Political science models of presidential elections can bite my shiny metal ass. This statement may come as a bit of a shock to regular readers who know that I have my own political science model of presidential elections that I think is the only thing that anyone should pay attention to. I’ll get to my model in a moment, but let me explain my current annoyance. I just read Alex Roarty’s new National Journal article, Predictive Intelligence. It is subtitled, “Think Hillary Clinton is likely to win? Think again.” I’m fine with that. I’ve been saying that under the right economic conditions, Louie Gohmert could be the next president of the United States. Anyone claiming that the Democrats are sure to win the presidency in 2016 — much less Clinton — is an idiot.

The problem with Roarty’s article is that it focuses on the absolutely stupidest predictive index: the number of years that a party has been in office. Why? Well, because this far ahead of the election, there really isn’t anything else you can base a prediction on. It’s kind of like the drunk looking for his keys under the street the lamp. “Where did you drop them?” “Oh, a half block south.” “Why are you looking here?” “Because the light’s better!” This is just madness.

Let’s look at just what a great predictor length of time in office is. First, there is Franklin Delano Roosevelt followed by Harry Truman. Together these presidents managed to win the White House five consecutive terms. When the Democrats lost, it was to war hero Dwight Eisenhower against anemic uncharismatic Adlai Stevenson. After two terms with Eisenhower, the Republicans only barely lost to Kennedy in the the closest presidential race in history: 49.7%-49.6%. But okay, that’s old news.

What about George Bush winning in 1988, giving the Republicans three straight terms? What about Al Gore winning in 2000, giving the Democrats three straight terms? Looking at the number of terms that one party holds the White House, I see randomness. And this is the problem with political science models. Designers throw in every possible parameter they can think of, do a linear regression and find out what fits bests. This is very lazy — most especially because there just haven’t been that many presidential elections on which to perform the regression.

Consider Alan Abramowitz’s “Time for Change” model that uses three parameters: “the incumbent’s approval rating, economic growth in the second quarter of the election year, and the number of terms the candidate’s party has held the White House.” Well, that strikes me as a little to a lot stupid. The current president’s approval rating and the state of the economy are themselves correlated. And the number of terms the party has been in office is probably just a fluke. Take a coin and flip it a hundred times and perform a correlation. You will find lots of correlations. And Abramowitz doesn’t have close to a hundred coin flips!

My model is simplicity itself: it is based entirely on unemployment. But it only works from the early 1970s onward. But unlike most of the political science models, mine actually has a good justification for that. Before the 1970s, economic gains were reasonably well shared. What’s more, the working class didn’t spend their whole lives worrying about losing their jobs. But since the 1970s, economics has been the defining issue for Americans. The fact that it has been used against them (as during the Reagan years) hardly matters.

I’ve never seen a political science model that was based on unemployment. They prefer the indirect measures of GDP and GNP. This is interesting because everyone knows that the economy can be doing well while the workers can be doing very badly. Just look at the last six years. The stock market is doing well; economic growth is reasonable; corporate profits are sky high. But the employment to population ration of prime age workers is still quite low compared to the anemic economy of the George W Bush years.

My model looks at the unemployment rate for the first nine months of the election year. If the trend is positive, then the candidate of the party currently in the White House wins. If the trend is negative, then the candidate of the party currently in the White House loses. Very simple. And it doesn’t just predict the win; it predicts the magnitude (although not as well). It even shows that the 2000 election between Bush and Gore was almost a dead heat, with Gore having a slight advantage.

The most common way that political science models look at the economy is by looking at the GDP growth in the three full quarters leading up to the election. I think this is a mistake. Voters don’t care what’s going on with the economy’s GDP. They notice if people are getting laid off or if people without jobs are getting hired. Now clearly, the rate of GDP growth is correlated with the growth of employment. But it isn’t correlated that well! So why use it? I think it is because the concerns of the power elite poison everyone’s thinking.

Regardless, it is clear that in a close election, candidates matter. But in 1980 and 2008, the economics were such that the party in the White House was going to lose. But in 1976, Ford had a slight advantage. If it hadn’t been for Watergate, I suspect he would have won. Gore could have lost in 2000, but he was no worse a candidate than Bush was. And Clinton was almost certain to win in 1992, but without Ross Perot running, the race would have been a whole lot closer.

It is madness to try to predict the 2016 election at this point. Even Alan Abramowitz’s model requires waiting until the party conventions before he can make a prediction. The best we can say now is what I always say: if the economy continues to grow jobs through most of 2016, the Democratic candidate will win; if the economy starts to lose jobs, the Republican candidate will win. And not to put too fine a point on it: Democrats should learn from Republicans in the past and nominate candidates they really like rather than candidates they think are “electable.” Because it was the economy and not great political skill that made Ronald Reagan and Barack Obama presidents of the United States.

Afterword

None of this should be taken as a criticism of Alan Abramowitz or any other political scientist. They are doing careful work trying to figure things out about the nature of politics. But there is no doubt that these models are underdetermined: there are so many variable and so little data that an endless number of solutions would fall within the error bounds. My model is very simple, but I haven’t seen anything to indicate that it is any worse than any other more complex model. And my simple model goes along with what I take to be the psychology of voters. Abramowitz might be onto something about people getting tired of the same old party. But I suspect that was more true in the 1960s and 1970s than it is now in our highly polarized political environment. The main thing is that people like Abramowitz are professionals and I’m just a tinkerer. I will be the first to admit that their models predict past elections better than my model does. But here’s the ultimate question: do they have a model that uses a single parameter to predict elections that works any better? And for the record: I’m sure that many people before me have done exactly what I’m doing now. I am simply skeptical that anyone has improved on this idea.