| A couple notes on FiveThirtyEight: First of all, yes, they got it wrong, in that they gave Clinton a higher chance of winning than Trump. They didn't give him a 0 chance, and a model that has 67%ish certainty should be "wrong" about 1 in 3 times. They were better than the other major forecasts, insofar as they gave Trump the highest chance of winning. I respect Nate Silver for his opposition to non-data-driven media narratives. In 2012, when the media was calling the race super close, Nate called them out for making a media circus when the race was actually looking really good for Obama. In this race, he did the opposite, constantly pointing out how likely a 30% chance was and how big of a polling error there could be (and also how many undecided voters there were). He also called out herding (the tendency of pollsters to not publish outliers) when he saw a ton Clinton+3 and Clinton+4 leads in the week before the election. His forecast may have been wrong, but his instincts are pretty good. Regarding the shy-Trump effect, we didn't see this play out in the primaries (Trump often underperformed his polls), which makes me skeptical. It might be that with all the reluctant republicans felt a need to vote for him in the general election, the effect got played up. Personally, I think there are 2 big problems in polling as we do it today: - Not enough pollsters are polling in Spanish. From the polling we do have, we know that Spanish speaking latinos vote very differently from English speaking latinos. - Our methods for finding people are a mess. Phone polls often use land lines (Cell numbers are often unlisted and may belong to people who live in other states despite their area code) which prejudice them towards the demographic of "people who still have/answer a land line". Internet polls suffer self-selection bias. We need a way to randomly sample people without them making the first move to join a poll. This problem needs to be solved. Lastly, I want to point out that even Trump's campaign didn't see this coming. No one (except Bill Mitchell's yard sign and Halloween mask model) saw this coming. We have a lot of work to do to fix polling. |
http://www.latimes.com/politics/la-na-pol-usc-latimes-poll-2...