> PredictWise has collected hundreds of thousands of respondents reporting party identification and models voters through survey, mobile data, and location data compiled through mobile devices
> PredictWise’s database includes verified data on 50 Million+ unregistered voters, sourced through credit files and portal registration data (e.g. Crate&Barrel), and is carefully purged of duplicates or registered voters, updating daily. PredictWise (PW) machine learning models predict the party of unregistered based on survey respondents from hundreds of thousands of Americans and can apply support scores to turnout your electorate.
> We make full use of the probability models on the file, ensuring that each individual will get a score reflecting his or her exact position. The end result is a continuous score for each cluster ranging from 0 to 100 for every American represented on the file.
So, they have a few hundred thousand survey fills/app installs. Then they combine the data against 50 million public records, and then they extrapolate it out for the entire voter population. So they may have one or two given devices in an area and then their model says what that represents across everyone.
> PredictWise has collected hundreds of thousands of respondents reporting party identification and models voters through survey, mobile data, and location data compiled through mobile devices
> PredictWise’s database includes verified data on 50 Million+ unregistered voters, sourced through credit files and portal registration data (e.g. Crate&Barrel), and is carefully purged of duplicates or registered voters, updating daily. PredictWise (PW) machine learning models predict the party of unregistered based on survey respondents from hundreds of thousands of Americans and can apply support scores to turnout your electorate.
> We make full use of the probability models on the file, ensuring that each individual will get a score reflecting his or her exact position. The end result is a continuous score for each cluster ranging from 0 to 100 for every American represented on the file.
So, they have a few hundred thousand survey fills/app installs. Then they combine the data against 50 million public records, and then they extrapolate it out for the entire voter population. So they may have one or two given devices in an area and then their model says what that represents across everyone.