Monday, October 29, 2012

The War On Averaging

Coming down to the wire, with 'Frankenstorm' Sandy bearing down on us, the war between the parties has centered on ... math. Specifically Averaging Polling.

From the Right
From The Left
Who's right? Who's wrong? Who knows? I, personally, will side with The Math / Results--but what is The Math? If the polls are all skewed then isn't there a problem with Garbage In - Garbage Out? Maybe. Let's look at the sides:

The Quants On the LeftOn the left we have Nate Silver (, Dr. Sam Wang (Princeton Electoral Consortium), and Andrew Tanenbaum at These guys have the following Track Record:

  • 538: Their model got 48 out of 50 states right in 2008 and got within 12 of the EV count (353 vs. a real 365)
  • PEC: Their model got 49 out of 50 and estimated 364 EV out of a real 365.
  • He called 49 out of 50 states in 2008 and got the same EV count that 538 did (within 12 EV).
Their current models look like this:
Obama 75% Likely To Win
Obama 90+% Likely To Win doesn't do a % to win but puts Obama, today, at a 280 vs. 235

What Are They Saying?
They are saying that Obama is pretty far up right now based on (a) the averaging of an unprecidented amount of polling and (b) the historical accuracy of said polling and (c) the way the electoral college works (i.e. national vote doesn't count for anything--it's all about like Ohio and shit).

Essentially, if Romney does not win Ohio--where he is behind right now according to almost all polls--he is in deep trouble. Here is on "Plan B"

The Quants On The Right
On the right we have:
  • Scott Elliot of Election He got 48 out of 50 states right in 2008 and was off by 27 EV
  • Dean Chambers of Unskewed who has no track record, but is re-weighting polls based on various external sources (was Rasmussen, now: something else after Rasmussen said 'don't do that').
  • Dave In Florida: Again, no track record--but he has a different method for re-weighting polls.

What Are They Saying?
They are all saying versions of the same thing--in slightly different ways:

  1. The national polling shows a dead heat with slight (very slight) advantage to Romney. The state polls should fall in line.
  2. Polls don't seem to "look right"--they estimate too many Democrats.
  3. Nate Silver seems kinda gay, doesn't he (this is Dean Chambers).

Here is the most credible of them, Scott Elliot, The Blogging Caesar (Election The Polls Will Shift To Romney In The Coming Days:
Things just aren't adding up out there. On the one hand, you have state polls in places like Ohio, Colorado, Virginia, Nevada and New Hampshire showing an extremely close race. On the other hand, national polls and job approval numbers are at losing levels for an incumbent president.
However, he also doesn't appear to put any stock into the idea that polls might be skewed in favor of Obama and other Democratic candidates. I do. And the prediction in the title of this post is an offshoot of that belief. Sifting through the plethora of factors in the run up to Election day - national polling, voter enthusiasm, candidate campaign scheduling, early voting results, to name a few - I have become convinced that Mitt Romney is on the path to victory.
I also know that pollsters don't want to be wrong. And that's why I believe we're in store for ever-improving Romney numbers in state polls that will be released between now and November 6. Before next Tuesday, look for Obama's 2.7-point lead in aggregate Ohio polling to shrink to zero, look for a couple polls coming out of Wisconsin to put Romney up by a point or two. Also, look for polls in North Carolina, Florida, Virginia, Colorado and New Hampshire (except, perhaps, for UNH) to show Romney pulling into a clear lead in those states.
President Obama is on the ropes and fading. All the signs are there. All, that is, except for battleground state polls. No worries there, though, they'll fall in line soon enough.
Dean Chambers isn't happy with Nate Silver and isn't sure why anyone would listen to him in the first place:
Nate Silver is a man of very small stature, a thin and effeminate man with a soft-sounding voice that sounds almost exactly like the “Mr. New Castrati” voice used by Rush Limbaugh on his program. In fact, Silver could easily be the poster child for the New Castrati in both image and sound. Nate Silver, like most liberal and leftist celebrities and favorites, might be of average intelligence but is surely not the genius he's made out to be. His political analyses are average at best and his projections, at least this year, are extremely biased in favor of the Democrats.
Here is Dean Chambers, for comparison:
Dave In Florida has a bunch of blog posts--but says in his most recent that Rasmussen wins by being the first company to come out showing Romney ahead in Ohio--which is what we'd expect, right?
Scott Rasmussen, come on down! You are the big winner!
I asked yesterday who would be the first to show an Ohio lead, and it was Rasmussen. His new Ohio poll is conducted completely on a Sunday (traditionally a bad day for polling Republicans) and it finally breaks through the wall, and shows Romney ahead 50-48. The D/R/I is 38/38/24 which at an Even split is a little surprising. Independents favor Romney by 4. Even so, we still get a silly result where 33% of voters claim to have already voted. As of last Friday the actual ballots returned were about 800,000. For 33% to be correct, 2 million votes would need to have already been cast.
What Do I Think?
In case it's not clear: I think that the people with (a) real track records who (b) are doing what is typically regarded as best-practices-math are right. The person I've picked "on the right" who I respect the most is Election Projection and I'll note that while he does unskew some polls (see his link) he doesn't use them on his projection. That's fine.

What Dean Chambers and Dave in Florida are doing is a lot more like conspiracy-theory-wishful thinking. The idea that Nate Silver gives lefties morale because he's propping them up with lies is, similarly, failure to face math: it's possible that this year the polls are garbage--but there's no evidence of that.

I want to talk about that for a minute.

  1. The large +D numbers are not per se evidence of a bad poll. They may be evidence of voter party ID shifting. They may be evidence of who picks up the phone. They could be evidence of enthusiasm. In any event, if you "unskew" to use, say, Rasmussen party-ID, you risk gravely damaging your poll in other ways. The unskwerers using Rasmussen are doing exactly that and are probably not aware of it. 
  2. Averaging polls is, I believe, a far better way to unskew polls than adjusting from the internals. Unless you believe that the vast majority of polling agencies are suddenly in a conspiracy, you are throwing out your best weapon in seeking the truth: a huge sample size.
  3. There may actually be something going on. For one thing, cell-phone polling is limited and a lot of people are cell only these days. Demographically the cell-only crowd tends to prefer Obama. So that won't help you.
  4. The response rate for polls is something like 1-in-10 calls. This may be the source of problems--especially if (disgusted) Republicans won't pick up the phones. If that's the case, things might, yes, be off. But skewing by Rasmussen's numbers won't correct for this. You might as well just pick a % you like and add that to the polling averages based on "Republican disgust." I'll say 4% since it gives Romney a clear win.
  5. Finally: It's not clear that turn-out will be that much lower than 2008. Yes: there are probably a great deal of people who are not happy with how things turned out--however, if they are still politically active or can be made active they may still vote--and despite what you think, they may still vote for Obama. The internal narrative that a lot of conservatives have bought into--that Obama was supposed to be a godling and turned out to be a mortal--is not the only one out there. A great deal of people thought Obama "got blocked by Republicans when he tried bi-partisanship." A great deal of people think "things were so bad he couldn't be expected to fix them." A lot of people still, well, like him and, despite his debates, do not especially like Romney. These narratives are all out there and they are just as true (and just as false) as the one conservatives like. 
So, no--while something might be going on that damages polls (cell phones, response rate), I think the available evidence is against there being a conspiracy that only unskewing can fix. However, I like The Blogging Caesar's approach: we have like 9 days left: let's watch and see if the polls shift over those days to favor Romney ... it's a testable hypothesis.

No comments:

Post a Comment