Articles and Analysis


POLL: Des Moines Register/Selzer & Co

Topics: 2008 , Iowa , The 2008 Race

The much anticipated final poll from the Des Moines Register and Selzer & Company is out. They conducted interviews from December 27-30 with 800 likely Democratic caucus goers and 800 likely Republican caucus goers (those who say they "definitely or probably will attend" the caucuses). The Register has full coverage of the Democratic results, the Republican results, leading issues, an overview from columnist David Yepsen and a description of their methods.

The Register's lead for the Democrats:

Democratic presidential candidate Barack Obama has widened his lead in Iowa over Hillary Clinton and John Edwards heading into Thursday's nominating caucuses [...] Obama's rise is the result in part of a dramatic influx of first-time caucusgoers, including a sizable bloc of political independents. Both groups prefer the Illinois senator in what has been a very competitive campaign.

Obama was the choice of 32 percent of likely Democratic caucusgoers, up from 28 percent in the Register's last poll in late November, while Clinton, a New York senator, held steady at 25 percent and Edwards, a former North Carolina senator, was virtually unchanged at 24 percent.

Next, the Republicans:

Mike Huckabee, a former Baptist minister riding a wave of support from fundamentalist Christians, tops Mitt Romney for first place [....] In a battle of former governors from Arkansas and Massachusetts, Huckabee leads Romney, 32 percent to 26 percent. [....]

The new poll [also] shows a resurgent Arizona Sen. John McCain grabbing third place in the Republican race for the first-in-the-nation caucuses. McCain tallies support from 13 percent in the poll -- a 6-point improvement since late November.

The two polls stories, plus Yepsen's column, are well worth the click. Unlike the last two releases, the stories include tables of results by demographic groups.

A few quick observations. What "will raise some eyebrows among party pros," as Yepsen puts it, is the fact that a "whopping" 60% of the Democratic caucus goers say this will be their first caucus and only 54% say they are Democrats (40% identify as independents and 5% as Republicans). Compare these results to what other polls have shown earlier in 2007 and it becomes clear that this Register sample predicts a very different set of caucus participants than in years past.

Yepsen also notes that if pollster Ann Selzer had weighted the new results by party identification "to look like they did in 2004, Clinton could beat Obama 31 percent to 29 percent."

However, one of the most critical challenges in polling the Iowa Caucuses is that no one knows for certain who will turn out this week. When different surveys produce results that vary beyond the margin of sampling error, their differing conceptions of the likely caucus goers explain most of the variation.

So is the Register/Selzer poll right or wrong about a the potential for "a dramatic influx of first-time caucus goers" and independents on the Democratic side? I know our comments section, and the political blogosphere, will be alive with speculation, but we really will not know for certain until Thursday night.

What I can say is that when we polled campaign and media pollsters last week, the Des Moines Register and Selzer were the runaway winners as the most trusted Iowa poll. My hunch is that their reputation results partly from an awareness of their past success and methods but mostly to an appreciation of what is at stake for the pollsters. Selzer is a Des Moines based researcher, and this survey is easily the most important her company has conducted since their last pre-caucus poll in 2004. Under those circumstances, other pollsters trust her to sweat the details.

Having said that, we all know that the conditions for survey research are treacherous this week, and even the best pollsters (and methods) are fallible even under the best of conditions. But with everything on the line, Selzer has done what good pollsters are trained to do: She put her trust in her methods and the results they produced, even when those results contradict conventional wisdom.

Update: The Edwards campaign issues a critical analysis of the Des Moines Register/Selzer poll.

Update II: Another critical assessment from the Clinton campaign.

Update III: New polls were also released late yesterday by InsiderAdvantage (showing Clinton and Edwards leading Obama) and early this morning by CNN/ORC (showing Clinton with more support than Obama but within the margin of sampling error). We will have full poll updates later this morning.

