|
Edited on Mon Dec-13-04 10:54 AM by davidgmills
On Stones-Cry-Out website this weekend Rick made it clear to me that the margin of error for Florida was 1.8% if you don't account for THE DESIGN DEFECT CAUSED BY CLUSTERING.
You apparently agree with him on that point and I presume he would agree with you on the other percentages you calculated.
But he insists that the Florida margin of error with "design defect" for clustering is 3.3% not 1.8% and that Ohio and Pennsylvania were 3.5%. Some states, due to low sampling, had margins of error of 4 or 5%.
So the question, to Rick and others like him (Mystery Pollster, Mytofsky, macdonald -- let's call them the "clusterists") at least, is how many of Bush's states are outside these margins of error not how many are outside a 2.5% margin of error.
My point to him was that if mathemeticians can not agree on what the effect of clustering is and how best to calculate it, let's have another National exit poll done right now (a random one to avoid the problems of clustering) of between 20,000 to 40,000 Americans (whatever number it takes to get a random sample that is highly accurate) done by reputable pollsters like Zogby (Democrat), Gallup (Republican) and Harris (middle of the road, I think). Have each pollster poll about 1/3 of the sample and make the results public immediately for academecians everywhere (even outside the US) to have a fair shot at analysis. There is no rule that says that an exit poll has to be done the day of the election.
Maybe do the same for key swing states, but make sure the sample again includes enough voters to be a highly accurate exit poll (I would hope down to .05% and preferrably less).
Otherwise, I just think we are going to have mathemeticians arguing with each other about the limitations of clustering and sample size. Let's take those arguments off the table.
If we made a demand to have this done, I am sure we could get the funding from places like MoveOn and insist that the Democratic Party help out. We could insist that the Republican Party help out, as a gesture of good will, though I doubt they would.
|