From Nate Silver’s FiveThirtyEight post, “The White House is Not a Metronome“:
In a series of articles last week, the writer Megan McArdle asserted that Republicans have about a 75 percent chance of winning the White House in 2016. “Mostly, the White House flips back and forth like a metronome,” she wrote. “Voters just get tired after eight years.”
As other commentators, like Henry Farrell, have pointed out, one can find almost any pattern in presidential results if one looks hard enough. By manipulating the definition of incumbency, the time frame that one examines or the measure of success (e.g., the popular vote or the Electoral College), or by selectively excluding “outliers” or exceptional cases, the potential for cherry-picking and overfitting is high. (In layman’s terms, an overfit statistical model is one that is engineered to match idiosyncratic circumstances in past data, but which is not an accurate picture and makes poor predictions as a result.)
…as Jonathan Bernstein writes, looking at wins and losses in such a binary way is probably not the best way to evaluate the evidence. Many United States elections, as in 2000 and 1960, have essentially been ties, where the most minor variations in the flow of the campaign might have changed the winner of the Electoral College or popular vote. With this in mind, it is better to examine margins of victory.
Incidentally, this is close to a universal principle of statistical analysis. It’s almost always more robust to evaluate the margin by which a given outcome occurs than to look at the variable as black or white, win or loss, hit or miss, on or off.
In conclusion, says Silver, after much wonky analysis:
…Popular presidents might seek and win another nomination, while unpopular ones might retire or be rejected by their party. It could be that voters react differently to parties seeking third terms in the era of the 22nd Amendment than they did in the years before it, and that the incumbent party’s batting average will wind up being meaningfully higher or lower than 50 percent in the long run. But we have little empirical evidence for this yet.
Instead, some humility is called for when interpreting the evidence. On the eve of an election — when the polls have become very reliable, and we know the identities of the candidates and something about the state of the economy and the mood of the country — it is possible to make relatively bold and precise forecasts about the outcome. But none of this applies three and a half years in advance.
Add to all that the fact that we have entered an uncharted reality, in which one party is increasingly defined by unhinged political nihilism, and it becomes even more clear that all bets should be off — until a couple of days before election day.