If I were polling in Mich:
One of the final question ... for whom did you vote in ‘12 (Trump or Hillary)?
Then weight by that horserace. So it’s 50/50. It’s a way of compensating for hidden Trumps and other sampling errors.
Can do similar in other states.
The Literary Digest of 1936 actually had the foresight to do that - but they did nothing with the information.
They reported their raw estimate, which was Landon 57-43.
Had they weighted their estimates by the 1932 response, they would have improved their estimates considerably but they still would have been wrong.
I’m not sure why.
One guess I had - did some states have poll taxes that were in effect in 1932 but waived in 1936?