What does this approach say right now? Assuming that the president’s job approval is unchanged in 2016, and that growth in the second quarter is the same as this year’s (2.3 percent), we would expect Hillary Clinton to receive 48 percent of the vote. Incidentally, this is pretty much what she receives in the RCP average today against Scott Walker, Marco Rubio, Jeb Bush, Rand Paul, Ted Cruz, and Mike Huckabee. To be sure, she leads, but she has universal name recognition.  We should keep 2014 in mind, when incumbent Democrats led early, but Republicans picked up undecided voters and the polls gradually converged on the fundamentals.

Now, does this mean Clinton will lose? Not at all. First, these are the fundamentals for 2015. It is the fundamentals for 2016 that matter. Things may well improve by then (or they could get worse).

Second, and more importantly, we should remember that this is how things usually work. We do have examples where things don’t cancel out, and the elections don’t really follow the factors nicely. We see these imbalances routinely in congressional elections, which is part of why predictive models of Congress are less successful than presidential models.

But we also see it occasionally in presidential elections. The classic example is 2000. The models generally picked Al Gore to win, often by substantial margins. A few had the actual result outside of the “error margin.” Of course, this isn’t what happened.  Some might chalk this up to Ralph Nader’s candidacy, but political science also suggests that a substantial number of Nader’s voters simply would not have shown up in the absence of his candidacy.