Because this is new-age stuff that has no real scientific basis?
I agree that the label New Age works well here, but it is very scientific.
Their hypothesis seems to be that by monitoring internet activity they can measure stochastic correlations and use these to assign a probability. Their experiment seems simple enough, although details are not provided in the linked article. Their conclusion, at least in the UK vote to exit the European Union, was that the yes vote was more likely. Here, they are not making a prediction. They are simply saying that in the "battleground" states Trump's index is slightly higher than Clinton's. Despite the somewhat misleading title "Trump's election to lose" the study seems rigorous enough to me, at least on the surface. It would be nice if they had shared their algorithm, but they may want to capitalize on it later if it turns out to have consistent predictive value, so you can't blame them for their secrecy.
Edit: That said, I think it is still Clinton's election to lose. Monday's debate will be important.