Does the model forecast the high numbers of people who will tell the pollster one thing - then turn around and vote for Trump?
The 2016 polls were basically (marginally) correct. Hillary did get more votes. However, the polls don’t reflect the significant factor....a state by state contest, and the Electoral College mattering in the end. To some degree, the polls aren’t built to work along these lines.
Another issue is that they typically go and poll people who normally vote. There was some evidence in 2016 that in some states....Trump got seldom-voters to come out and vote. Adding to this....black voters that came for Barak Obama’s two elections....didn’t come out that much for Hillary.
In the end, the polls simply couldn’t reflect these values.
Exactly