The APPOR (American Association for Public Opinion Research) investigated the 2016 U.S. election polling and found that national polls were generally correct and accurate by historical standards. However, the state polls showed a bigger than average error rate by historical standards. What accounted for the errors? One key factor was that state polls weighted on education less often. In 2016., there was a strong correlation between education and presidential votes in key states. Adjusting for over-representation of college graduates was critical, but many polls did not do it.
Random Probability Sample was the classic paradigm but with rapid adoption of cell phones and increasing non-response rates, the Modeled Sample has emerged as the alternative paradigm. This allows for non-response correction and leveraging auxiliary data to validate weighting targets.
Mark Blumenthal, Head of Election Polling at SurveyMonkey