Site Meter

Wednesday, October 06, 2010

Final Battle for the Bride of the Gallup anomaly.

I gulped when I say the TPM smoothed congressional generic ballot. The Republicans have pulled ahead again. This happened because Gallup switched on their likely voter filter. I warned in advance not to take the number too seriously.

My I've written a lot on the Gallup anomaly.

I believe that Gallup switched on the filter roughly when the biases with registered and likely voters are roughly equal, meaning both are biased in opposite directions. In particular, the Gallup likely voter filter has very different properties in late October and August (would have that is, I don't think they even ask the questions in August). One of their 7 questions is "do you know where your polling place is ?" The answer clearly conveys different information if the question is asked October 30 or October 6th. Someone who still doesn't know then is someone who is not likely to actually vote. Someone who doesn't know now is just someone who hasn't voted there before (that is another of the 7 questions).

Pollsters are judged based only on the last poll which they report. Gallups last poll is a very good predictor of electoral outcomes. Is Gallups early October poll a good predictor of Gallup's last poll ? I'm off to check. It is now 7:52 AM EST.

I have been over at www.pollingreport.com. The few numbers I looked up provide no support to my claim. I am looking at the Gallup/USA Today poll because it's what I found.

Gallup D-R

Year Last, first Oct, date Registered last first Oct
2008 12 6 10/10-12/08 15 ---
2006 7 13 10/6-8/06 11 23
(damn last in September was tied)
2004 1 -3 10/22-24 4 2
2002 -6 1 10/03-6 5 5
(ruined everything)

The likely-registered gap improves on average 0.33% for Democrats and the last poll is 0.75% better for the Republicans than the first.

I still believe what I claimed.

No comments: