A political intelligence platform.
A political intelligence platform.
A political intelligence platform.
Methodology
Methodology
At the core of our platform is a machine learning ensemble that converts weekly survey data into stable small areas estimates. The result is a continuous picture of American public opinion at every county, district, and state, every week of the year.
At the core of our platform is a machine learning ensemble that converts weekly survey data into stable small areas estimates. The result is a continuous picture of American public opinion at every county, district, and state, every week of the year.
At the core of our platform is a machine learning ensemble that converts weekly survey data into stable small areas estimates. The result is a continuous picture of American public opinion at every county, district, and state, every week of the year.
At the core of our platform is a machine learning ensemble that converts weekly survey data into stable small areas estimates. The result is a continuous picture of American public opinion at every county, district, and state, every week of the year.
FIG 0.1
Collect
We survey voters every week, with hard quotas on race, age, and region so the sample mirrors the electorate of every place we cover.
FIG 0.2
Model
Our models predict the probability that a voter turns out and how they will vote for each of 146 million voter-profile-by-place combinations.
FIG 0.3
Smooth
A state-space model weights the full trajectory of responses (yesterday's, last week's, and the weeks prior) so real movement comes through clearly and daily noise washes out.
FIG 0.1
Collect
We survey voters every week, with hard quotas on race, age, and region so the sample mirrors the electorate of every place we cover.
FIG 0.2
Model
Our models predict the probability that a voter turns out and how they will vote for each of 146 million voter-profile-by-place combinations.
FIG 0.3
Smooth
A state-space model weights the full trajectory of responses (yesterday's, last week's, and the weeks prior) so real movement comes through clearly and daily noise washes out.
FIG 0.1
Collect
We survey voters every week, with hard quotas on race, age, and region so the sample mirrors the electorate of every place we cover.
FIG 0.2
Model
Our models predict the probability that a voter turns out and how they will vote for each of 146 million voter-profile-by-place combinations.
FIG 0.3
Smooth
A state-space model weights the full trajectory of responses (yesterday's, last week's, and the weeks prior) so real movement comes through clearly and daily noise washes out.
FIG 0.1
Collect
We survey voters every week, with hard quotas on race, age, and region so the sample mirrors the electorate of every place we cover.
FIG 0.2
Model
Our models predict the probability that a voter turns out and how they will vote for each of 146 million voter-profile-by-place combinations.
FIG 0.3
Smooth
A state-space model weights the full trajectory of responses (yesterday's, last week's, and the weeks prior) so real movement comes through clearly and daily noise washes out.
Methodology
How it works
The methodology is a seven-step pipeline that takes weekly survey responses in one end and produces reliable local estimates of public opinion out the other.
Methodology
How it works
The methodology is a seven-step pipeline that takes weekly survey responses in one end and produces reliable local estimates of public opinion out the other.
Methodology
How it works
The methodology is a seven-step pipeline that takes weekly survey responses in one end and produces reliable local estimates of public opinion out the other.
Segment the Electorate
We divide American voters into 13,824 profiles. Each profile represents a type of voter, not an individual. The framework spans 10,600 geographies, covering every county, congressional district, and state legislative seat in the country. That gives us roughly 146 million profile-by-place combinations the system tracks at once.
Collect
We survey 2,400 likely voters every week. Voters are reached through online panels, SMS, email, and our internal respondent panels. The sample is designed before collection begins, with hard quotas on race, age, and region, so it mirrors the likely voter population of every location in the country.
Model Turnout
For each of the 146 million profile-by-place combinations, our turnout model produces a probability that a voter of that type, in that place, will cast a ballot. The model uses individual demographics alongside the context of the place: historical turnout, past election results, and local socioeconomics.
Predict Opinion
A second model predicts what each profile is likely to say about a given race, candidate, or issue. It uses the same approach as the turnout model and the same inputs: age, party, race, education, income, religion, where they live, and the political character of that place. The output is a predicted opinion for every kind of voter in every place.
