Analysis of election results has just begun, but crude signs of foul play are blatantly in evidence. I'm reporting from Boston, working with Jon Simon who pioneered the idea of exit poll analysis with the unlikely Bush upset of 2004.
Briefly, the method is to capture exit poll results reported by the media immediately after voting has ended. At this point, the reported exit polls have not yet been mixed and massaged with official results, because there are no official results yet reported.
Comparing these virgin exit polls with the final official tally, we can see whether the official results match the expectation. Simon coined the term "red shift" to describe official results that are more Republican than the exit polls, and "blue shift" when the official results are anomalously Democratic.
If the exit polls are accurate and the official count honest, we expect an equal number of red shifts and blue shifts. The numbers should be small, and within the pollsters' margin of error.
Here's what we find for yesterday's results. There were 18 states in which the Senate races were exit polled. 15 of these had "red shifts" and only 2 had "blue shifts". (In one race, the result was unshifted, or zero.)
11 of the red shifts were statistically significant. The only significant blue shift was the Senate race of Harry Reid in Nevada. In other words, Reid did better than the exit polls would have led us to expect, while every other Democrat (except Ellsworth in Indiana) did worse.
The overall probability of this situation occurring by chance was less than 1 in a million.
The average red shift for all Senate races was 3%. This is what we would expect if a "programming glitch" inside the touch-screen machines were taking one Democratic vote out of every 30 and flipping it to Republican. As Simon explained in interviews last week, polling methodology has changed in the last few years to match up more closely with the (suspect) election results. This means that the exit polls themselves might be redshifted , and, if so, the tampering may be larger than what we are able to detect with this method.
The data for governors' races shows a similar pattern. We were able to download exit polls from 12 states, and 10 of these were red-shifted. 8 of the red shifts were statistically significant. The average shift was just under 3%, and the overall probability of this occurring by chance was less than 1 in a million.
California showed no red shift in either Senate or Governors' races. We should all be writing thank-you notes to California's Secretary of State, Debra Bowen.
Of course, these calculations are tentative at this point, but I thought OEN readers might want to see them when they're hot from the spreadsheet.
I hope to expand this column later in the day, with reports on the US House.
