From what I’ve read about past campaigns, it’s not so much that campaigns have better data than the pollsters, so much as they are better at getting a specific question answered. Clinton isn’t ignoring certain states because her campaign thinks those states are safe, it’s more likely she’s ignoring those states because her campaign feels she can move the needle more in other states. A state like Michigan is going to be very polarized, with a large minority population and large working class white population. I don’t think there’s much Clinton can do to change the dynamics there. Seems to me that the key swing demographic is college educated whites. Most of those voters usually vote Republican but are leaning towards Clinton. That’s where the winnable voters are, so that’s where she’s focusing her efforts.
This is wrong. They have better data. *Much *better data.
This is for a couple of reasons, but mainly it’s money. They can pay to make sure the pollster makes 100 attempts to reach each person randomly selected, to avoid biases from who chooses to answer the phone. They can pay for the algorithm that determines whether you’re a likely voter based on what you bought from Target last weekend. And much, much more.
My apologies in advance if someone already linked this somewhere but I found this to be an interesting read from someone in the know: David Plouffe, former Obama insider.
I admit that I am panicky and prone to bouts of cynicism to the extreme, but this read seemed cogent and reassuring. And he’s not trying to be overly-sunny – either. Plouffe does say that she needs a good debate showing. But he’s optimistic that Trump has a monumental job of trying to win even one state among PA, FL, and CO.
We shall see I reckon, but fingers are crossed.
That’s targeting, which is what I was talking about. What I was disputing is whether their state polling is any more accurate than say, Nate Silver’s aggregations.
It is. Re-read my post. One big reason for the better polling is that they can afford the money is takes to get a more random sample.
Also, targeting and polling aren’t independent. The more you know about the population, the easier it is to build a representative sample.
Agreed 100%.
Yeah, Adaher, it’s not just questions, it really is the data – the data is gold. Karl Rove comes to mind – the 2000 and 2004 version, not the 2012 version that hung around past his prime like Willie Mays in the OF and got owned and shut down by Fox News anchors for trying to insist that Mitt was gonna win Ohio and other states long after it was already clear he wasn’t. The Obama team was equally up to the task of number crunching – maybe the best we’ve observed.
This is why having a real campaign matters. Number crunching, finding out which precincts and even which voters in precincts need to be worked. It’s the grinding work of campaigning, posting signs, registering 1st time voters, knocking on doors, sitting at phone banks, robocalling, emailing, sharing, tweeting, liking. This can be the difference between winning and losing, and without question, Obama’s data crunchers did a better job than Romney’s and helped him win the election.
I read this earlier, (but did not share it). Great article!
So 538 has her back at low-mid 50s again in all forecasts including nowcast. Wang has bumped her back up to 87%. Is it just me or did he just change Florida from tossup to Clinton? I don’t know which polls would have done that, though the three new national polls listed on Huff Pollster have Clinton +1, +3 and +7.
Why would Wang bump her to 87%? What polling would justify that?
Not sure, but Wang now has FL at medium blue and NC at light blue on 270 to win. This was all from yesterday’s mid-afternoon (pre-debate) update.
Not sure. I think he uses HuffPo Pollster and the last 3-way polls they put up were Clinton +4 (Monmouth) and Clinton +5 (NBC/SurveyMonkey)
http://elections.huffingtonpost.com/pollster/2016-general-election-trump-vs-clinton-vs-johnson
Wang’s number have seemed to bounce around a little more than they did earlier, istm.
I’m repeating myself a little, but I think it’s under-appreciated how the phasing out of old polling affects the models. Suppose your rolling average of polls looks like this for the Clinton margin: 10, 9, 8, 7, 6, 5, 4, 3, 2, 1. If you get a new poll that’s not any different from the last poll, it looks like this: 9, 8, 7, 6, 5, 4, 3, 2, 1, 1. That moves the average by a full point, even though nothing has really changed in the race in the short-term. Obviously, the same can happen in reverse too.
538 and other models don’t just chop out polls crudely like that, but there’s a similar phenomenon at work as new polls get highly weighted and old polls have their weights drop off. IIRC, PEC takes it’s polling from HuffPo, but which has a secret sauce for how they arrive at their topline numbers from the polls. But it’s some form of rolling average, so the same principle applies–namely, if there was a big swing about a month ago, and then the polls show a steady race for a month, the models will continue to swing even though all that’s happening is that months-old polls are washing out rather than the race moving from the new equilibrium.
Wang pretty specifically does not have much of a “secret sauce” as I understand it. He looks at polling variations in the past and calculates for a similar possible variations in current polling. DSeid could no doubt explain it better than me.
Eta: oh, I guess you were actually referring to a HuffPo secret sauce. Dunno about that.
Right.
And I’m not saying it’s some advanced algorithm. It’s some form of a rolling average that smooths things out. I just don’t think they report the exact formula anywhere. The point being that the topline number reported on Day X includes polls in the field from Day X to Day X-n, and we don’t really know what the n is, or whether it’s a simple average or some more advanced weighing system.
That’s at least once nice thing about 538, since they report exactly what weight they’re giving every poll. So while their predictive model is less transparent, their report of the national polling aggregates is more transparent than PEC/HuffPo. If you are so inclined, you can look at the constituent polls and, using their weighing, set the look-back period to be shorter in order to get a clearer picture of when the race has actually moved.
HuffPo does give some explanation but I’m not sure it’s enough to reverse engineer their sauce:
Congratulations to Sec Clinton. Coming up to 60% from the downside sure feels a lot better than coming down to 60% from the upside. I really hope I get to see how 70% feels.
I’m praying we all witness 100%.
I miss >80%.
Do I get to wish? OK, I wish for the margin for tired ol’ Hillary to be solid and reliable, but somehow all the wishy-washy liberals, the blacks, the hispanics, women and the young are worried enough so they are pounding on the polling place doors at 7am. And vote Big D, all the way down.
Works for me.