Coburn opens a lead?
By Adam C Posted in Elections — Comments (0) / Email this page » / Leave a comment »
The last KWTV weekly poll showed Carson ahead of Coburn 40-38, but it's sample seemed odd to me. SoonerPolitics has some strong analysis that may suggest a Coburn lead of 7 points outside the margin of error. We will see if this is corroborated next week, but the details are below.
The KWTV poll looked fishy to me and I was not alone in positing that it could be an outlier. SoonerPolitics has some good information and analysis on the poll. First, it points out the jump in liberal voters:
Last week, 12% of likely Oklaoma voters identified as either “liberal” or “somewhat liberal.” That number jumped to 17% of responses, a statistically significant gain that occurred mainly among the “somewhat liberal category.” The number of self-identified conservatives dropped from 50% last week to 44% this week, while self-identified moderates remained fixed.
Remember that Carson was up 2 last week and remains up 2 this week. So if this sample overweighted liberal voters, Coburn must have gained a bit. How much? Well SoonerPolitics tackles that one too:
We’ll set ideology at the levels expressed in the poll last week: 12% liberal, 31% moderate, 50% conservative. Then we’ll reweight the results of this week’s poll to reflect last week’s ideology. The following results come out:
Actual (Reweight)
President
Bush-R 50 (54)
Kerry-D 38 (38)
US Senator
Coburn-R 38 (42)
Carson-D 40 (35)
Those Presidential numbers look much closer to reality to me; however, Carson polling lower than Kerry seems suspect. I don't believe the reweight numbers are accurate, but I am sceptical of the original polling numbers as well. We'll see if the level of liberalism in the next sample (next week) is similar to Oct 11 or Oct 4. But there is a possibility that the Meet The Press debate, the Presidential debates (Oklahomans thought Bush won both), and a backlash against the tornado attack ads has push Carson back down to his base 35 and opened the door for Coburn to step through.
Finally, SoonerPolitics(should I charge them for this free advertising) has Chris Wilson of Wilson Research Strategies explain their fixed party registration methodology:
This methodology was first proposed by Green and Gerber in 2003 as a better predictor of likelihood to vote than Random Digit Dialing among registered voters. The methodology was later studied and endorsed by Christopher B. Mann, Yale University professor and can be read about in the attached paper, “Improving Pre-Election Forecasts from Registration Based Sampling: Using Voter Registration Data to Predict Partisan Vote Intention and to Allocate Undecided Voters.”
To read Mann’s paper, click here.
(Cross-posted at Okie's Corner)