Reply #17: Yes, it should be interesting

[Febble]

Regarding complicity, I will risk my neck here by supporting Mitofsky even on the two points on which you regard him as "complicit":

1) Whether or not people think that exit polls "should" not be adjusted to the count, there was no secrecy about it. The procedure was clearly explained on the E-M web-site in advance:

How are projections made?
Projections are based on models that use votes from three (3) different sources -- exit poll interviews with voters, vote returns as reported by election officials from the sample precincts, and tabulations of votes by county. The models make estimates from all these vote reports. The models also indicate the likely error in the estimates. The best model estimate may be used to make a projection if it passes a series of tests.

and is the way it has been done in previous years (give or take tweaks to the algorithm). The point is that the exit poll is not designed as an audit, or a parallel election - it is designed to allow the networks to "call" the state for one candidate or another before the count is complete. It therefore uses the best data available at any given time - the exit polls at first, then the vote returns from precinct and county as they come in. The prediction is thus continuously updated in line with current data. It's as if the odds on a horse race were continously updated during the race. It aims to predict the counted result, not to check the count is correct. However, it potentially does the latter as a by-product, as the early predictions ARE made solely on the basis of the responses, and in any case the raw data are always released (as they were, as usual, in January). Unusually, this year, E-M also issued an evaluation in which they gave their "3rd call" estimates, so no more captured screen shots are required. We know the data from the exit poll responses alone differed significantly from the the count.

2) which brings me to your second point: the raw data are released, as usual, complete with demographic details of each respondent. You can download them here:

ftp://ftp.icpsr.umich.edu/pub/FastTrack/General_Election_Exit_Polls2004/

What are not released are the precinct identifiers, and they never are. Also, the counted vote for each precinct is also not provided in what is released, as this would allow for precinct identification. The reason for this is very simple. The professional body that covers pollsters, like the professional body that covers my own profession, psychology, has strict ethical rules regarding confidentiality of data. No data may be released in such a way that repondents could be identified. Releasing precinct IDs would violate this.

Now, we know it is possible for "blurred" datasets to be prepared and released, as this was done with the ESI data in Ohio - precinct vote totals were slightly altered, randomly, to prevent precinct identification. It might be worth raising funds for futher datasets to be similarly prepared for further study. If the whole thing came to a legal challenge, presumably the precinct IDs could be subjected to subpoena.

But there is no reason to suppose that Mitofsky, or Joe Lenski for Edison, were "complicit" in "withholding" the "raw" data. The raw data is released. To release the precinct IDs would involve them in violating the ethical guidelines of their own professional body.

The sad thing, I think, is that a lot of the flak that has been fired at Mitofsky is a result of the peculiarities of your electoral system and the way the networks handle it. In the UK we have exit polls, and the "prediction" made by the polls is issued at 10pm when the polling stations close (of course we only have one time zone). A "swingometer" predicts the results based on the exit polls alone, and gives the pundits something to talk about before we get any results. But as soon as the first result comes in a couple of hours later, the swingometer starts to swing wildly, and is widely regarded as good entertainment value. As the night wears on it settles down, and eventually only twitches a bit with the arrival of each straggler. And by morning it's all done (and yes, we do have all hand-counted paper ballots, and yes, it is quick....)

So the exit poll prediction and the predictions made by exit poll plus count are never confused.

Whatever. It happened. And in the light of the Florida shenanigans of 2000, and the insecurity of the system, people were right to be suspicious.

It will certainly be interesting to hear what Mitofsky says.