For poll conspiracy-theorists
October 31, 2012 11 Comments
To follow-up my previous post on the Sonic Charmer Electoral Model™, we can now use it for something really important which is to help hone, calibrate and fine-tune your poll conspiracy theories.
As I’ve said, the quandary faced by righties is that they need to disbelieve state polls in order to think Romney isn’t a long shot. Now, there are plenty of half-assed reasons one can come up with to disbelieve state polls, and I’ve come up with some of them myself. It’s fun! I encourage it! Maybe there’s a ‘Bradley effect’. Maybe pollsters tend to be biased lefty Smart People so they fudge whatever fudge factors in whatever direction. Maybe righties Don’t Answer Their Phone As Much because they’re more likely to Have Jobs And Families. All three.
On the flip side, as you know, I think (D)s are Better At Cheating. This means they’ll outperform (accurate) polls which means that a poll that is (D)-biased, in the exact size of (D) cheating, might actually be ‘correct’.
In any event, when I mentally pull these poll biases out of my butt and add them together, I get that maybe polls are consistently underestimating (R)s by ~1.5%. And if I add that number to all the state polls, and run it through the Sonic Charmer Electoral Model™, I get that Romney has a ~55% chance of winning.
Which is good to know. But for folks playing along at home, I realized it might be helpful to know the answer to this: IF…I want to fantasize that the election probability is different from what Nate Silver says, HOW MUCH…do I have to fudge state polls?
Now that I have the Sonic Charmer Electoral Model™ at my fingertips, this is easy to do! Let’s run the model by bumping Obama’s advantage in each state by X%. Remember that for X=0% (no bump), I just come out with the ~78% Obama advantage that Nate Silver has. As stated above I have to bump Obama’s polling by -1.5% (i.e. 1.5% in favor of Romney) to make Romney the (slight) favorite.
Here’s a fuller table below, to aid you in your poll second-guessing/election fantasizing:
| poll fudge (O+): | Obama% |
| 2.0% | 98% |
| 1.5% | 95% |
| 1.0% | 91% |
| 0.5% | 86% |
| 0.0% | 78% |
| -0.5% | 68% |
| -1.0% | 57% |
| -1.5% | 45% |
| -2.0% | 34% |
And the good news for conspiracy-theorists/poll skeptics everywhere is, the sensitivity really is pretty big. You only have to think polls are off (in Obama’s favor) by 0.5% to lower Obama’s probabilities by ~10% or so. That’s not that hard a fantasy to pull off!
So, good to know. Just another public service from the staff here at RWCG, Inc.
There might not be a conspiracy, just simple dumb human error.
After 2004, there was a fair amount of handwringing about the exit polls, which showed a Kerry landslide and turned out to be totally wrong. Too lazy to google it, but Mickey Kaus wrote a bunch about it after the election.
The explanation was boringly prosaic. The exit poll firm used a lot of minimally-trained female graduate students, who did a poor job of selecting a random sample of people leaving the polls (i.e. they interviewed people who looked like themselves or wanted to talk, rather than every nth person).
My main reason for believing that the current polls are flawed is that they reach a tiny fraction of the population – 9% vs. 60+% a couple of decades ago. So it’s probably pretty easy for that sample to become skewed in some significant way. The best indication that there is a such a skew has occurred is that polls have consistently reported a more Democrat-leaning electorate than in 2008. Can you think of a single piece of non-poll evidence that would make you think that’s a factual description of American political attitudes?
Anyhow, great post on the huge sensitivity of a Silver-esque model to polling error. Even a Smart Person like Krugman could learn a thing or two from it!
Right, so if something like that is what’s going on with phone polls, that’s sort of an (R)s-Have-Jobs/Families-And-Don’t-Answer story. I can go for that, to some extent (~2%ish)
As far as the mismatch between Presidential-poll party breakdown and the results of party-alignment polling, that is definitely interesting and requires explanation. One thing I’m just not sure of though is which is the ‘correct’ one. How do we know that it’s not vice versa, that the party-identity polls aren’t flawed because of not matching the party-breakdown showing up in these (perfect) Prez polls? (Honest question, I don’t know the answer)
Also, a lot of (R)s (dhingers) are busy attending Ku Klux Klan meetings and have their phones switched off or turned to vibrate.
Good point. You can’t have a ringtone interrupting a Klan meeting, everyone knows that.
the polls showing obama winning ohio are completely accurate:
obama will win ohio if he gets a bigger turnout now than in 2008.
is that likely?
nope.
pew says early voting favors gop.
ergo: the polls show a landslide for romney.
Ok.
1. How much do you want me to fudge Ohio’s poll numbers towards Romney? I can do that. Even if I do that by 20 points, Obama will still come out the favorite to win the Presidency, FYI. As shown above I basically have to fudge *all* state polls to start thinking of Romney as the favorite.
2. ‘Early voters’ may or may not be a representative of ‘all voters’. They may have a skew one way or the other for all I know. You need a theory/explanation for why early voters voting X means that the overall vote will also be X. Not seeing it here but maybe there is one.
I’m just trying to point you on the right track re: how to build a convincing Romney fantasy. (And that’s precisely what I want, a convincing Romney fantasy!) The main idea is, don’t worry about Nate Silver or electoral-projection models. It’s correct to focus on how best to ignore/discredit state polls, so keep at it.
Obama won in 2008 because of early voter turnout in swing states he lost in Election Day turnout. He doesn’t have that level of early voter turnout for sure this time and it is pretty unlikely Romney will have lower Election Day turnout than McCain (who beat Obama on Election Day turnout in several swing states despite one of the lowest R turnouts in recent history).
In fact, the 2008 race looks pretty different if you take out states Obama won due to early-vote advantage and give them to McCain. This is a surprising aspect of 2008. All those people who voted for the cool black guy didn’t want to do so on Election Day apparently.
That doesn’t seem right, you can’t just arbitrarily ‘take out’ people who early-voted – they’re still voters. Similarly it’s not correct to say those people ‘didn’t want to vote for the cool black guy on Election Day’; they (probably) still did, it’s just that they had already voted so they didn’t have to.
It might be true that (early voters) as a group were more pro-Obama than (election day voters) in ’08. I understand the argument that this bodes well for Romney given how early voting is (reportedly) going, but it’s a somewhat weak argument in the face of continued battleground-state polls.
Indeed, the more I look at polls and the electoral college, the more unlikely a Romney win seems.
Also, Obama’s spending quite a lot of money in states he supposedly is well ahead enough to win in, which is not exactly a sign that the electoral college is lined up for him.
Well, there’s also the generousness of likely voter screens. Strict ones lead to Romney up, loose ones lead to Obama up in polling.
I just don’t see how Obama loses a dozen or more points in polling advantage in swing states from 2008 and still wins when he’s clearly from that same state-level polling lost a bunch of states he had in 2008.
And an oldie but goodie from Mr. Silver:
http://www.fivethirtyeight.com/2009/11/independent-voters-and-empty.html