Explore Harvard's Nieman network Nieman Fellowships Nieman Lab Nieman Reports Nieman Storyboard
Supporters raised their hands New Year's Day when Barrack Obama asked if they would take part in the Iowa Democratic caucuses. (AP photo)

How not to conduct a presidential poll

COMMENTARY | January 02, 2008

From its bumper-sticker mentality to its gaping margin of sampling error for subgroups, the Des Moines Register's new poll of likely Democratic caucus-goers provides a great example of what not to do in an election year.


By Barry Sussman
bsussman@niemanwatchdog.org

The Des Moines Register’s poll of likely Democratic caucus-goers published Monday — widely regarded as the single best pre-caucus measure — exhibits many of the worst characteristics of election-year polling.

It could serve news organizations that do polls as a “how not to” guide for the rest of the 2008 campaign.

The poll focused only on the horse race and on vague bumper-sticker voter motivations, like desire for change or unity. That’s a disservice.

In a 33-paragraph story and in sidebar charts, the number-one issue for many Democrats, Iraq, was never mentioned. Not once. This seems standard procedure for the Register; earlier on its editor banned discussion of Iraq in televised Democratic and Republican candidate debates she moderated.

Did the poll questionnaire even try to determine what bearing Iraq or any other issues might have on people’s choices? Not as far as I can tell — but I can’t rule that out, either, as there was next to no transparency in the paper’s report.

The poll had Barack Obama with 32 percent of likely caucus-goers, Hillary Clinton with 25 percent and John Edwards with 24 percent.

Eight hundred people were surveyed by telephone Dec. 27 to 30. A striking assertion was that 40 percent of the Democratic caucus-goers will be Independents, another 5 percent Republicans and that 60 percent of the 800 have never attended a caucus before.

Perhaps these unlikely-sounding predictions will be correct; there’s a lot of interest in the Democrats this year. But the Register should have explained its turnout conclusions. How did it determine who’s likely to show up and who isn’t? In years past, the Gallup poll had a nine-question filter to determine likely voters. What’s the Register’s filter? The story doesn’t say or even give a hint.

Another problem is the difficulty of polling during the Christmas-New Year’s holiday period; fewer people than usual are available for interviews, and there’s no easy way of knowing if those who are available have somewhat different candidate preferences.

In addition, the poll report goes too far in analyzing candidate preferences among subgroups. The overall margin of sampling error for a survey of 800 people is about 3.5 percentage points, as the Register story states. But for smaller groups, such as women under the age of 35, or between 35 and 54, or 55 and older, sampling error is substantially higher. But that didn’t stop the Register from stating, without noting the higher error margin, the preferences of each group.

Polls can tell us a lot about the electorate — but not when they are so fixed on the horse race to the exclusion of everything else.

Bandied about over the years is the question of whether pre-election polls change voter behavior. If they do, it seems certain that the change would be only slight — but slight differences could result in different outcomes. For that reason alone, news organizations should be extremely cautious in their polling.

Two of the first items to appear on this Web site in 2004 were primers on election polls, one by Barry Sussman, the other by the late Leo Bogart.



Sussman responds
Posted by Barry Sussman - Editor, Watchdog Project
01/04/2008, 01:22 PM

Mr. Washburn is correct in several of his important criticisms. Although I looked for it online before I wrote, I didn’t see the sidebar describing the methodology and questions – but it was there.
>http://www.desmoinesregister.com/apps/pbcs.dll/article?AID=/20071231/NEWS09/71231041/-1/iowapoll07<
Neither did I see the “influential issues” chart. It’s fair for him to say I should do a little reporting next time. He also could have mentioned that the poll was highly accurate in key points: Obama won going away and while the survey got the second and third place order wrong, it was off by only a miniscule amount. It was correct in anticipating a “dramatic influx of first-time caucusgoers,” a main factor.

I feel strongly that omitting Iraq as an issue in the Register’s poll story and in candidate debates were disservices. Iraq is the elephant in the room. The chart I missed showed Iraq as the No. 1 issue. A simple crosstab would have helped explain whether it had any bearing on how those interviewed were choosing candidates. Sadly, the Register isn’t alone in ignoring Iraq – almost all the coverage I saw focused on how, for Iowans, change trumped experience. Change from what?

It’s during election campaigns that Americans are most attentive to national issues and problems. The press needs to deal with important issues, not put them aside or bury them.

As for poll reviews, I’ll try to do better next time.
--Barry Sussman





The NiemanWatchdog.org website is no longer being updated. Watchdog stories have a new home in Nieman Reports.