Reply #147: A couple points...

[flintdem]

I did overstate my argument in my first post. Exit polls CAN be used to determine winners but if we follow the NEP standard there are quite a few elections where no winner can be determined from the exit poll. That I think is part of the misconception of the public and the media as to what exit polls can do. They CAN and they CANNOT determine winners and the closeness of the election appears to the significant dividing factor.

In terms of what the data can and cannot "prove" otherwise... In my experience if a colleague or reviewer tells me my data is not "suitable" for the analysis I am doing -that is usually a kiss of death for the article (no R&R). If they are not "well-suited" I probably have an opportunity to defend the validity of my usage. But now I have to do more than just show my results, I have to convince others that my usage of the data is valid and that I am measuring what I claim to be measuring- a much more difficult job (which hasn't been done with any of the analyzes here- lots of results, no validity proofs).

To claim finding a fraud of 2% in precincts with an N of 50 strains credulity. Regardless of precinct size your odds are 1 in 50 of interviewing someone whose Kerry vote was electronically changed. Match that with the idea that "bush voter aversion to pollsters" only requires an average of one aversion per precinct adding one more Kerry respondent to produce a 2% shift, and you cannot prove one hypothesis (aversion or fraud) over the other with this data (a recount with a paper trail could though prove fraud, but not a data analysis- a recount showing a pattern would be proof). Jump the precinct N to 200 respondents, add more than one interviewer to insure wire to wire coverage and actually have supervisors check the distance between the pollsters and the exits and in the aggregate across the nation or state you can produce a convincing design and possible case for fraud.

edited for sloppy grammer