Tuesday, July 29, 2008

The Early Bird gets the Worm

As my anti computer virus strategy, I am always late posting. In particular I want to post about the USA OK Gallup outlier poll which shows McCain ahead for the likely voter sample (and not for the registered voter sample).

The left blogosphere is dumping on Gallup. I would like to defend them. Yes the likely voter number is unconvincing. I note that Gallup insists that it is not the number of interest and is not to blame. Very much to Gallup's credit they say exactly what their likely voter filter is, so it is clear why it works well in October and not at all in July.


As e.g. Steve Soto notes, as regular as the seasons, there is a gallup anomaly every 4 years . Clearly the most likely explanation is the likely voter filter which differs across pollsters. Here the internals basically prove that it must be the likely voter filter (no calculations but given the overlap the difference between the results with the two samples is clearly significant).

So what is it about the Gallup likely voter filter ? Well *as Gallup insists* it is useless in July.
http://www.gallup.com/poll/109135/Who-Likely-Voters-When-They-Matter.aspx

Back when they were a near monopoly they didn't even release the numbers. Now they told people that the number to look at was support among registered voters (absolutely explicitly and no one listened). The reason that the crazy number is crazy is, I think, due to the integrity of the Gallup organization.

Let me explain. IIRC they use the same filter they have used for decades. This is honest. It means they aren't fiddling the numbers. It means that old Gallup performance is relevant (and their last polls before election day do very very well). It means the filter is transparent, publicly described and can be implemented by hand without a computer. It also means that it is much worse than worthless in July (as they have been saying for decades).

Oooops they are fiddling the numbers. I was trying to find the 7 Gallup likely voter questions and I discovered that they used only 3 this time. My point (which is totally invalid) was that one of the seven questions is "where is your polling place". Not knowing that in late October is a sign that the registered voter is not likely to vote. It doesn't mean much of anything in July and Gallup didn't even ask.

Looking at Gallup performance in past elections based on their last poll and the outcome is a total waste of time if one wants to decide if they or the other pollsters have good estimates now (or in the regular interval when they have numbers which are better for the Republican presidential candidate). They are using a different filter now. The full 7 question filter (which works very well near election day) makes no sense now.

Gallup does have the integrity to publish the likely voter filter which they are using* . Click the link for details but the 3 questions are
At this point, the USA Today/Gallup poll is using a set of three questions to determine likely voters:

"1. How much thought have you given to the upcoming election for president -- quite a lot, or only a little?


2. How often would you say you vote -- always, nearly always, part of the time, or seldom?



3. Do you, yourself, plan to vote in the presidential election this November, or not?"

My guess is that to be a likely voter one has to answer very much, always and yes. Importantly question 2 refers to past elections. Participation is low and rises sharply with age roughly up to age 30. Question 2 thus excessively screens out young voters (even compared to their low turnout). It also can't pick up the unusual enthusiasm for a young Black charismatic super star world historic candidate (if any).



*ought to be required by news organizations as in "if you want us to ever name you on the air, send us your raw data and your numbers and if we can reproduce your likely voter numbers we will call you a pollster and if we can't we will have a long investigative series about whether or not you are a total fraud. But hell, before they do that, they might point out when candidates say things which are demonstrably false, and, if they did that, we wouldn't have to worry about polls this time, because McCain clearly can't help himself.

No comments: