Thursday, November 10, 2005

OK, so who was right?

In an earlier post, I wrote about the potential differences in polling practices. The risks of variant polls is misinformation in the political system. In theory, widely disparate polls could also influence who comes to the polls. One of the major pollsters had the integrity to publish a temperature assessment of how well each of them did. Survey USA did a compilation of the accuracy of the polls in the Special Election. Their totals were as follows -

For the big 4 propositions (73-77) the margin of error was was pretty different.

POLLSTER AVG. ERROR
Field 3.36
SurveyUSA 5.32
PPIC 5.76
Stanford/KN 5.96
Polimetrix 6.24
LA Times 7.44

The numbers varied a bit when comparing the average margin of error by the number of propositions polled.
POLLSTER ERROR (Number of propositions polled)
SurveyUSA 5.32 (5)
Field 6.05 (8)
Polimetrix 7.20 (7)
LA Times 7.85 (8)
Stanford/KN 8.80 (8)
PPIC 9.57 (7)

When one looks at the 3 outlyer propositions (78-80) one would think that accuracy would be improved. If you look at the Secretary of State's election maps (the red/blue splits) you find that the bottom three had a pretty resounding thumping. Also, the real interest and financing came to the first five (73-77). But that assumption is not necessarily true.

The results raise two additional questions - which are fundamentally more interesting than the level of error (although that is pretty interesting). First, we do not know the directions of the error (although with the newcomer Stanford poll there was a relatively consistent error on the side of overestimating support). My suspicion is that two of the leading pollsters were pretty consistently off in estimating the conservative vote. That could be a function of two things either sample error or turnout - there is considerable evidence that conservatives in Southern California did not vote in normal numbers. The second issue is more important. With more than 40% of the voters casting absentee ballots we do not know whether the polls in the last two weeks had any influence on the final results. Both issues could be a good topic for a student of politics to write on in a dissertation or for someone with a little time and a calculator to work through.

No comments: