|
Even sampling an urban area might be trickier than one might think. The pollsters define "geostrata" in each state, intended to pick up on salient regularities such as the one you mention (I know nothing about WI, just taking your word for it!), and then use a sampling method designed to cover the partisanship spectrum more or less uniformly. Their models look at deviations from past performance. So if some region or 'type' of precinct is out of trend relative to the others, they may not immediately get a good measure of what is going on, but they at least have some early indication that something is.
On average in 2004, their precinct samples were reasonably accurate, but from state to state, there were pretty big discrepancies in both directions -- i.e., the vote counts in the sampled precincts (plus attempts to model absentee vote, etc.) didn't yield good predictions of the state totals. (Apparently the biggest discrepancy was in Kansas: paging Thomas Frank!)* Whether those discrepancies were driven by rural areas, urban areas, or neither in particular, I don't know. I just don't want people to think that there is some slam-dunk explanation, based on the exit pollsters stupidly undersampling rural areas and not compensating for it: that's not how the exit polls work.
* Here I'm drawing on pp. 29-30 of the evaluation report.
|
Printer-friendly format
Email this thread to a friend
Bookmark this thread
(1000+ posts)
Send PM |
Profile |
Ignore
