Definition
The National Swing Voter Score predicts how likely a person is to have voted for presidential candidates from both major parties since 2000.
Technical Details
The model was built to target labels generated from the following two survey questions:
Have you voted for a Republican candidate for U.S. president since 2000 (e.g. Bush, McCain, Romney, or Trump)?
- Yes
- No
- Don’t know or I have never voted for president
and
Have you voted for a Democratic candidate for U.S. president since 2000 (e.g. Gore, Kerry, Obama, Clinton, Biden, or Harris)?
- Yes
- No
- Don’t know or I have never voted for president
Respondents were labeled according to the following rules:
- Respondents who answered “Yes” to both questions were considered to be swing voters.
- Respondents who answered “Yes” to one question and “No” to another were considered to not be swing voters.
- Respondents who answered “don’t know” to either question were excluded from the dataset.
- Respondents whose later responses were inconsistent with these 2 questions (for example, those who reported never having voted for a Democrat, but in response to a subsequent question reported having voted for Kamala Harris in the 2024 presidential general election), were excluded from the dataset.
The National Swing Voter Score was trained on polling data from the Summer 2025 Murmuration poll. We collected 7,721 responses online from registered voters in July and August 2025. The survey used stratified random sampling to ensure national representativeness across age, race, gender, partisanship, education, and geography.
Scores range from 0-100, where higher scores indicate greater likelihood that an individual voted for both Republican and Democratic candidates in U.S. presidential elections since the year 2000.
The model was trained using gradient-boosted decision trees. Model features were drawn from the Atlas by Murmuration dataset, which includes demographic (age, race, gender, etc.), commercial, geographic, and vote history information for all registered voters nationally.
We validated the model’s accuracy using a held-out set of 1,352 polling respondents (20% of the original survey sample amongst 6,721 qualified respondents), and whose data was not used during model development. The model achieved an AUC on this test set of 0.633, indicating useful discriminative ability in identifying national swing voters. Further, among individuals with scores in the top 20% of the National Swing Voter Score, 38% were actual swing voters in presidential elections, which makes this group 47% more likely to be swing voters than the baseline proportion across the whole held-out set. Of individuals in the bottom 20% of the National Swing Voter Score, only 12% were truly swing voters, making this group 54% less likely to be swing voters than the baseline proportion across the whole held-out set.
Use Cases
Partners can use the National Swing Voter Score to identify and mobilize voters who may not be committed to party-line voting in national elections. These voters may be more open to being persuaded by outreach efforts on a particular topic than a committed partisan might be. Layering the National Swing Voter Score with the Partisan Score will allow targeting of nominal Democrats or Republicans who may be more likely to vote across party lines in a national election if the candidate appeals to them or a major issue at stake is of particular importance to them.
- Identifying likely Republican issue-based ‘defectors’: Let’s say a partner is working in a (federal) Gubernatorial race in a swing-state, where the Republican candidate has a long record of opposition to reproductive rights, and the Democratic candidate has been a strong supporter of reproductive rights. They create a segment of voters statewide with low Partisan Scores (20 and below) who are also most likely to be swing voters (scores of 30 and above), and who are also supportive of reproductive rights (Reproductive Rights Support Scores of 60 and above).
- Identifying ‘strategic primary voters’ in open primary states: Let’s say a Democratic primary is uncontested for a race a general election a partner is working on, and they strongly prefer the more moderate Republican candidate win the primary. They could segment on voters who have voted in past Democratic primary elections, but who also have high National Swing Voter Scores (35 and above), with information about the upcoming election and Republican candidates, as well information about the primary rules in their locality.
- GOTV targeting of infrequent swing-voters: A partner running a Get Out The Vote (GOTV) program in a presidential battleground state wants to identify voters who could swing the election but historically don't always turn out. They create a segment with high National Swing Voter scores (35 and above), low-to-mid turnout scores (30-70) identifying the prime GOTV targets.
Targeting Table
The table below shows the score values associated with each decile to help you more easily target using the National Swing Voter Score nationally. Note: these score cutoffs may be different in your local districts.
| To target the top... | Set the minimum score value as... |
| 10% | 37 |
| 20% | 33 |
| 30% | 30 |
| 40% | 28 |
| 50% | 26 |
| 60% | 24 |
| 70% | 21 |
| 80% | 19 |
| 90% | 16 |
| 100% | 5 |