On the predictive properties of the Iowa & NH primaries

Great Debates
1

I don’t know that this is really GD stuff, but this is the political forum, and people say things about how likely the winner of Iowa or NH is to win a party’s Presidential nomination. It seemed simple enough to look at just how accurate they are, so I’ve done that.

I’m considering them together, which means I’ve stuck with 1976 through 2004, the period during which Iowa has been significant. That gives us 8 nomination cycles on each side, or 16 altogether.

I’ve discarded nomination cycles where a Presidential incumbent wasn’t seriously challenged for the nomination. For the Dems, only 1996 falls into this category. On the GOP side, 1984, 1992, and 2004 all do so. One can make an argument for Buchanan’s challenge to Bush Sr. in 1992, but I think it’s a weak one, and anyway, it doesn’t affect the stats much.

So we’ve got 12 nomination cycles to consider. Not a huge sample space, but WTF.

In five of these 12 cycles, one candidate won both Iowa and NH: Ford in 1976, Carter in 1976 and 1980, Gore in 2000, and Kerry in 2004. In all five cases, the dual winner went on to win the nomination. (Including Bush 1992 makes this six for six. Like I said, it doesn’t affect the stats much.)

The real question is, what happens when Iowa and NH render a split decision? We’ve got seven such instances, and the verdict is split down the middle: the Iowa winner won his party’s nomination in three of the seven cycles, the NH winner fared equally well, and in one cycle (Dems in 1992), neither the Iowa winner (Harkin) nor the NH winner (Tsongas) won the nomination.

Here they are (party year: Iowa winner, NH winner, nomination winner bolded)

R 1980: Bush, Reagan
D 1984: Mondale, Hart
R 1988: Dole, Bush
D 1988: Gephardt, Dukakis
D 1992: Harkin, Tsongas
R 1996: Dole, Buchanan
R 2000: Bush, McCain

There’s not even any party bias: Two R nominees won Iowa but not NH; two R nominees did the reverse. One D nominee won Iowa but not NH; one did the reverse; one won neither.

Conclusion: winning Iowa or NH is only a good predictor of winning the nomination if one wins both. If you win just one or the other, you’re in much better shape than if you won neither, but it’s still a crapshoot.

But if you win neither, you’re a longshot.

2

And then there’s the vaunted SC primary on the GOP side. Four for four: Reagan 1980, Bush 1988, Dole 1996, Bush 2000. (There was no SC primary in 1976.) All four years, incidentally, are years where two GOP candidates split Iowa and NH.

The problem with patterns such as this on small sample spaces is that they’re made to be broken. Just because five of five candidates who’ve won both NH and Iowa have won their party’s nomination doesn’t mean it’ll get to be six out of six. Just because SC’s gone four for four, doesn’t mean it’ll be five for five. Patterns on a limited data set usually go awry sooner or later.

3

Neat OP.

Just as a note, '92 candidate Harkin was from Iowa, and so that state wasn’t really heavily contested for that reason. Which may be why that year is the only outlier in which the eventual nominee won neither of the first two races.

4

Someone should do the statistical work on this one- I would, but I’ll admit that it’s probably one of the weaker areas of my poli sci background. It’s a small sample size, but it’d still be interesting.

< shows herself to be a total idiot >

You know, the thing where you do the 98th percentile . . . thing? Is it chi square? No. . ok. I’m going to go back and review my statistics book, because this is sad.

5

Thanks. :slight_smile:

I’d expect so. With Harkin essentially being handed (an accordingly devalued) Iowa, all it took was NH being won by a candidate who couldn’t turn that win into broad-based support, which Tsongas wasn’t able to do.

I think I’d shy away from statistics tools with samples this small, and stick with probabilities. 1/16 chance of getting ‘heads’ on four predetermined coin flips, and all that.

6

The way I see Iowa and NH, is that the value is not in who wins, but who loses. This is where the also-rans fade… and #2 or 3 can get a boost.

