Site Meter

Saturday, October 30, 2004

Strategic Polls from Strategic Vision ?

The Strategic Vision anomaly in state polls is much less famous than the Gallup anomaly in national polls, but it is critical to the meta meta analysis of state polls. Strategic Vision polls for Republican candidates. Partisan pollsters are suspected of getting polls favorable to their party. for this reason Strategic Vision polls are not included at however they are available at the strategic vision web page and at the excellent race2000 which is Sam Wang's spot for hot hot fresh state polls.

I checked using polls from the battleground staes Iowa, Michigan, Minnesota, New Jersey, Ohio, Pennsylvania and Wisconsin with sample periods ending on or after October 20. This is a tiny subset of the available data with13 strategic vision polls. On average Strategic Vision polls give Bush minus Kerry 2.64% higher than the average of other polls for the same state. The standard error due to sampling of the average Strategic Vision poll is 0.83% the standard error of the average state average of other polls (with perfect correlation within states and 0 correlation across) is 0.40. This means that the standard error of the difference Strategic Vision minus average is 1.05 and the actual difference is 2.52 standard errors.

This is proof at standard significance levels that Strategic Vision is indeed biased for Bush or that the average of other polls is biased for Kerry (or both). As a result, I consider it wise to exclude Strategic Vision from my calculations, whcih I have been doing all along.

No comments: