The polls before the 2016 election didn’t screen accurately for likely voters. This is an intractable problem given that any correction based on this year’s errors may not apply to a future election. Even worse are misleading probabilities of victory for candidates created by the compilers of polls.
This year, these assessments ranged from about a 71-99 percent likely victory for Hillary Clinton. Such probabilities, however, exhibit the fallacy of false precision because they are no more accurate than the underlying polls.
It is past time to stop using polls as predictors and to turn to historically based models.
– Allan Lichtman, Distinguished Professor of History, American University