The polling failure in Michigan, and so on.
From
Rich Ulrich@21:1/5 to
All on Sat Mar 12 16:58:03 2016
Last week, Bernie Sanders won the Democratic primary in
Michigan, upsetting the expectations of the 8 or 10 pollsters
who had surveyed there. I think their reports had showed
Bernie losing by 20%, some more and some less, with the
smallest loss being 7%.
The demographic explanation, which I have not heard applied
to the case, would be that the pollsters
(a) did not weight for "age" -- which has never mattered
that much to them in the past;
(b) were still relying on their telephone land-line polls --
whereas, the younger generation is far more likely than their
elders to not-own land-lines; and
(c) Bernie is capturing up to 85% of the voters under
the age 30, versus Hillary.
Some of the pollsters do attempt to adjust for party affiliation,
race, previous voting experience, and maybe sex. I have not
read of them compensating at all for age, except when offering
warnings that "these results may not be reliable because we
cannot predict very well who, exactly, is going to show up."
- The energetic, youthful supporters of Bernie Sanders are a
bit less likely to vote (it would seem) because they do not have
long habits of voting. So, even if the pollsters try to "adjust",
the results will be shaker than usual.
Polls also have some trouble, I imagine, in the "party affiliation"
aspect of assessing Donald Trump. He has fared better in the
"open primaries" where the voter declares his party when he
shows up; and has fared less well in "closed primaries" where the
registration had to be done some longer time in the past.
Some of the supporters of Trump, who may vote when they can,
are folks who, a year ago, would have identified themselves as
independents or Democrats.
How are these people weighted for "party affiliation" when doing
cross-party polls? How likely are they to show up for their first
time in a Republican primary?
Pollsters have new problems this year, to add to the high
refuse-to-reply rate for the phone polls. I saw that stated as 90%
not long ago -- If it were my poll, and the rate is really that high,
I would be eager to try to find a reliable way overcome the biases
of pollingl on-line.
--
Rich Ulrich
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)