Friday, August 3, 2012

EV Models: What's Up With The Polling!?

Today's electoral vote histogram
The Princeton Election Consortium Gives Obama a 90% Chance of Victor and Predicts A 330 EV Landslide. WONDER HOW THAT GUY VOTES??
Depending on who you ask, if the election were held right now, Obama would have a 7-in-10 chance of winning, a 9-in-10 chance of winning, a 6-in-10 chance of winning, or 4-in-10 chance of winning. What the heck is up with this?

A note on reading that chart above: That red-line at 270 isn't Romney's votes--the picture only shows Obama's EV votes. That red line is the "finish line." Those blue bars are where the guy thinks Obama's final EV count will wind up past the finish line. Does that even pass the "is this totally absurd" sniff test?

What Is Going On With This?
Put simply, Obama seems to have some kind of edge. That edge is a little better in the swing states, and after that? Polling is wildly all over the place. The question is: how good is the polling? That's a hard question to answer. The numbers suggest that most polls seem to over-sample Democrats (or at least a lot of them do). Let's look at the recent Pew Poll that gives Obama a staggering +10 lead. Here's the breakdown from HotAir:
Almost Double The Number Of Donkeys

So, okay: they polled--presumably--presumably in good faith--and got 813 Democrats and only 460 Republicans? Why? Did their "random sample of voters file" come from the DNC? Did they, I don't know, use ACORN forms to get names from? What's going on?

One Thing To Keep In Mind
Before we try to answer the "what the heck" question--and I'll tell you now, I can't do it with any authority--it's important to note that the polling houses do adjust for their samples:
The combined sample was then weighted to match gender, age, race, Hispanic origin and nativity, education and region to parameters from the March 2011 CPS and to match population density to a parameter from the Decennial Census. . . . The following table shows the sample sizes and the error attributable to sampling that would be expected at the 95% level of confidence for different groups in the survey.
So presumably they got skewed respondents and tried to mathematically "right" the numbers for what they think "reality" is. Do they think "reality" is an equal number of Republicans and Democrats? I don't know.

So What's Going On?
So what is going on? I have no idea, not being a political scientist--so I asked Google. Here are the results from the front page ... -- Again: Media Skew Swing State Polls To Boost Obama
This poll, and many others like it, are simply propaganda. They are meant to reinforce the idea that Obama's and the media's attacks on Mitt Romney are working. They are meant to reverse the growing belief that Obama's reelection is in serious jeopardy. But, the days of the media controlling the campaign narrative are over. -- What's Up With All These Democratically Skewed Polls? 
One possible reason is that not everyone who is a registered Democrat actually votes that way. Geraghty notes that in North Carolina, for example, there actually are many more registered Democrats than Republicans, though Democrats tend to lean conservative.
National -- Why Are So Many Pollsters Oversampling Democrats?
I’ll make a quite modest defense of PPP on North Carolina, in noting that there are a lot of conservatives down there who are still registered Democrats.
 But the best is Commentary: -- Five Reasons There Are Bad Polls
1. They like the result. (They are LIBERAL!)
2. Polling is expensive. (Good Polling is Expensive Polling)
3. They blew the sample for other reasons. (Trying To Include Minority Samples Is Hard)
4. They polled on the wrong days. (Time of Day, Day Of Month--Makes A Big Difference)
5. Polling is dying. (Cellphones and Other Things Make It Hard)
What Do I Think?
I think the idea that this is a national media conspiracy is wishful thinking. It's conspiracy thinking and should be beneath serious consideration. Bias? Hell yes. Conspiracy? No. And to be fair, Commentary dismisses it--and I'd accept nothing less from Brietbart anyway! So I don't know. But I suggest everyone look at this: The guy from Princeton talks about InTrade betting markets:
Even when lots of data are available, such as political polls, InTrade can still fail. One simple reason is bias: InTrade bettors appear to skew Republican. This could explain why there is such a mismatch between the poll-based Obama win probability (>99% for an election today, probably >80% in November) and the InTrade price (equivalent to a probability of about 0.56). This could be excused on the grounds that the election is far off, and there is uncertainty as to what will happen in the next 4 months. However, there is a third flaw.
As I’ve written before, InTrade bettors are habitually underconfident in the face of polling data, even on the eve of an election. Even a 10-point lead in a race is insufficient to drive a market-based probability estimate above 80%. This is perplexing since such a lead is basically a sure thing.
This is important: if, indeed, InTrade bettors are more Republican and they are ignoring polling data ... when it turns out to be strong ... and right ... then maybe the problem is not with the polls themselves. To believe all the polls are bad is nonsense. To believe that only Rasmussen is any good seems unlikely. Trying to get objective data about various track records is hard because all the polls seem to "tighten up" just before an election so it is hard to say if their record was correct during the time before.

Jay Cost and Sean Trende do seem to favor Rasmussen. Nate Silver finds Rasumssen to have a fairly strong house-effect but its focus on likely voters explain a lot of that (Republican demographics are more likely voters than Democratic). So I'm not sure.

A lot of people dismiss any unfavorable poll out of hand and I think that's a very human thing to do. But so is believing heavier objects fall faster as was the conventional wisdom for centuries--I have to believe that mathematics can adjust for variance in methodology (Nate Silver's model, for example, revises that +10 Pew poll to a 'good' but not news +4). Although Silver leans Democratic, I have to think that is the right kind of thinking and, if there is any value in the underlying data whatsoever, I have to think that the numbers the collective is showing have at least some merit.

If I were Romney, I would be concerned (and, note, he is starting bigger buys in the swing states).

No comments:

Post a Comment