November 26, 2012
Why Did The Washington Poll Miss So Badly On The Presidential Race?

Almost all the votes have been counted here in Washington state, so we can look at just how close "Wave 2" in the Washington poll came to the almost-final result.

And what we find is genuinely puzzling.  President Obama currently has 56.16 percent of the vote, and Governor Romney 41.30 percent, so Obama has a 14.86 percent margin.   But the poll predicted that the margin would be almost 21 percent.  (57.1 percent to 36.4 percent)  Before the election, I said this was implausible, and predicted, correctly, that Obama would have a smaller margin in Washington this year than he did in 2008.

Now you might just attribute this 6 point error in the margin to the kind of bad sample that affects every poll from time.  Except, the poll was quite close on the governor's race.  But then not on some other, down-ballot races.

So I don't see any obvious explanation for those inconsistent results.  But I do know that, if I were running KCTS, I would want to ask Professor Barreto about those inconsistencies.

(A Democratic pollster, PPP, erred even more badly, in the opposite direction, predicting a 5 point margin for Obama.  I didn't believe that poll, either.

If you would like to make some comparisons, here are the Gallup errors for all the presidential elections except the last.)

Posted by Jim Miller at November 26, 2012 07:10 PM | Email This
Comments
1. In the words of a former state political operative; "Polls schmolls"

I don't know why, but I do know that the GOP is being duped and is unable to make their case in the negotiations about the fiscal cliff. If the GOP does not learn to communicate effectively and keep laying out their case, they will be blamed by the public no matter what happens.

Republicans - Raise taxes on the rich - compromise between 35% and 39.6% -say 37.5%, but not unless there is entitlement reform by the Democrats. Also, give in for the sequestration - military cuts for at least one year - they are wanting it both ways if they don't and will not have a leg to stand on.

What they should also do - The negotiations should be televised on CSPAN - they gave in to the Democrats secret meeting wishes again. It may be too late for that and if that's the case, when are they going to learn to be shrewd and grow a pair ? Secondly, call Obama's bluff - he needs to inform all of his plan - no wonder he is teflon/ the GOP is aiding and abetting this. Again the GOP is not being shrewd and are afraid what the perception might be. The screaming message to them is stop being afraid and do something about it !! Go on offense and hold daily press conferences - McConnell did today and if he is as press saavy as the Democrats - he will have one every day. Also get rid of the consultants - they are akin to the ones that led Romney to his defeat !

Posted by: KDS on November 26, 2012 08:39 PM
2. "Fiscal cliff" - A few more elements that need to be brought to the table - Obamacare and spending cuts and the ratio of spending cuts to increases of revenue should be at least 4:1 for this year or no deal.

Right now, it looks like the Republicans are unwittingly doing the Democrats bidding. They are capitulating to raising taxes - which may be OK if they get something substantial in return. Sequestration will make it a lot easier to obtain the 4:1 ratio of spending cuts to revenue increases. The GOP does not even have a tax plan and in the meantime, they appear to be letting Obama get away without showing us his plan...

I hate to say this, but the GOP blew the elections in 2012 because they can't help themselves from being the stupid party and we are seeing evidence of it again right here !

Posted by: KDS on November 26, 2012 08:54 PM
3. Jim: You are either innumerate or you're deliberately misrepresenting the Washington Poll results. Judging from how most GOP commentators were incapable of understanding poll results before the election, I'll go with the former.

The Washington Poll did not explicitly predict a 21 percent difference between Romney and Obama. It estimated that 57.1% of likely voters prefered Obama with a 3.9% margin of error. In fact, 56.2% of WA voters selected Obama, so that's well within the margin of error. In fact, it's a pretty good prediction.

The poll also estimated that 36.4% preferred Romney, also with a 3.9% MOE. Actual result was 41.3%. That's an "error" of 4.9%, only 1.25X larger than the standard MOE. Having a result that deviates from an estimate by 1.25 times the standard deviation is not unlikely, especially when you consider that the poll showed that 2.8% of likely voters had no preference.

As someone who dabbles in statistics, you should know that if you estimate two values, each with an MOE of 3.9, the MOE of the estimated difference is not 3.9, it's 5.5. Let me know if you need an explanation for this. Or you can ask Professor Barreto to explain it if he can stop laughing at your "analysis". From the Washington Poll data we can infer an estimated difference of Obama vs. Romney of 20.7% with a MOE of +/- 5.5%. The actual difference was 14.9%, which is just barely outside the expected MOE.

Polling is an inexact science, but in this case, the Washington Poll was within or near the expected MOE.

