Forecasting Elections with Dirty Data

Summary: Y'know how online polls are unscientific? You can science that shit up with a statistical technique affectionately called "Mr. P".

Article gives a famous example of a magazine in 1936 doing a massive poll of its readers and botching their prediction because the readers were a biased sample. (Duh.) But the authors took a massive poll on Xbox and, by adjusting the responses, predicted the election outcome accurately. They even nailed predictions for demographics (like retirement-age women) who are pretty much completely unrepresented on Xbox.