Oh, I hate when this happens. Henry Farrell not only
beat me to an obvious response, but came up with exactly the XKCD strip that goes with it.
I suppose I should back up. This is about a Megan McArdle item this morning...actually, two items, but the relevant one is a follow-up in which she defended her
assertion of a 70% chance of unified Republican government after the 2016 elections. Why? Because there's only been one time since World War II that a president has been elected to replace a same-party president. Henry nails it:
Human beings are cognitively predisposed to perceive patterns in the world. Many, likely most of these patterns are garbage. Without good theories, and good ways of testing those theories, we’ll never be able to tell the garbage patterns from the real ones.
Okay, but what we need is an intermediate step. I mean, sure, political scientists can look at election results carefully (that is, using the well-developed information we have about how US elections work in general) and get a general sense of which patterns to test for, and then design sophisticated tests to get that final step. But that's not available on the fly to lots of people who are still going to want to figure out what to do with a pattern they find. So what can we recommend?
Here's a three-step test to check to see if the pattern you've noticed is worth tossing into a blot post or not.
First of all, look at how many times the pattern has recurred. In McArdle's case, we're talking about times when a president stepped aside (making a same-party succession possible). That happened in 1952, 1960, 1968, 1988, 2000, and 2008. So her pattern, to begin with, is one out of six. That's perhaps something...but it's not exactly an Iron Law of Politics, is it? 0 for 10, or 1 for 50, would be a lot stronger.
Then, next, we can check the qualifiers to see if they're making the pattern look stronger. In this case, there's one: postwar. If we put that aside and go with "20th century," then we add 1908, 1920, and 1928 -- and get two hits, with TR/Taft and Coolidge/Hoover. Is there some special reason that the postwar era should be different? Not that I can think of, and if we include those the pattern drops to three in nine -- hardly something to get worked up about. Note that the more qualifiers you toss in, the more likely you are to be creating the pattern that you're seeing, so this is an important test.
What's next? Well, are the cases you are using strong evidence of something, or weak? Here, out-party replacements by Ike in 1952 and Obama in 2008 were both pretty solid...but so was George H.W. Bush's counter-pattern win. The rest were toss-ups: Nixon/Kennedy, Humphrey/Nixon, and Bush/Gore, with the latter of course counting the other way on the national vote. Overall, that seems a lot closer to a coin-flip than an Iron Law.
Actually, I suppose that this suggests a simple rule of thumb for presidential general election patterns: if the pattern depends on how one scores Bush/Gore, it's probably not a pattern worth caring very much about.
But of course, as Henry suggests, this kind of thing comes up all the time, and it's going to; we are pattern-finders. And yet it's unfair to say that reporters and pundits should rigorously test every pattern they notice. In this case, it is in fact probably true that extra terms in the White House make electoral defeat more likely...but to the extent that kicks in after two terms, it's going to be a fairly small effect.
So, go ahead and pattern-spot. Just be aware that (as Henry says) most patterns are garbage, so be careful -- and do a quick self-exam of your pattern before you invest any meaning in it at all.