> In 2016 Trump polled way worse than he performed in key states.
That's not really true. The state-level polls were generally within normal polling error, although there was a systematic error due to educational attainment. In 2020, more pollsters have incorporated educational attainment into their weighting, so this source of error should be reduced.
People are also more paranoid today than in 2016. If the communists take over, you do not want to be on some voted—nazi list.