Trends
The model estimates an underlying latent opinion path over time. Daily or weekly survey estimates are treated as noisy observations of that path, with less certain observations given less weight. A random-walk-type structure smooths the trend, linking neighboring periods so information propagates across the series. The model also estimates many related opinion paths simultaneously. Because these series are connected, evidence from one can stabilize estimates in another, especially when data are sparse. This sharing reduces overreaction to isolated jumps while still letting each series move independently when data show a sustained shift.
Project
We combine each profile's predicted opinion, its turnout probability, and how many voters of that profile actually live in each place. The result is a single number for any geography we cover: county, congressional district, state house seat, state senate seat, statewide, or national.
Quantify Uncertainty
Vantage estimates come from a full modeling pipeline, so uncertainty is measured throughout rather than added after the fact. Sampling uncertainty comes from resampling the completed survey and rerunning predictions to see how much estimates vary. Turnout uncertainty is included by incorporating different plausible turnout scenarios into snapshot predictions. Time-smoothing adds a further layer, producing a range of plausible opinion paths rather than one trend line, with uncertainty growing when forecasting beyond the latest data
Segment the Electorate
We divide American voters into 13,824 profiles. Each profile represents a type of voter, not an individual. The framework spans 10,600 geographies, covering every county, congressional district, and state legislative seat in the country. That gives us roughly 146 million profile-by-place combinations the system tracks at once.
Collect
We survey 2,400 likely voters every week. Voters are reached through online panels, SMS, email, and our internal respondent panels. The sample is designed before collection begins, with hard quotas on race, age, and region, so it mirrors the likely voter population of every location in the country.
Model Turnout
For each of the 146 million profile-by-place combinations, our turnout model produces a probability that a voter of that type, in that place, will cast a ballot. The model uses individual demographics alongside the context of the place: historical turnout, past election results, and local socioeconomics.
Predict Opinion
A second model predicts what each profile is likely to say about a given race, candidate, or issue. It uses the same approach as the turnout model and the same inputs: age, party, race, education, income, religion, where they live, and the political character of that place. The output is a predicted opinion for every kind of voter in every place.
Smooth Over Time
A single day of survey data is noisy. Real movement and random wobble look the same. The model looks at the whole trajectory: yesterday’s data, last week’s, the week before. Real shifts come through clearly. Noise washes out. When opinion moves, you see it.
Project
We combine each profile’s predicted opinion, its turnout probability, and how many voters of that profile actually live in each place. The result is a single number for any geography we cover: county, congressional district, state house seat, state senate seat, statewide, or national.
Never Miss an Opinion
Every Election, Every Region, Every Week
Never Miss an Opinion
Every Election, Every Region, Every Week
Never Miss an Opinion
Every Election, Every Region, Every Week
Never Miss an Opinion
Every Election, Every Region, Every Week
Sign up for our Newsletter
Election Updates & Results
Select Top Line Poll Results
Future Product Updates
Sign up for our Newsletter
Election Updates & Results
Select Top Line Poll Results
Future Product Updates
Sign up for our Newsletter
Election Updates & Results
Select Top Line Poll Results
Future Product Updates
Sign up for our Newsletter
Election Updates & Results
Select Top Line Poll Results
Future Product Updates
Sign up for our Newsletter
Election Updates & Results
Select Top Line Poll Results
Future Product Updates
The first-ever membership-based polling platform. Revolutionizing public opinion data with unparalleled speed, accuracy, and affordability.
Account
Voter Panel
The first-ever membership-based polling platform. Revolutionizing public opinion data with unparalleled speed, accuracy, and affordability.
The first-ever membership-based polling platform. Revolutionizing public opinion data with unparalleled speed, accuracy, and affordability.
Voter Panel
The first-ever membership-based polling platform. Revolutionizing public opinion data with unparalleled speed, accuracy, and affordability.