7

Can’t have one without the other. :slight_smile:

Well, this is the problem: what the limited history of IA/NH shows is that, Bill Clinton notwithstanding, #2 or #3 just isn’t worth that much: you’ve gotta win one of these two to have much of a chance. So far, the only exception we’ve had to that rule was the one year where home-state Sen. Tom Harkin was handed Iowa for free. If he hadn’t run, and a national candidate other than Clinton had won Iowa, and Clinton had still been able to use his second-place showing to get him on his way to the nomination, then I’d ascribe some merit to that position.

But weeding out also-rans so that, say, three or four real contenders can have a less-cluttered race is what IA and NH don’t apparently do. And of course they’ll do it less than ever this year, since all those also-rans have been with us in a plethora of debates in 2007, and the ‘weeding’ of IA/NH will take place only a month before the nomination is resolved, so there’s really no reason for anyone to shut down their campaign on Jan. 9 next month. If you’ve spent a year campaigning already, at that point you might as well hang in there through Feb. 5.

8

New Hampshire’s prominence was due to a long streak of the winner ending up in the presidency (note that even then, the winner sometimes didn’t even get the nomination: Henry Cabot Lodge won in New Hampshire in 1964). For several elections, though, it was true:

1952: Eisenhower
1956: Eisenhower
1960: Kennedy
1964: Johnson
1968: Nixon
1972: Nixon
1976: Carter
1980: Reagan
1984: Reagan
1988: Bush

1992 was the first time the eventual president did not win in New Hampshire (Clinton was defeated by Paul Tsongas). Since then, the 2000 primary did not pick the president (McCain won instead of Bush), but 1996 and 2004 did.

So one winner of the NH primary has been elected president 12 of the last 14 elections, with 10 in a row at one point. This has been a big influence on commentators and given the primary importance in addition to it’s status as being first.

9

The problem with that from a predictive standpoint is twofold:

  1. It’s backwards: what it’s saying is that if you won the Presidency, the likelihood that you won New Hampshire several months earlier is excellent. So it’s very good at ‘predicting’ the past, but that’s of limited utility. :slight_smile:

1a) Those backwards stats are also considerably burnished by the effect of incumbency: five of those 12 winners were incumbents not facing any serious competition for the nomination. If they won the Presidency, then of course they won New Hampshire.

If we take those out, we’re down to 7 of 9. (I know, how’d we get from elections to Star Trek? :)) There’s a 9% chance of getting 7 or more heads in 9 coin flips.

1b) And this coin is weighted: if you won the nomination, you won most of the delegates, so chances are better than even that you won any particular primary. If you postulate a 60% chance that if you won the Presidency, you also won a randomly selected primary (which I think is still pretty low), then the odds of the Presidential winner having won 7 of 9 of a particular primary go up to 23% - an event that can easily happen by chance.

Not that our pundit class is up to thinking through (1), of course, and (1a) and (1b) are just plain right out. So the commentators will continue to speak their usual nonsense.

  1. Depending on whether we’re talking about predicting the presidency or the nominations, the NH primary is decent but not terrific at predicting forward in time.

With respect to the Presidency, the problems are (a) too many elections, especially in recent years, that have pitted two NH winners against each other (2004, 1988, 1980, 1976, 1960); (b) a 7/9 winning percentage in the remaining elections from 1952-present; with (c) five of those wins being in the 1952-72 period, before the nominating process became similar to what it is now. (Up to 1968, only a handful of states had Presidential primaries; I regard 1972 as a transition year. It was a different game back then. Between that and 1976 being the first year where Iowa played its current role, I think 1976-present is the appropriate time period for comparisons.) From 1976 on, the NH winner has beaten the non-NH winner twice in four tries.

With respect to the nominations, NH has a decent 8-of-12 record in the 1976-present era in cycles where the nomination was contested.

In the 1952-72 period, it stunk: once we toss out the wins of Nixon in 1972, Ike in 1956, and Johnson in 1964 against trivial opposition, the NH primary went 4-for-9 in predicting the winner. (The NH winners/nomination losers in this period were Estes Kefauver (D) (1952, 1956), Henry Cabot Lodge (R) (1964), Lyndon Johnson (D) (1968), and Ed Muskie (D) (1972).) Overall, that leaves NH at a mediocre 12-for-21.

Personally, I’d leave the pre-1976 years out of it. They may matter to David Methuselah Broder and his compatriots, but that’s just one more reason to ignore Broder and friends.