Not only did the polls overstate the Democrat's chances, they did so by more than the reported error. Sam Wang uses only polls. He warned that there are many types of error that go unreported and that it was possible for the error to favor a Republican sweep. Given his dependence on polls, it is not surprising that he was the furthest off. Best of the commenters, by far, was the Washington Post. It was a very low standard.
This is a bit of what I was getting at in the other thread. Was the correct call 100% GOP, and thus the closer to 100% GOP your call was the better it was? Or, because the polls ended up being so wrong, was Sam Wang's call of 60% +/- 15% actually the best because it worked uncertainty into the prediction?
I intended to. The point of the quote was that you cannot treat it as a outlier. In practice you should be very cautious about apparent outliers, because they sometimes are valid. You can disrupt an entire data set that way.
There are two things here that I think are getting confused. First, there is a firm, purely mathematical definition of an outlier that is based on the standard deviation of a data series compared to the potential outlying point. I'll admit I didn't do the calculation because I don't have my high school stats book with me, but since the data range in this race was around a tie with a fairly narrow deviation, it looked like it would fit. It's definitely harder to identify them, though, because the variance in the polls this time around was really high and there were fewer than in presidential election years.
Second, what we would need to dive into to determine which poll was more accurate, regardless of outlier status, would be the cross-tabs. Did the Walker +7 poll get the demographic percentages right for their sample? Or at least closer than the rest of the polls clustered around a tie? Or did it have the same voter breakdown but the people who answered, by a random fluke, happen to line up their preferences with election day? I haven't done that either because I have enough work to do and I figure one of the stat gurus will do it in the coming weeks.
Yea the polls were quite off. Hagan for instance was supposed to have been leading, even leading up to the election. Of course not the result that manifested itself
Hagan's lead was always pretty narrow, though, usually 1-2 points for her over Tillis. From other races, we know that at least a couple 1-2 point leads go down on election day, it just so happened it was Hagan this time around. In 2010, the same thing happened in the Angle-Reid race, Reid was supposedly losing by a few points but pulled out a win on election night. There are dozens of examples.
Democrats have benefited from that before, this time around it clobbered them. Along with the midterm electorate, an unfavorable map, an unfocused and quite likely mismanaged set of campaigns, gaffes, etc. it was one of those perfect storm metaphors.
Is there anyone here from VA who can explain why it was so close there? Gillespie wasn't supposed to have any chance, yet he came within 15k of winning.
If you compare Warner's gubernatorial and first senate race with this map, you will see the 2014 map looks a lot more like the 2008 and 2012 presidential election maps. His older maps indicate he had more support in rural counties which he lost this time around.
I don't know if there is a fully comprehensive, exit-poll-backed theory as to why this happened. But it looks like a combination of low turnout, a heavily-nationalized race, and the local polarization where people were voting down party lines. If anything this might end up being an ironic bright spot for the Democrats: when you look at the last gubernatorial race and this one, both went a hair for them under mostly unfavorable circumstances.