Skip to main content
ABC News
The Best Pollster In Iowa Just Released Its Final Survey — How Accurate Has It Been?

The final Des Moines Register poll was just released, showing Donald Trump leading the Republican field in Iowa with 28 percent, Ted Cruz with 23 percent and Marco Rubio with 15 percent. Hillary Clinton was ahead of Bernie Sanders, 45 percent to 42 percent, on the Democratic side. The political world — us included — was eagerly awaiting this survey, as Ann Selzer, who has conducted the Register’s polls since the 1988 caucuses, has a very good track record. But just how predictive of the final results have Selzer’s polls been? History suggests they’re a good indicator of what will happen in Monday’s Iowa caucuses, though there is room for a candidate or two to surprise.

I’ve gathered Selzer’s final caucus polls1 for Republicans and Democrats since she started at the Register. For each race, I’ve calculated the difference between all candidates’ polled percentage of support and their final share of the vote.2

Selzer’s final poll has correctly projected every single winner except for Republican Rick Santorum in 2012.

How accurate has Ann Selzer’s final caucus poll been?
1988 Democratic Dick Gephardt 4.3 8.4
1988 Republican Bob Dole 3.6 10.4
1996 Republican Bob Dole 4 9.7
2000 Democratic Al Gore 2.2 2.5
2000 Republican George W. Bush 4.3 7.7
2004 Democratic John Kerry 4.3 10.1
2008 Democratic Barack Obama 3 6.1
2008 Republican Mike Huckabee 1.3 4
2012 Republican Rick Santorum 2.6 8.5

Source: Des Moines Register

And the Register poll’s successes haven’t been limited to blowout races. She caught the late momentum for Democrat John Kerry in 2004 and famously called Barack Obama’s victory in 2008.

Even when Selzer’s final poll missed the winner in 2012, it still indicated the potential for Santorum to win. The survey was conducted over four days, and in addition to the topline, aggregate numbers, Selzer released day-by-day results. The final day of her last 2012 caucus poll showed Santorum within a percentage point of Mitt Romney. In fact, the average error per candidate of the last day of her final 2012 caucus poll was just 1.9 percentage points.

Indeed, what makes Selzer truly special isn’t just that she calls winners but that her error rates are fairly low across all candidates. Her average error per candidate per year has been just 3.3 percentage points. That means that what a candidate receives in her poll is probably going to be pretty close to what he or she gets from voters.

That’s not to say the Des Moines Register poll is perfect. It sometimes misses on a candidate by a lot. Selzer’s final poll in 1988 missed Republican Pat Robertson’s eventual vote share by just more than 10 percentage points. Same thing with Kerry in 2004.3

Who might benefit from that type of miss this time around? History suggests there are two types of candidates who tend to outperform their polls. The first is a candidate who does well among Christian conservatives. Selzer’s final polls on the Republican side in 1988, 1996 and 2012 all missed the candidate favored by Christian conservatives by at least 8.5 percentage points. That could be good news for Cruz. Secondly, candidates with late momentum, such as Kerry in 2004 and Santorum in 2012, also tend to beat their polls. That could be beneficial to Rubio, who seems to be gaining in some polls.

Could there also be a big error on the Democratic side? It’s possible, but Selzer did particularly well in 2000, the last Democratic campaign with only a few candidates running. Fewer candidates means voters have an easier time settling on one candidate and reallocation of support becomes less of an issue.

Of course, we won’t know how accurate Selzer’s final poll is this year until Monday or the day after. It’s worth remembering, however, that even the best pollsters — and Selzer is one of the best — aren’t perfect.

Read more:

Four Roads Out Of Iowa For Republicans

What Happens If Bernie Sanders Wins Iowa

Check out our live coverage of the Iowa caucuses.


  1. Selzer did not conduct the Des Moines Register’s 1992 poll.

  2. To account for undecided voters — polls include undecideds; final results do not — I’ve allocated them to candidates. There’s some question about whether to allocate undecided respondents proportionally (the leading candidate gets the most) or just split them up equally; I’ve split the difference, divvying up the undecideds half equally and half proportionally.

  3. Although, because Democratic caucus rules require candidates to meet certain thresholds of support, and caucus-goers can move camps if a candidate fails to get enough support, it’s possible that Selzer’s final poll — or any poll — accurately portrayed the pre-caucus preferences of voters.

Harry Enten was a senior political writer and analyst for FiveThirtyEight.