Update IV: Overlooked this earlier this morning - Marc Ambinder notes: "Obama's internal polling does not show this high a proportion of independents choosing to caucus" (via Kaus).

Update V: Kaus also notes that the 2004 Register poll "came out only a day before the caucuses--not three days before." True, but although the Register poll was released on the day before the Caucuses (Sunday, 1/18), they completed interviewing two days earlier (surveys conducted over four days, from 1/13-16, Tuesday to Friday). So that poll had the same three-day gap as this one.

One intriguing footnote: That final 2004 Register poll showed Kerry with 26%, Edwards with 23%, Dean with 20%, Gephardt with 18%. On the same Sunday, Zogby/Reuters also released a three-day rolling average tracker with fresher interviews (conducted Thursday to Saturday, 1/15-17) but with Edwards running five points lower (Kerry 24%, Dean 23%, Gephardt 19%, Edwards 18%). Edwards had 17-18% over the nights of interviewing that coincided with the Register poll.

Now to be fair, Zogby did show Edwards rising at the rate of a little less than one percentage point a day over the week leading up to their Sunday release. However, on the final Zogby release, on Monday (interviews conducted Friday to Sunday, 1/16-18), Edwards suddenly jumped 3-points (showing Kerry at 25%, Dean 22%, Edwards 21%, Gephardt 18%). Given the 3-day rolling average, the Edwards number on the final night of interviewing had to be in the 25-30% range. Such a result is plausible, given that Edwards received 26% on the network entrance poll, and Edwards was certainly surging in the campaign's final week. But draw your your own conclusions as to why the Register caught more of the Edwards surge earlier.

PS: Happy New Year!




Everybody is saying DMR's Iowa polls set the standard. But nobody pointed out they actually missed 2004 presidential election by 4 points, basically they picked up the wrong winner in Iowa.

Their last poll was Kerry 48, Bush 45. The actual result was Kerry 49, Bush 50.

Their 2004 caucus poll number was not much different from Zogby's percentage wise. They were just lucky to get Edwards/Dean's order right.

The results are pretty shocking and we shall see.


Jeff Winchell:

Good point. Thanks. I would also add that polling Iowa's presidential election is MUCH MUCH easier than polling their caucus, not to mention polling it over Christmas. I predict that the media will overexaggerate the relevance of this poll.


Jeff Winchell:

Where can I find the detailed results from this poll. I am particularly interested in learning the data on 2nd and 3rd choice (and those questions about who would you prefer to pick 1st, but won't and why?)

I looked on the DesMoinesRegister website, but only see the "executive summaries" within the stories.


Great job, Mark - happy new year to you and yours.


Mark Lindeman:

"hello"'s point doesn't seem all that good to me. 45/48 in a poll with a 3.5-point statistical margin of error (that's on the vote share, not the difference between the candidates) = "too close to call," not "basically they picked up the wrong winner."

Of course, by the same token, the poll doesn't deserve huge credit for predicting the order of finish in the last caucus.

(Speaking of Zogby, I think they had Kerry +5 over Bush in Iowa. Stuff happens.)


I'll take momentum. A quarter of caucus-goers will decide in the final three days.

From the DMR's David Yepsen discussing his paper's poll:

"the nightly results show Obama's support flat over the last two nights, Clinton's declining each night, and Edwards' support increasing each night."

Mark Lindeman:

Petey, I'm unable to find Yepsen saying that. I can only find Cillizza quoting Hickman as saying that Yepsen said that.What Yepsen said, I would be happy to know.

What he says here is that "Obama seems to be opening up a last-minute lead...." That doesn't contradict the assertion about the nightlies, although they apparently would be based on so few interviews that I don't see how they could prove much.

I don't especially think Obama is pulling away in Iowa -- it mostly seems to come down to guesswork about who will actually vote. I just like quotations to be accurately attributed.


Mark Blumenthal:

Mark & Petey,

I have not tried to find that Yepsen quote. However, the numbers in question appear in a chart at the very bottom right of the Register story on the Democratic sample.

The caption reads:

"During four days of polling, Hillary Clinton's two-day rolling average declined slightly while support for Democratic rivals Barack Obama and John Edwards rose somewhat."

The changes on the chart from the first two days to the last two days of calling are:
*Obama +5 (29% to 34%)
*Edwards +3 (23% to 26%)
*Clinton -4 (27% to 23%)

However, if I assume 400 interviews in each two-day subgroup, none of these changes achieve statistical significance (the p values I get are .13, .32 and .19 respectively, and that assumes simple random sampling).


nick panagakis:

The Edwards campaign critical analysis says "Usually, the number of first-time caucus goers is no more than 20%."

Entrance polls in 2000 and 2004 (Dem only) show much higher first-timers, 41% to 46%. Don't where they got 20%.

I posted this elsewhere on this site yesterday.

"This is off subject but a couple of weeks ago, Mark posted sample characterisitcs of Iowa polls. Most were way low on first-time causcus attendees compared to past entrance poll results.
"In 2000 and 2004 entrance polls showed 41% to 46% showing up for the first time at IA caucuses.

"I am not surprised at this given low overall turnout at two hour event debating their candidate virtues.
Most voters are simply not prepared to do this every four years."
2004 Dems only
First time caucus-goers. 45%.
2000 Dems.
46% first time caucus-goers.
2000 Reps.
41% first time caucus-goers.

Posted by: nick panagakis | December 31, 2007 10:06 AM





Mark can you verify this model from the DMR


Mark Lindeman:

Mark, thanks, but that isn't precisely on point. Hickman apparently claims that Obama's support was "flat over the last two nights" and that Edwards' support "increase[d] each night." Again, that is perfectly possible, but then we're down to 200 interviews per night. So, say Obama went 34%-34% in the last two nights, and Edwards maybe went 25%-27%. With 200 completions per night. Shrug.



If, in fact, the main discrepancy between the DMR poll and other polls is due to differences in projected participation of independents and Republicans, I'm wondering what factors in polling methodology would lead to such a discrepancy. I can think of several possibilities:

(1) Looser screen. If the DMR poll allows a looser screen for likely voters, that could conceivably allow in many more independents who might not pass a stricter screen.

(2) Different screen. Perhaps the DMR screen is not looser, just different, based more perhaps on stated likelihood of participating in the Democratic caucus, rather than other factors such as prior participation in the caucus (which would favor registered Democrats).

(3) No weighting. I'm guessing that registered Democrats are getting huge numbers of calls from campaigns, and are thus not answering their phones. Are more independents answering their phones? Did other polls "weigh" for this and the Seltzer poll refuse it, believing it reflected the actual enthusiasm for participating in the Democratic caucus?

(4) More updated phone lists? Does the Seltzer poll operate off of different phone lists than other polls?

Are there other possible explanations? Or is my basic assumption incorrect--i.e., are there other factors besides projected participation of independents and Republicans that account for the discrepancy between this and other polls?



Having worked through these three polls, Des Moines Register favors Obama, Insider Advantage favors Edwards after the 15% second ballot and delegate apportionment and CNN favors Clinton. Can we assume the two most significant factors influencing these disparate results are (a) the possible influx of a different mix of caucus goers and (b) the final delegate apportionment process in contrast to whether a particular poll is focused only on the initial round "straw vote". The actual delegate count is what should matter. However, the delegate count will probably get lost in the media hoopla over percentage reports. In the final analysis, the question becomes one of momentum which is triggered by media spin, campaign spin, campaign strategy and finances. One would think based on all these factors, that all three lead candidates continue to New Hampshire and beyond. The remaining candidates appear to have no chance.

On the Republican side, the Register has Huckabee by 4 and CNN has Romney by 5, with McCain 3rd or 4th. Giuliani has lost at least half of his support in both Iowa and New Hampshire in the last month. He clearly is playing the "low expectations" game in these two states and taking a national approach unlike any other major candidate. If Giuliani's strategy ends up being successful, it will be interesting to see whether candidates in the future place all or almost all of their bets on the very early primary states. We shall see.



Let's not forget that Opinion Research's parent company is owned by a Clintonite--Dick Bennett.

There may be a conflict of interest here.


Andrew :

Correction. The owner of Opinion Research who has donated and given contracts worth millions to the Clintons is Vinod Gupta, not Dick Bennett. I had the name Bennett in my head because he was wrongfully accused of being a Clinton donor the other day.

Here is the NYT article from July, 2007 noting that Gupta once "gave consulting contracts worth $3 million to Mr. Clinton" and "nearly $1 million to the Clintons and other candidates from the 1980s on."




Of course it is a good idea not to read too much into the tracking data in the DMR poll, but it does seem to provide some evidence in support of your holiday effect hypothesis.

Incidentally, one of the significant differences between the CNN and DMR polls may just be that the CNN poll started interviews on 12/26, and the DMR poll waited until 12/27. If we hypothesize a holiday effect along the lines of what the DMR tracking data would suggest, that extra day on the front end could do a lot to skew the results.


The DMR poll director made an incredible statement on cspan this morning. I am waiting for the video or transcript to go online:

She said that the large projected increase in turnout on the Democratic side could be due to MORE VIABLE CANDIDATES and MORE TIME SPENT cultivating voters.

That is patently untrue:

Clinton,Edwards, Obama = 3 viable
Dean,Kerry,Edwards,Gephardt - 4 viable

Plus ind. and republicans didn't have a choice last time it was dem caucus or nothing. This time there is a republican race with candidates ranging from Paul (anti-war) to McCain (pro-war), Guiliani (pro choice) to Huckabee (pro life)

I think they got lucky in 2004 because the facts show they issued 2 sets of data:
definite vote:

Their dv screen was more accurate for Kerry and Dean but had the wrong order.

Their lv screen was less accurate for all three but had the right order.

If that isn't luck, what is it?



I guess it depends on your definitions, but it seems to me that Biden, Richardson, and Dodd are all viable candidates, and I believe they have in fact spent a considerable amount of time (and money) campaigning in Iowa.



Regarding second choices, Yespen says:

The three front-runners each get a few, but it doesn’t change the order of finish: Obama would win with 28 percent; Clinton would still finish second with 20 percent and Edwards would capture third with 19 percent.

This is confusing to me. The wording in Yespen's column implies that these are the votes after the redistribution of second preferences, but that can’t be right. The three leaders can’t all decline. So, perhaps these are the second preferences themselves? In that case, there are either a lot of voters undecided–Yespen does not say–or who prefer other nonviable candidates, as these totals add to only 67%.



This will sound cynical, but isn't what the DMR did in this poll just setting themselves up to have it both ways? If Obama wins, it will most likely be because of an unconventional turnout model. If Obama does not win, then everyone can shrug and say I's and R's or new D's jsut didn't participate. I'd like to have heard more from Slezer or Yepsen on the rationale for this model and the screen used.
Also - is it really responsible to list nightly numbers? Are these "partials" meaningful? (200 A NIGHT??)



I checked the link you provided for the "accuracy" of DMR's 2004 poll. And what I saw was that their "likely voters" poll was just as flawed as everyone else -- only their "definite voter" poll proved better than other polls. And since the numbers for the poll you are citing are of "likely voters", not "definite voters", the idea that this poll is somehow more "reliable" than others doesn't make a whole lot of sense to me.

Did DNR release its "definite voters" polling results?



The Iowa Poll also got Gephardt REALLY wrong in 2004. Sure they got the order right, but they polled Gephardt at almost twice the level (18%) he actually received (10.6%). Insider Advantage (13%) and Survey USA (11%) were much closer.


Post a comment

Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.