this sampling narrative that keeps coming is certainly interesting:
http://www.weeklystandard.com/blogs/morning-jay-are-polls-tilted-toward-obama_653067.html?page=1
I mean unskewed seems to be completely off its rocker, but the sampling breakdown/assumptionms stuff is curious in any event. I definitely is the component that makes polling a nightmare I imagine.
It would seems the LV measure should simply be generated fresh, and the sampling weights should be determined empirically rather than historically generated.
If one rings enough phones one should have a sample that defines D/R/I split and the enthusiam/LV likelyhood of each component and then sample accordingly.
It seems that a D/R/I weighting is being applied based on previous elections rather than what people say today.
Or not and it just a smoke screen. I can't tell from the articles whether the sample weighting is deliberate up front or simply a consequence of what they found when polling.
But is does look bimodal.
One can find some comfort in the notion that even with the more R weighting(Rassmussen e.g.) both the national and the swings are blue, just that it is much tighter when that sampling method is used.
| 