Posted by: scottd on November 26, 2012 10:49 PM
4. They no longer count ALL the votes. Say in California, if GWB and GORE were at it.....and they hit a point where GORE has 1,000,000 votes more than GWB.....they stop counting when they have like 800,000 votes left, because if all of those votes went to BUSH, GORE would still win the vote.

Posted by: go to these guys on November 26, 2012 10:55 PM
5. While scottd already did the heavy lifting, statistically-speaking, one of the things I liked most about the Washington Poll is how the self-identified Tea Party voters were equally likely, or more likely, than self-identified Republicans to vote for the Republican Party candidate. This is just further proof that the Tea Partiers are independent patriots, equally-disillusioned with both major parties.

It is also good news for John McCain.

Any luck in un-skewing any other polls, Jim?

Posted by: tensor on November 26, 2012 11:43 PM
6. tensor: It's like they live in an alternate reality.

For weeks we all looked at the same polling data. In aggregate, the data clearly showed that Obama very likely had at least an electoral vote win. Yet, folks like Jim Miller and the SP peanut gallery managed to cherry pick a few results, "unskew" the rest, and convince themselves that the same data somehow showed a strong Romney win -- maybe even a landslide! That's remarkable obtuseness.

In the same vein, Jim has twice written about the Washington Poll. This poll has shown demonstrably solid results. Most of its findings are well within expected margins, a few results are just outside of those margins. Nothing that's statistically unexpected. Yet Jim (who wisely wimped out of making his promised pre-election predictions) tries desperately to convince us that the poll "miss[ed] badly". The most amazing thing is that he might actually believe this.

Posted by: scottd on November 27, 2012 07:12 AM
7. I don't see the topic being that much of an issue. Virtually all polls correctly predicted the statewide winners with varying accuracy. A good number of the National polls and conservative pundits incorrectly predicting the POTUS is more of a significant issue, but that was not brought up as a topic, probably by design as it has already been cussed and discussed at the National level.

That is why I introduced the topic of negotiations of the fiscal crisis at the Federal level. IMHO, it is a more stimulating and crucial issue that will eventually come up than this topic - no offense, Jim. It is not looking particularly good right now either - predictably both sides are not putting the country first. The eventual outcome does not look good for the short term and maybe it is better to have nothing before Jan. 1st. What say you ?

Posted by: KDS on November 27, 2012 08:22 AM
8. KDS @7
It looks like Chamblis, at least, is considering country first. Norquist's antiquated pledge is out of date and not right for the times. It is good to see Republicans finally waking up and not letting this yesterday's news person (Norquist) drive their today negotiations.

Posted by: tc on November 27, 2012 12:46 PM
9. Jim:

As scottd and tensor have already pointed out, there doesn't seem to be much of a problem here. The poll was within one point of Obama's final number. It was off a bit more for Romney, but I would guess that was mostly caused by people like the local peanut gallery here telling posters they were not committed to Romney (because he was a filthy liberal RINO, or some such reason), and yet in the end voting for him.

I would suspect the Governor's race had fewer Republicans claiming to be unsure as to who they would vote for. Sure, McKenna was a filthy liberal RINO also, but the ultra-Conservatives around these parts are willing to swallow some pride after 30..., sorry, soon to be 34 years, of Democratic rule.

Posted by: Lionel Hutz Esq. on November 27, 2012 12:58 PM
10. I used the spread because that's what you want to measure. If Rob McKenna had a +1 spread, rather than a -3, he'd be planning reforms to our state government, rather than looking for a new job.

(If Barreto had allocated the undecided votes, I could have used the candidate predictions, directly, but he didn't.)

Years and years ago, I rejected the frequentist approach to statistics, and so arguments based on margins of error, and arbitrary confidence limits mean nothing to me.

Simplifying greatly, I think that the farther a poll is from the actual result the worse the poll is, and so that 6 point error is a serious mistake.

But what makes it interesting is that the same poll was almost exactly right on the governor's race. And that oddity calls for explanation from Barreto.


At this point, I thought of asking "scottd" to do me favor, but realized that would be silly. Judging by his comments here, he sees the SP contributors as enemies, rather than political opponents. (Not so incidentally, I don't consider him an enemy, but a political opponent.)

So, instead, I will suggest that scottd do himself a favor, and consider the truth of what
Ovid said, long ago: "Fas est et ab hoste doceri."

And now I am going to close this post, because few others appear to be interested in the subject.

Posted by: Jim Miller on November 27, 2012 01:25 PM