|
I agree that we do need to verify where the exit poll data came from. Mitofsky/Edison have been refusing to make their complete data set public and we are going on "leaked" numbers for the most part. I am very skeptical of some reports of "final" exit poll data since they have been tainted by applying some kind of factor to more closely match "actual" vote data. We will have to wait until Sunday when TIA returns so that we can have that discussion.
As far as the MOE and SD, TIA gave an example calculation in the original post. However, look at the chart above. I do not rely on MOE or SD in the chart - it is simply the difference between the exit poll and the "final vote". In a perfect exit poll, all differences would be zero. In a more real-life case, the differences should be small and clustered somewhat evenly around zero on both positive (favor Kerry) and negative (favor Bush) sides. What we see is a whopping disparity between how many shifts to favor Bush versus how many shifts to favor Kerry (41 to 9). Also, we see that Bush has many more states shifting to him in more dramatic fashion (let's say more than 2 points).
As for the randomness of the initial polling data, again, we will have to try to get information from Mitofsky/Edison. However, I do not buy any argument that there is enough inherent bias in the exit polling techniques to cause this "red shift".
As a statistician, can you suggest other tests that would be appropriate to apply in this case? I've looked at skew and kurtosis, but they didn't seem quite right for this case.
|