87% to 99%?? That's a rather large margin of error. This is a long-tail election, and - as at least one other pollster has pointed out - in those instances many of the assumptions on which a model is built no longer hold. Garbage in/garbage out. The more fruitful approach would be to look at least at that 1960 election and see how it compares to this election. That was the election in which incumbent VP Richard Nixon ran against JFK. That was the election in which the first televised general debate between the two candidates for president was held. That debate was a long-tail event because it involved television for the first time, an unfamiliar medium that Kennedy and his team figured out how to exploit and which Nixon and his team did not. As a result, Nixon tanked in those debates, and that was seen by millions of people who would otherwise probably not seen either of the candidates in person. There's a bit of a discussion on the debate here:
http://www.history.com/topics/us-presidents/kennedy-nixon-debatesThis is a similar situation; despite the fact that social media has been around for at least 2 presidential elections, how to use it - and how to manage negative exposure on it - is still being worked out and, furthermore, the republicans - including Trump (giving him the benefit of the doubt) - are woefully behind the democrats in understanding the media, how it works, and how it can be used/managed.
Social media certainly played a role in the 2008 and 2012 elections, but this time around both candidates have gotten onto social media in a way that earlier republican candidates have not. As a result, it seems to me that social media's role will be crucial to this election in a way it has not been in earlier elections. And, in this regard, it seems to me that the democrats are still running rings around Trump because Trump's principal foray into social media appears to be his own thumbs on his own phone; he does not appear to have a solid or substantial social media team out there running a social media campaign for him. The democrats do.
Then there is the partisanship of the mass media. That partisanship was not nearly as severe, nor as one-sided, in the earlier elections as it has been since the 2000 election; that is when it started to get bad and it's just gotten worse from there.
Then there is the fact that Trump is - as his supporters like to crow - a political neophyte who has never run for office, nor had to fulfill the duties of an office, and who is ostentatiously running against politicians as such. In all of the earlier elections, both candidates had political experience; they had run for at least one office and they had won at least one office.
Then, there is the fact that Trump has spent more of his time attacking republicans than he has spent attacking Clinton or the democrats. That was never the case in any of those earlier elections.
Finally, there is the fact that this election has been turned into an intensely personal campaign in which the personalities and personal idiosyncrasies and peccadilloes of the candidates are more important - in terms of the time devoted to them - than are the actual issues. Compounding this is the fact that there are very serious issues, with dramatically different proposed solutions from each party, and with a lot of people very interested in having those issues resolved.
All in all, this election is substantially different from all of the elections on which the model has been tested and, given that the model is built on the results of those elections and therefore necessarily builds in assumptions based on those statistics, the model almost certainly has substantially less predictive value in this election than it may have had in earlier elections.