Thursday, October 19, 2006

Taxing Polls

Just a disclaimer, there's some boring math and so forth in this post so read at your own risk.

As the election draws nearer I will be a bit more political than usual (well actually anything is more political than usual since I haven’t posted in 5 weeks, but anyways). So with the campaign season comes the usual festivities. Pollsters and pundits can find work again, producing wildly offbase polls for the former and using those absurd polls to predict the death of the Republican Party for the latter. Lefties believe said polls and pundits along with other fever swamp theories about the Bushitler personally supplying and testing North Korea’s nuke, then personally cutting Corey Lidle’s gas line so he would crash into a New York building, and finally rushing back to the White House to send emails to the NFL threatening dirty bomb attacks on football stadiums. Sigh, if only he would use his genius for good. The Democrats are on their usual no holds bar quest for power. When not outing gay Republicans they are busily trying to divine whether a GOP senator used the “N-word” some 40 years ago and hiring Nazi genealogists to determine if this senator also has the blood of the Juden in his veins. If you think that would be the limit you'd be wrong, they are also throwing oreo’s at a black GOP senate candidate while having Nazi ethnologists decide if his contributors have features of the Juden. Ahhh, makes you feel like a kid Christmas time, you just wish it could go on all year.

The key to all this election year behavior is the polls. I know, it seems like every election we go through the same routine. Polls predict GOP rout, lefties get irrationally exuberant, election happens, their bubble pops, and then we get to listen to them cry about it and come up with dark conspiracy theories about stolen elections, questioned patriotism, created terrorism, and such until the next election season when we start the cycle all over. So in the spirit of trying to help some lefties become rationally exuberant, and some conservatives to not be so irrationally depressed, I’ll analyze a recent poll and point to some other analyses of polls that support my theory that polls are becoming more and more useless.

This is a recent New York Times-CBS poll for the Ohio Senate race. I picked it not only because it is at the high end for Sherrod Brown but also because it is the only one of the five recent polls to include internals. It has DeWine losing by 14 points. The first thing to note is that the poll has almost 100 questions and fills up 30 pages! As is one of the oldest complaints about these polls, obviously the middle aged middle income couple with kids who make up the bulk of voters and the bulk of the Republican Party do not have the time to answer these things. Nevertheless, delving down to the bottom of the report they have the poll's weighting. Of all respondents the Rep/Dem breakdown is 30/34. Not too unreasonable, the 2004 exit poll had the party breakdown in Ohio at 40/36 so it’s sort of close. Then they weight it to a 27/34 breakdown without saying how or why they did it.

They then further adjust the poll for “likely voters”. They only show the likely voter total (693 out of 1164) and like the weighted total they do not say how this number was calculated. It seems to have been done by simply multiplying the weighted respondents by the number who said they will “definitely vote” (page 15, question 2) since I arrive at 684 “likely voters” and also the same Brown by 14 result doing that. I thought pollsters had some elaborate method of foretelling who would vote and who wouldn’t, but I guess I was wrong. So after correcting for “likely voters” the poll sample is (R/D) 27/37. Again, this is compared to 2004’s 40/36. The change per party (with independents tossed in) is -13/+1/+14. Anyone who has followed politics for more than just this year would recognize this result is ridiculous.

It is possible this year will have lower Republican turnout, but 33% lower? We can also compare these weights to historical turnout (state election website). In 1998 and 2002 (picked since they followed motor voter and such laws that caused a major increase in registration and corresponding drop off in turnout) turnout was 49% in Ohio. In the 2002 primary turnout was 19.4% with 586,000 Dems and 659,000 Reps voting. In 1998 primary turnout was 28% with no partisan breakdown. This year’s primary turnout was 24% with 872,000 Dems and 870,000 Reps voting. So yes this indicates higher Dem turnout but not a 27/37 advantage. It also means that if turnout is 49% or thereabouts again and Republicans are only 27% of voters then Republicans will only increase their turnout 12% over the primary. Further, the poll has likely voters at 70% of registered voters which means a full 30% of them are lying and shouldn’t count. This causes the poll to be worthless, are those 30% DeWine voters, or Brown voters? Or they could be equally both, no one knows. My own guess, for what little it is worth, is that the turnout will be (R/D) 36-37/38-39.

Others have picked up on this line of reasoning as they do before each election. Rich Lowry at the National Review has a good review of the partisan weighting of the “Generic Ballot” polls (which are even more useless than Senate polls). In short, they are also grossly over-weighted in favor of the Democrats. Also at the National Review, Jim Geraghty reviews a lot of House polls from across the country. He looks at polls from the same district released during the same time. Despite being the same district and the same time frame the polls have radically different results. For example, in Iowa-1 a poll has the Democrat up 11 points while another has the Republican up 13 points. So either it’s a major win for the Dems or it’s a major loss for them. You can see this in the DeWine-Brown race. Looking at the Real Clear Politics list of polls, Brown is either up 4 (Zogby) or 14 (NYT-CBS). Overall, it’s the same as it is every election. Media polls show Democrat victory, election reveals Republican victory (though certainly less of one this year than most), sun rises, sun sets.

Update: Looks like every leftie's favorite pollster (Zogby for those who didn't know) has a series of good Senate polls for the Republicans. I swore off Zogdby long ago (Kerry winning Florida and Ohio my, ahem) and put no credance in any of his polls but since lefties love him there you have it.

1 comment:

  1. You're writing again. Looking forward to reading the new posts.