…Of deconstructing polls there is no end; and much political analysis is a weariness of the flesh. (A gold star to the first reader who catches the reference)
It’s less than a week until the midterm election. We’re being bombarded by polls. But simply put, they’re not all created equal. In some cases, the problem is easier to see than in others. Take for example the new CNN/Time poll on Colorado races with a mix of credible and incredible top-line results (H/T Complete Colorado). The Denver Post‘s Curtis Hubbard picks up on some of the survey schizophrenia. This one merits a closer look.
First, the seemingly credible: Republican Ken Buck leads appointed Democrat incumbent Michael Bennet in the U.S. Senate race, 47-46.
Then there’s the incredible governor’s race finding:
John Hickenlooper (D) 51, Tom Tancredo 37, Dan Maes (R) 10… A 14-point margin? A few days ago Hickenlooper’s lead fell somewhere between 1 point (in a Republican poll) and 3 points (in a Democrat poll). No dramatic stories or news events happened, and all of a sudden we’re supposed to believe that Tancredo lost all that momentum?
There may be other shortcomings, but the CNN/Time poll carries the burden of one very significant problem: the partisan breakdown of likely voters. The pollsters don’t share the breakdown of R’s, D’s & U’s, but you can deduce them by running the numbers on the cross-tabs. And folks, it ain’t pretty.
If you believe that the breakdown of Colorado voters in this election will be roughly 34 percent Republican, 38 percent Democrat and 28 percent unaffiliated, then by all means the CNN/Time poll is highly credible. But if you inject the conservatively realistic sample of 38 percent R, 35 percent D and 27 percent U, you get the following results:
- Ken Buck 50, Michael Bennet 44
- John Hickenlooper 48, Tom Tancredo 40, Dan Maes 11
If you inject the percentages based on actual early voting returns this far, you get the following results:
- Ken Buck 51, Michael Bennet 43
- John Hickenlooper 47, Tom Tancredo 40, Dan Maes 11
Let us hear the conclusion of the whole matter… In other words, based on sample alone, the CNN/Time poll skews the results 5 to 7 points in the Democrats’ direction. You get into the internals, and it gets sketchier.
In the U.S. Senate race, your first clue that something is wrong should be seeing Buck leads Bennet 49-36 among unaffiliated voters. How in the name of reality could both candidates hold their respective base parties at 90 percent, and the Republican lead 13 points among independents but only lead overall by 1 point? At the same time, that 13-point advantage among unaffiliated voters (based on other polls) seems too optimistic for Ken Buck. Or is it?
In the governors race, other recent polls consistently have shown Tom Tancredo leading or tied among unaffiliated voters. But not CNN/Time, where he trails 45-41. And while Tancredo consistently pulls in about 9 or 10 percent of Democrats in other polls, he only grabs 4 percent in CNN/Time. Is the poll good news for the independent conservative gubernatorial candidate? No, he has an uphill battle to the finish line. But even the 7- or 8-point deficit projected by a realistic voter sample looks too large.
Finally, this one from CNN/Time will knock your socks off: John Hickenlooper leads Tom Tancredo among rural voters, 42-31. Do you really think so?