Articles and Analysis


In Iowa, Somebody Was Right**

Topics: 2008 , Exit Polls , The 2008 Race

Needless to say, from the pollster's perspective, there were three big winners in Iowa last night: Barack Obama, Mike Huckabee and J. Ann Selzer, the pollster for the Des Moines Register.

As regular readers know, the final Register "Iowa Poll," released on New Year's Eve, showed Barack Obama leading Hillary Clinton by a seven-point margin (32% to 25%), followed closely by John Edwards (at 24%). The result was stunning, indicating a better Obama showing by far than in other recent polls. Only two polls released during the latter half of December had shown Obama even nominally "ahead," and neither margin was large enough to attain statistical significance. Earlier that day, our "sensitive estimate" (based on all other recent polls, but set to be more sensitive to the most recent) gave Clinton a 4.5 point lead over Obama. The Reuters/Zogby tracking survey released that morning (and conducted even more recently than the Register poll) showed Clinton leading Obama by 4 points (30% to 26%).

The explanation for the difference was even more stunning. "Obama's rise," the Register reported, was "the result in part of a dramatic influx of first-time caucusgoers, including a sizable bloc of political independents." The poll showed 60% saying this would be their first caucus and 40% identifying their party preference as Democratic. No other poll that disclosed similar results to Pollster.com came close on either measure. Within an hour or so of its release, the Register poll had been condemned by the Clinton and Edwards campaign pollsters as inaccurate, "at odds with history" and based on an "unprecedented new turnout model." The only way the poll would be accurate, Edwards consultant Joe Trippi said later, is if "220,000 people vote."

Despite the dark insinuation, Selzer had not changed her methodology. She had "assumed nothing" about the demographics or party allegiance of the likely caucus goers interviewed by her company. Instead, as she explained it to the News Hour, "we put our method in place, and we let the voters speak to us." In the face of massive skepticism, some of which seemed to come even from the Register's most prominent columnist, she stood by her numbers.

And yesterday, those findings were vindicated. Obama won by an 8-point margin in the official results and by approximately seven-points on the (more comparable) entrance poll head count estimate. The turnout was 239,000, nearly double the number from 2004. And the entrance poll put the share of first-time caucus participants at 57%. The only mismatch was on party -- 23% were independent or Republican on the entrance poll as compared to 45% on the Register survey.

I would urge some caution on making too much of the inconsistency on party. The jury may still be out on that issue. Pollsters sometimes get slightly different results on that question depending on the exact wording, the structure of answer categories or on the nature of other questions asked earlier in the survey.

In this case, it is also worth remembering that 100% of the caucus-goers were registered Democrats by the time the Caucuses got underway last night. Those who had been registered as Republicans or who had a "no party" status had to fill out a form to switch their registration as soon as they arrived. A friend within the Obama campaign tells me they went to great lengths to prepare newcomers for the process of switching their registration: they described the process in radio spots, campaign literature, canvass scripts, email and four direct-mail pieces sent to identified supporters. Other campaigns presumably provided similar preparation for identified caucus newcomers.

So we can safely assume that most non-Democratic registrants arrived at their caucus location expecting to fill out a form switching their party registration to Democratic. Is it a stretch to assume that many, who described their party leanings as "independent" in a telephone interview a few days before, were more likely to check "Democrat" on a paper questionnaire they filled out as they entered their caucus location?

Of course, party identification and party registration are not the same thing. The entrance poll party question asked explicitly about how "you usually think of yourself." Also, I am not sure there were enough party switchers to support a the 45% result, even if my theory about a "measurement" shift on the entrance poll is valid. However, the answer is knowable. The Iowa Democratic Party will be able to report on the number of new registrants and party switchers in a few weeks, and while she has not yet reported it, Selzer can compare those statistics to the number of non-Democratic registrants that qualified for the survey (remember, though they are highly correlated).

Either way, Selzer and the Iowa Poll got the big story exactly right, and they saw the results coming when other pollsters were showing something very different.

