And that is why the polls are now rubbish: with response rates that low, the conclusions are unstable and liable to be biased (though you can't tell how) due to some systematic cause for non-response. You can't publish a paper based on a survey in any reputable journal in the social or behavioral sciences with a response rate of under 30%.
Of course, it may be that 35K represents the number of digit strings dialed, and that a lot of them were not phone numbers assigned to any users, in which case we don't really know anything about the response rate other than that it was greater than the 3.2% it would have been had all the numbers dialed been valid phone numbers.
"And that is why the polls are now rubbish: with response rates that low, the conclusions are unstable and liable to be biased (though you can't tell how) due to some systematic cause for non-response. You can't publish a paper based on a survey in any reputable journal in the social or behavioral sciences with a response rate of under 30%."
Unless, of course, you've just dialed every voter in Michigan and completed the population sample. Close!
And answered. Even if all were valid we don’t know how many were answered and how many actual live adults accepted or declined to be polled.