I didn’t comment before on Bricker’s “No President since World War II has won reelection with an unemployment rate above 7.4.” because, frankly, it was just too silly to waste time on. It’s exactly as meaningful as saying that every president elected in a year ending in 0 will die in office. That was true until it stopped being true. It had no predictive power any more that tossing 8 heads in a row predicts what the ninth toss will be.
Since the subject has been raised by others, though, let me throw in my two cents. The predictive power of econometrics has been examined at length by Nate Silver and it’s worth your time to go back on his blog and read those entries. (You can read the books and papers on the subject as well, but I think he does a good and fair job of laying out the arguments in readable fashion.)
The short version is that econometric predictions work until they don’t. Other things are never equal. The farther you go back the less likely you are to have good equivalent numbers. Things other than the economy occasionally become the dominant issue in elections. Third party candidacies throw the final percentages way off the curve. And the way modern elections are run are simply difficult to compare to older ones. So even if econometric formulas were perfect, the data entered into them is always fuzzy.
Yet everyone agrees that economic issues play a major role in deciding the presidency. It is an overwhelming temptation to look for economic numbers or trends that will predict how they will play out. One obvious way of doing this is to look at the indicators and find the ones with the best **postdictive **predictive ability, i.e. the ones that in past elections correlate well with the party that won. Of course if you have all the numbers in the world to play with you will find some that correlate very well. The skill - or art - lies in choosing the ones that truly matter over the ones that are just coincidence.
The next question that needs to be answered is how far back you go to look for postdictive answers. Most everyone stops before they get to Roosevelt, just as Bricker did, for the simple reason that Roosevelt won four landslides with economic numbers that would send anybody else in any other time out on his ass.
That just emphasizes that economic factors play a major role but not the entire role. Sometimes other factors will trump them. The corollary to that is not obvious. It says that the closer the election, the less predictive ability pure economic indicators will have. Nobody know what swings the last two percent of the electorate but it’s not economics because you’ve already used up all of its predictive power.
Econometric predictions have been around for several presidential elections, so we have a track record on some of them. As expected, they’re pretty good at getting close to the final percentages. The problem is when the final result is 48.4% to 47.9% getting close is no better than a coin flip - especially since the larger number went to the loser, as in 2000.
People still make these predictions. Professors want to earn tenure, bloggers need something to talk about, and an actual working predictive model would be the Higgs Boson of politics. I think even Nate Silver, who dissects these flaws in others, puts a little too much faith in them. He’s ostensibly neutral and has to fill his blog, after all. Could he have simply stated that the nominee would be Romney and then dropped the subject during the wonderfully insane days of 2011? No, he was forced to natter on through endless paragraphs of filler about the chances everybody had for what would happen in the next week. Besides all those real world events do allow for tweaking the model for future campaigns.
There’s only one thing everybody agrees upon. Basing a prediction on one economic number is supreme silliness. The quarter in my hand has exactly the same predictive ability. Both are 50/50.
Heads. I’ll claim victory in November.
