First, as I recently explained at The Christian Science Monitor, any effort to model presidential nomination races or general elections suffers from the small-n problem: there just are not enough cases for making sound projections.
Second, to the extent that models are useful, they depend on numbers that are not yet available. For instance, we do not know what the economy will look like several months from now.
Third, the Trump phenomenon is hard to gauge. Are online surveys overstating his support, or are live-interviewer surveys understating it? And will his supporters actually show up in primaries and caucuses? We have no way of knowing until people actually vote.
Fourth, as Michael Barone writes:
Besides the known unknowns, what unknown unknowns lie ahead? The Las Vegas debate centered on terrorism, which probably would not have been the case but for the San Bernardino attack 13 days earlier. That probably helped some candidates, notably Christie, more than others.
That was not the first time that this contest was affected by what the British Prime Minister Harold Macmillan told a junior colleague determines history: "Events, dear boy, events." And it probably won't be the last.
None of the Republican or Democratic candidates in 2008 expected that financial markets would collapse that fall, and they might have campaigned differently had they known. We can see the calendar ahead, but not everything that will appear on the pages.
