has all kinds of different data information on these specific individuals. their vote history, likely candidate preference. all kinds of things that you can look at. what you get then from the county auditor or from the secretary of state are individual level returns. so you know exactly who voted early, and many times you know how they voted early, either in person or by mail. then you can imagine that back to your existing database or voter file. >> and you know enough about those people you sort of profiled the them whether it's through their age, their education level, their ethnicity to be pretty confident who they're actually voting for based on the models you have. >> no, exactly right. you look at two separate data points. the first is what we call a preference candidate model. every voter will have a score from 100-1. out of 100 mitch stewarts, barack obama would get 1 vote.