Taking all the fun out of election punditry

The 1968 presidential election is the best example of the validity of this kind of analysis. Following the disastrous Democratic Convention in Chicago (the event which, more than other led to the realignment of the political parties from economic and Civil War factors to the current social/cultural ones), Vice President Humphrey was so far behind that the New York Times Labor Day weekend article on the race projected HHH as not leading in a single state.

The race closed, not because of the punditry's explanation (Muskie drew in disaffected Democrats, HHH announced his "plan" for the Vietnam War etc.)but because Democrats were looking for excuses to return to their party during a time of economic prosperity.

It could be argued that Humphrey came close to his predicted outcome only because Governor Wallace took 13% of the vote which otherwise would have gone mostly to Nixon. I think the absence of Wallace would have helped Nixon, but not as much as we, with the benefit of hindsight as to how American politics unfolded, think. The Southern white working class was still several elections away from switching to the GOP. Look at Carter's results in 1976 and even in 1980.

The problem, appropos of a recent post, is that this makes for lousy punditry — it says that there is little in the day-to-day nature of the campaign that will have any effect.

I dispute that. The results show the joint effect of certain broad forces combined with a serious effort to mount a Presidential campaign (OK, that is charitable to The Duke and John Kerry, but still...). The current numbers are so gloomy for McCain that it would predict that only a brilliant campaign against a Dem disaster of a nominee could have given him a hope.

That is clearly not the current case, and Obama is unlikely to make enough mistakes to lose this. But that is different from saying he could take the next four months off and then win. or that serious mistakes are neither possible nor relevant.

The Yankees (or even the Red Sox) would be heavily favored against a solid AAA ball club. But they would still have to take the field.

While Abramowitz's barometer is accurate, because there is no precedent for a black nominee of a major party we don't know what voters will make of that. Thus, Mr. Crook makes short-shrift of the race variable, which is real.

It's not especially difficult to find and arrange variables to "predict" the past. For something like ten straight election, the party of the winner corresponded with the NFL conference that won the Super Bowl -- obviously a mere fluke.

One problem with the barometer is that, in the past, a candidate running after eight years of his own party in power has generally been a vice president (Nixon in '60, Humphrey in '68, Bush in '88, Gore in 2000) very closely associated with the president. McCain has (1) a strong individual identity and (2) a plausible claim to have been a Bush critic on a number of issues.

Speaking of arranging variables...

"Postwar presidential elections" -- why was the postwar limitation chosen? And what happens to the forecast results without it?

Speaking of arranging variables…

“Postwar presidential elections” — why was the postwar limitation chosen? And what happens to the forecast results without it?

The reason probably has to do with less reliable polling stats from the 1930s. The econ stats may be a little shaky too.

Of course, in 2000 the winner of the popular vote was not the winner of the election. Difficult to model, that. Another "wrong win" by McCain in 2008 is not hard to imagine.

There are, of course, possible wild-card factors that can work against Abramowitz's simple equation, and I think there's a lulu of one in this campaign -- which works to McCain's advantage and explains why he's still running so stubbornly close to Obama in the polls.

That's the simple fact that he was not only held as a POW but actually tortured -- and, whatever else you say about the man, the fact that he stubbornly refused to fold under those conditions does make him a genuine hero, and makes it extremely difficult to accuse him of selfish opportunism or deliberate dishonesty. Instead, when McCain says something dumb or flips 180 degrees on a position (as he frequently does), Obama is forced to fall back on saying that McCain is guilty of wishful thinking or plain mental fuzziness. This gives McCain a gigantic advantage that no other GOP nominee could possibly have had, and I think it will enable him to continue breathing down Obama's neck all the way to November.

Any idiot can predict the past. There is a formula called the Taylor Series that shows how you can match any arbitrary continuous function as close as you want by plugging more terms into the formula. It takes remarkably few such terms to create a graph that is virtually indistinguishable from the original.

Wall Street charlatans have for years come up with formulas that match previous results and sell them to gullible investors wanting to beat the market.

All you need to do is find the right coincidences. For example, if the Washington Redskins win the week of the election, this means a win for the incumbent party. This has held true since 1936.

If Abramowitz declared his formula, say, 12 years ago and it was subsequently right in every election since then, then he might have something.

It’s not especially difficult to find and arrange variables to “predict” the past.

Well, only when you have few data points to get right. It's not crazy to model the past--that's pretty much all you can do sometimes, and in general it's quite hard--but modeling a small amount of data might well be meaningless. Here they're predicting 15 either/or events from 3 indicators. Seems kinda iffy but I wouldn't say it's meaningless. It's certainly interesting that you can find indicators that seemingly don't relate to a new candidate's identity.

Sorry for the re-post:

It’s not especially difficult to find and arrange variables to “predict” the past.

Well, only when you have few data points to get right. It's not crazy to model the past--that's pretty much all you can do sometimes, and in general it's quite hard--but modeling a small amount of data might well be meaningless. Here they're predicting 15 either/or events from 3 indicators, though it's hard to account for all the possible indicators one might have chosen and all the different weights one might have given to each, so there might be more degrees of freedom than that. Seems potentially iffy but I wouldn't say it's meaningless. It's certainly interesting that you can find indicators that seemingly don't relate to a new candidate's identity.

It seems like an interesting problem to try to identify all known political indicators and measure their interdependence. If you're lucky, this might give you some notion of politics being approximately N-dimensional for some large (but finite) N. Then you could say more accurately how meaningful it is that it's possible to choose some three indicators that correctly predict 15 elections.

Dan, do you know of any attempts at this kind of analysis? Seems quite ambitious, I know, but someone out there must be dreaming.

Any idiot can predict the past. There is a formula called the Taylor Series that shows how you can match any arbitrary continuous function as close as you want by plugging more terms into the formula. It takes remarkably few such terms to create a graph that is virtually indistinguishable from the original.

You mean an arbitrary smooth function. Continuous functions need not be even once-differentiable. Even in the smooth case you have examples like e^{-1/x^2} around x = 0 (all Taylor coefficients are 0).

In any case the larger point is that yes, you ought to be able to model anything with enough degrees of freedom, but it's a matter of how much flexibility you allow yourself in your model that makes it meaningful or meaningless.

Predictive value is important too, but it's not the case that any model has to prove its predictive value to be meaningful. Such a model might, for example, indicate when a qualitative change in our society takes place.