Wednesday, September 10, 2008

RealClear Gallup Anomaly

Recently I have been compulsively going to www.realclearpolitics.com for the latest polls. Right now they show a huge gigantic Gallup anomaly with McCain ahead on average 7.5% in 2 Gallup polls and 0.8% in non Gallup polls. What the hell is going on ? Part if it is (like everyone as far as I can tell) realclearpolitics insists on reporting likely voter numbers if they are available. Gallup doesn't even report results for a likely voter sub-sample of its tracking poll (McCain + 5 among registered voters). This is, according to Gallup, real information. I see no reason to take it less seriously than the polls which show a closer rate and a clear reason to count it more than most as it has a huge sample size. On the other hand, Gallups headline number for the USA/Today Gallup polls is *not* McCain + 10 but McCain + 4 among registered voters.

Gallup makes it very very clear that they don't believe their likely voter filter is useful in early September (it has a very good record of performing well in late October). It is based on a few questions. One is "do you know exactly where to go to vote ?" If it is late October and someone doesn't know he or she is not likely to vote. Not true in September. The question removes the young and people who move around a lot. Other questions relate to enthusiasm and how closely one is following the election. Obviously these questions introduce bias right after a party convention (imagine they were 1) were you exited by the most recent convention and 2) did you watch a lot of it on TV then explain how they are really not like that).

Gallup knows this. Gallup says this in the strongest possible terms. At Gallup the headline is about the registered voter sample. Gallup didn't even release the sample size of the likely voter poll. Yet many organizations focus on the results with the likely voter sample. I suppose this is for "comparability" with other polls, but that is silly -- the Gallup likely voter filter is so different from that used by other agencies that the results are not comparable.

Note, however, that Gallup doesn't even report likely voter numbers for their tracking poll which also has a huge sample. The McCain +5 in the tracker is very important information. The McCain +10 in the USA today Gallup poll should not be reported, since Gallup insists that their best estimate based on the raw data in that poll is McCain + 4.

I am having trouble finding the Gallup report on the USA today Gallup poll but I just found this "Candidate support by educational attainment" Obama's support is lowest among the "high school or less" subsample. This fits the "elitist Democrats" line which the Republicans use when they mean "elite Democrats" so well that people might not notice that it is extraordinary. Normally support for the Democratic candidate is higher for "high school or less" than for "some college" or "college graduate" (which are usually quite similar, also Usually and for this election, support for the Dem is higher for "post graduate" than "some college" or "college graduate"). Obamas trouble (relative to the average Democrat) among people who never went to college is a sign that the elitist arugala idiocy issue is costing him.

No comments: