Rich Ulrich
2016-11-05 01:14:50 UTC
It was 3 or 4 weeks ago that the NY Times had a couple of articles
about the science or craft of polling.
One of them reported an "experiment" that showed the variability
of poll results due to "method." That is -- Someone handed to
six different polling agencies the raw data on a Trump v Clinton
poll for one state, N about 1000.
Everyone used /some/ version of weighting, rather than report
the simple numbers. What was available included, I think, age,
sex, registration, and something about likelihood to vote; and
probably some information about the state.
The pollsters, all using the same data, reported conclusions
that ranged from a 3 point advantage for Clinton to a 1 point
advantage for Trump.
In the 2012 election, I wondered whether "voter suppression"
by the GOP would be unmeasured by the polls. The pollsters,
generally, were very successful. A GOP pol in my state, Pa., did
claim that they had cut back Obama's margin by several points.
Overall, the Obama victory was credited, my many people, to the
effectiveness of the Obama ground-campaign in getting out the vote.
But, I never saw any comment on this: Did the polls /attempt/ to
predict suppression and get-out-the-vote; or did these two contrary
factors just happen to offset one another?... to the effect that,
the pre-election polls were accidentally more accurate than they
deserved to be.
For this election, I'm again wondering about the suppression
factor, and whether any pollsters feed that into predictions.
For this election, I also wonder whether the pollsters have tried
to measure and account for the fraction of voters who have
changed their party affiliation during the recent campaign -- that
is, the pollsters ask, Which party do you identify with? The high
level of acrimony on both sides might affect how many people,
today, claim to be Democrats or Republicans; and I have not
seen anything in months that reports on the changes in self-
identifications. But that should effect the weights applied to
get their estimates.
about the science or craft of polling.
One of them reported an "experiment" that showed the variability
of poll results due to "method." That is -- Someone handed to
six different polling agencies the raw data on a Trump v Clinton
poll for one state, N about 1000.
Everyone used /some/ version of weighting, rather than report
the simple numbers. What was available included, I think, age,
sex, registration, and something about likelihood to vote; and
probably some information about the state.
The pollsters, all using the same data, reported conclusions
that ranged from a 3 point advantage for Clinton to a 1 point
advantage for Trump.
In the 2012 election, I wondered whether "voter suppression"
by the GOP would be unmeasured by the polls. The pollsters,
generally, were very successful. A GOP pol in my state, Pa., did
claim that they had cut back Obama's margin by several points.
Overall, the Obama victory was credited, my many people, to the
effectiveness of the Obama ground-campaign in getting out the vote.
But, I never saw any comment on this: Did the polls /attempt/ to
predict suppression and get-out-the-vote; or did these two contrary
factors just happen to offset one another?... to the effect that,
the pre-election polls were accidentally more accurate than they
deserved to be.
For this election, I'm again wondering about the suppression
factor, and whether any pollsters feed that into predictions.
For this election, I also wonder whether the pollsters have tried
to measure and account for the fraction of voters who have
changed their party affiliation during the recent campaign -- that
is, the pollsters ask, Which party do you identify with? The high
level of acrimony on both sides might affect how many people,
today, claim to be Democrats or Republicans; and I have not
seen anything in months that reports on the changes in self-
identifications. But that should effect the weights applied to
get their estimates.
--
Rich Ulrich
Rich Ulrich