Jonathan Bernstein asks a great question pursuant to the release of our state-by-state forecasts today. If we want to know what will happen in New Hampshire on Jan. 10, should we be looking at the Iowa polls as well as those from the Granite State?
What I’d love to see Silver plug into his equation is whether Iowa polling helps predict New Hampshire (and other states). I suspect it wouldn’t work — what matters is the spin from Iowa, not necessarily the results, and even then Iowa polls aren’t really a reliable enough guide to Iowa results for it to (probably) work. But I think it would be worth it to try running it, anyway.
The way that I evaluated this was to look at the polling-based forecasts our model would have put out for both Iowa and New Hampshire on the day before the Iowa caucuses in past years. I then ran a simple regression analysis to see what best predicted the actual results in New Hampshire:
Unless you read this blog a lot, you probably aren’t used to seeing Stata output in articles about politics. But the key thing to notice is that the variable attached to the Iowa polls has a very small coefficient attached to it and is only modestly statistically significant. In plain English: there’s not a lot of value in looking at Iowa polls in trying to predict what will happen in New Hampshire.
The reasons are the ones that Mr. Bernstein describes. First, the Iowa forecasts themselves are a fairly crude tool since the caucuses are hard to predict based on polling. If we knew exactly what was going to happen in Iowa, we could do a bit better — but we don’t.
But second and more important, although Iowa can very much affect New Hampshire, it does so in strange ways. Historically, the average “bounce” for a candidate who wins Iowa is about 7 percentage points in New Hampshire. But it varies greatly from election cycle to election cycle. Candidates like the elder George Bush and John Kerry got a 20-point bounce in New Hampshire based on their wins in Iowa. Walter Mondale, on the other hand, substantially underachieved in New Hampshire and lost to Gary Hart despite having won the Iowa caucuses with almost 50 percent of the vote! Somehow, Mr. Hart’s distant second-place finish in Iowa was perceived as a moral victory, and he got the momentum out of the day. (I hope he was paying his media team well.)
One of the determinants of the Iowa bounce, in fact, has been how much a candidate overachieves or underachieves in the polls there. Performance in an absolute sense has mattered very little — it’s mostly how a candidate performs relative to expectations.
In fact, I need to revise my statement from before. Iowa polls are useful for predicting New Hampshire. Ironically, however, they’re useful only after the voting in Iowa takes place. A candidate who gets 15 percent of the vote in Iowa when polls had him at 25 percent will be perceived as a big loser. A candidate like Mr. Hart who gets 15 percent when he was supposed to get 5 percent can get big momentum out of the state.
This is why it’s always a mixed blessing for a candidate to get a favorable poll in Iowa. If he or she fails to live up to the poll on caucus night, its existence can be quite harmful. Conversely, a candidate like Newt Gingrich — who has been slipping in the Iowa polls of late — may benefit from diminished expectations.