Is there any reason for me to believe the polls when they showed him losing Michigan Wisconsin
He barely won those states...? Polls have margins of error: in many cases, +/- 3 percentage points. That means a poll that shows a tie could mean the result is anywhere from one candidate being up by 3 to that same candidate being down by 3; the margin could be off by as much as 6 points.
| State | Poll Avg | Result | Diff |
| IA | -2.9 | -9.4 | 6.5 |
| OH | -1.9 | -8.1 | 6.2 |
| WI | +5.3 | -0.8 | 6.1 |
| WI | +5.3 | +0.8 | 6.1 |
| ME | +7.5 | +3.0 | 4.5 |
| MI | +4.2 | -0.2 | 4.4 |
| NC | +0.7 | -3.7 | 4.4 |
| PA | +3.7 | -0.7 | 4.4 |
| MN | +5.8 | +1.5 | 4.3 |
| VA | +5.5 | +5.3 | 0.2 |
| CO | +4.1 | +4.9 | 0.8 |
As you can see, every state according to the 538 model was either right on the line of that 6-point variance or well within it. So the notion that polling was a "flop" in 2016 really is exaggerated: basically, the polls were off by as much as their disclaimers always say they can be.
You're not going to have a situation where the polls are showing Trump's approval underwater by 20 points when it is really A-OK in reality. There's room for variance, but it could just as easily tilt in the direction you don't want it to tilt as it could in the other direction.
Looking at that chart had the opposite effect than your argument goes. When the polling error isn't random but systemic or correlated with other factors, that's a polling error. The polls were off in the same direction in 9 out of 10 cases. Even worse, these are polling
averages, and the whole point of aggregating polls like 538 does is to reduce random error. Clearly, the polling was off in crucial Midwestern swing states with lots of working class whites. Pollsters tend to overpoll middle class respondents. That's also why they slightly underestimated Obama in 2012.
