Reply #99: My son

[Febble]

when he was eight, wrote a poem with the superb title:

"Don't let your caution waver" - and I don't. It's part of being a scientist. Pollsters are scientists too.

So I will defend Mark Blumenthal, whom I count as a friend, against the charge of putting lipstick on pig. What scientists have to do is to check all livestock for authenticity, regardless of the cosmetics used. I would agree that the discrepancy between poll and count in Ohio 2005 needs investigating. However, investigating any discrepancy between two findings involves the investigation of both findings. Blumenthal is a pollster. He's done a good analysis on the half of the problem he knows about, and he makes good points, as do the posters in this thread:

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=401315&mesg_id=401509

But just because it is perfectly possible to propose plausible hypotheses as to why the polls might have been wrong doesn't mean it is not equally possible to hypothesise why the count might have been, as you have done. Both hypotheses need to be tested.

All I would argue is that it can be counter-productive to regard a poll-count discrepancy as prima-facie evidence of fraud when there are so many ways in which polls can give you wrong answers. All social scientists know this, which is why failing to take into account perfectly valid arguments as to why a poll might have given a discrepant result risks alienating those who might (as I do) argue that the fraud charge should also be seriously investigated.

As well, of course, as arguing the case that no electoral system in a democracy should allow any room for doubt about the result (ours doesn't, which is why we know that polls are often wrong).

Which brings me back to my bottom line (above, if a bottom can be above....)