Did you read the article. More importantly, did you understand it?
His "models" are based upon two categories -- whether or not the outgoing president served an eight year term and strength of each candidate in primary elections. Trump pretty much dominated the GOP primary and Clinton did not. Past elections give the greater advantage to Trump. Also, we are exiting an 8-year presidential term. In other elections, the candidate from the other party usually wins in these "turnover" elections, so again, advantage Trump. Taken together, Trump wins 87% of the time when comparing historical elections back to 1912. Trump should win if history is a guide. What's missing in Norpoth's model is any capacity accommodate for the unusual aspects of this election, such as the dismal performance by Trump in the debates and the scandals, real or otherwise, of both candidates. Also teensy sample size. Also the model is based upon empirical evidence and not on fundamental principles. In other words, it's a pretty flimsy model without a lot of theory to back it up.
Not rejecting his model. Its just that public opinion polls and models from them are better at accommodating new information than Norpoth's "Primary Model" and they point overwhelmingly toward Clinton. What I gather from Norpoth's modelling results isn't that Trump will win in November but that Trump had the upper hand early in the general election cycle. He had an advantage and he blew it.