Reply #79: Well, to save winged Mercury trouble

[Febble]

I'll post this response here:

I'm going offline for a bit as I HAVE to finish my dissertation, and it still needs re-ordering.

I did, in fact, look carefully at your model - what I hadn't looked in detail was your 1st November post. I knew you had a high probability for Kerry winning the EV. I hadn't appreciated that you had extended it to the popular vote.

Your model does assume that the polls were random samples (Wang's did not, although my understanding is that he assumed that overall there was no net bias). This is the problem I have with it. I did understand your assumption about undecideds - that was not the problem. Although I think that you should also have put a probability value on your assumptions being correct, as Wang did. But your model allows us to put in our own assumptions, which is fair enough. But it still assumes that the only error in the polls was sampling error. It does not allow the possiblity for bias in the polls. The point of my "mantra" is simply that your probability estimates depend on the assumption of random sampling, and we simply cannot make that assumption when it comes to polls.

That is one respect in which we differ. The other is that I do not accept that what has happened before will necessarily happen again - or that trends will continue as they are going. It's why I took issue with your 4th degree polynomial, although I completely agree that the approval numbers look terrible for Bush. But my experience with trends is that they often change direction. So I did not put a lot of certainty in the incumbent rule, nor in the undecideds-break-for-the-challenger rule. Although I desperately hoped they would hold.

As for the Gore-Bush thing - as I said in the post, and I've said many times before, given what we KNOW about the way people report past vote, I do not see this as a clincher. Yes, those proportions are impossible IF people reported their 2000 vote correctly. But we know they do not, and that they often misreport having voted for the incumbent. This has been shown in longitudinal studies where the pollsters KNOW how their respondents said they in a given year, yet the SAME voters report having voted for someone different when asked several years later, in both the UK and the US. So we know we cannot rely on "reported vote" data.

Anyway, happy holidays to you and to all! See you after I've submitted my dissertation.

Lizzie

PS: link to OTOH post re recalled vote:

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=390193&mesg_id=390341