This.
There were some models that gave Hillary a 7% advantage over trump but reporting about a model like that will give you in no way the same amount of attention as a model that has a clear victor.
You can see the same thing happening in the """"science"""" part of many newspapers, where correlation and causation are happily mixed and statistical confidence is thrown out the window for sensationalism.
Personally I'm also speculating that predictive models fail to factor in the levels of enthusiasm present on either side.
The decision to vote for a candidate is very different if you have to actually leave your house and stand in line instead of just being asked which candidate you favor with no further personal inconvenience.
I think Trump voters were on average more invested in their candidate therefore more people actually showed up on election day.
I imagine it's kind of like with caring about the environment.
If you ask people randomly if they care about the environment many will say they do.
However, in situations where people are actually making decisions you'll find that even small personal inconveniences outweigh the abstract damage damage done to the environment.