater propensity to vote in local elections.
Models were evaluated on a holdout set comprising 73 elections held across 1,808 geographies — all elections with available vote history data held after the model training cutoff, excluding 2025 statewide off-year general elections. Among voters in the top 20% of the Local Voter Score, an estimated 65.8% voted in their most recent local election — roughly 195% more likely than the average voter, whose turnout rate was 22.3%. The population-weighted AUC across all holdout elections was 0.87, indicating strong discriminative ability between likely voters and non-voters.
Use Cases
The Local Voter Score is intended to support targeting for local elections where unique circumstances mean the upcoming race has no direct historical analogue. Unlike Murmuration's other turnout scores — which are trained on past elections of the same type to predict future turnout in similar elections — the Local Voter Score is not a direct prediction of an individual's likelihood of voting in a specific upcoming election. It should be understood as a general propensity score: a voter with a score of 70 is more likely to vote in local elections than a voter with a score of 50, but a score of 70 does not mean a 70% probability of voting in any particular local election — the actual probability may be higher or lower depending on the election context.
As a result, score thresholds should be applied with care. In particular, for elections with very low anticipated turnout, thresholds may need to be adjusted downward. The score is most useful when combined with issue support scores to sharpen targeting universes:
- Persuasion/Name Recognition: Target voters at the higher end of the score distribution — those most likely to turn out and complete their ballot. This is especially useful early in a campaign or in races where candidate name recognition is low, to ensure outreach reaches people likely to actually vote.
- GOTV: Target voters in the middle range — those who may be on the fence about participating and for whom a reminder or nudge is most likely to make a difference. GOTV targeting with this score should generally be combined with issue support scores or candidate affiliation to ensure outreach is reaching the right voters.
For example, suppose you are campaigning in support of a school district bond measure to fund new classroom construction. The election is being held off-cycle — separate from any statewide or federal election — in a district without prior vote history for standalone ballot measure elections, making the Local Voter Score the appropriate choice.
- Persuasion universe: high Local Voter Score and mid to high Education Voter Score
- The high Local Voter Score threshold ensures outreach is focused on those most likely to turn out in this off-cycle election. A slightly lower Education Voter Score threshold casts a wider net among voters who care about education issues, including those who may need persuasion on the specific merits of the bond measure.
- GOTV universe: midrange Local Voter Score and high Education Voter Score
- The middle Local Voter Score range targets voters who are on the fence about participating — those most likely to respond to a nudge or reminder. The higher Education Voter Score threshold ensures GOTV outreach is concentrated among voters with a demonstrated interest in education, who are most likely to support the measure if they turn out.
Targeting Table
The table below shows the score values associated with each decile to help you more easily target using the Local Voter Score nationally. Note: these score cutoffs are built off of turnout patterns in local elections from a national sample, so you can use these as a starting point, but should adjust the thresholds according to your desired universe size since they are not finely-tuned to elections in your particular geography.
| To target the top... | Set the minimum score value as... |
| 10% | 65 |
| 20% | 46 |
| 30% | 31 |
| 40% | 21 |
| 50% | 14 |
| 60% | 9 |
| 70% | 5 |
| 80% | 3 |
| 90% | 1 |
| 100% | 0 |