And what about the critics that spun so furiously to discredit her results? ABC's Gary Langer adds this postscript:

On the press plane flying from Iowa to New Hampshire [this morning], our off-air reporter Eloise Harper reports, “Mark Penn admitted to knowing that the trend was shifting towards Obama this past week.”

That means that at the very moment Penn was accusing the Des Moines Register of producing unreliable data, and saying it was Clinton who had the momentum, he knew otherwise.

So to be clear. Ann Selzer changed nothing about her methods, stood by her result and earned vindication, demonstrating the considerable integrity and courage that had already earned her the respect of her colleagues. Mark Penn trashed her poll and then changed his story.


Charles Franklin posted his thoughts earlier on how our polling trends compare to the entrance poll and official results, and we will have more to say soon about the exit the "accuracy" of Iowa polling. Stay tuned.

**Apologies to Langer (for the title)



Which is the poll to watch in New Hampshire?


Jeff Winchell:

I am not so sure about the 239,000 number. The Des Moines Register's website has *actual votes* listed that total up to about 250,000. I was shocked to see this webpage listing actual votes because I thought the Iowa Democratic party basically forbid this, but then that might be why I found it really hard to find the webpage today when I looked (which is why I saved a hard copy of it in case it goes away completely). Also, the Republican list only has 93% reported precincts so they seem to have given up updating this page in favor of their less easy to view formats.


The data above is at the county level. If anyone finds this data for the dems at the precinct level, let me know so I can update my model from the simulator I wrote.



The good news is that I think Selzer has now added to her already very high reputation (and rightly so).



On a slightly different note, can we please finally put these ARG polls in the garbage where they belong?



Does Anne get a win bonus???:)


Mark Lindeman:

Jeff, no, sorry, those aren't actual votes. Mark B. posted an explanation from Joe Lenski in the entrance poll open thread comments, but it can also be deduced from the data, which are spookily consistent with the state delegate equivalent allocations. In fact, they are basically 100 times the SDE values (there is an implied decimal).

Dem precinct-level data -- the fullest available AFAIK -- can be found at http://s3.amazonaws.com/caucuspublic/upload/full_data.txt ; the format appears to be county number / precinct number / candidates in alphabetical order (I'm not sure of the exact definitions of the last two columns). These aren't head counts of voters. I believe they are delegates elected to county conventions; these figures aren't proportional to state delegate equivalents. But you can get SOMETHING out of these data if you work at it, and information wants to be free. (The counties appear to be in alpha order.) If you chop the URL after caucuspublic, you can see some other aspects of the site design.


Mark Lindeman:

Sorry, I got sucked in by the site bug. Oh well, that was such a great post, I'm sure everyone wanted to read it twice!


Jeff Winchell:

Bummer. But that makes sense about multiplying by 100 since there are 2500 state delegates.

Thanks for the link the precinct level county delegates. I have tables that convert the county level to state level, so that's no problem. I will keep searching the net periodically for more details (for example I was able to find Story county and Jefferson county attendance by precinct for 2004). I know there's chaos at the precincts and it is difficult to report as much detail as we'd like, but the precincts do maintain a list of everyone who signed in, plus they *might* (or at least some precincts might) have kept the final votes per final preference group before they did the math to figure out delegate counts. Even if that is only partial data, as an analyst, I always feel a partial model/data is better than none at all.


Nick Panagakis:

Mark B. makes a good point. Not all Dem caucus goers were Reg Dems before the caucus. So the XPoll party ID question, even something like "regardless of how you voted, do you think of yourself as..." could be ambiguous to someone who is otherwise independent who had just registered as a Democrat to participate in a Democratic caucus.

Since party affiliation is such an important part of election analysis, perhaps XPolls should be asking a party registration question also,in IA, something like "before today, were you registered as...." In NH, an open primary state, they will probably ask a straight registration question in addition to party ID to determine how unaffiliated registered voters voted.

Party ID data are not as stable as you might think. They zig-zag up and down over time. I see it my numbers and in Gallup numbers I got a few years ago.

This is important. My experience is that pre-election party ID is *never* comparable to XPoll party ID, even though in my phone polls, "independent" is an option not read to respondents. "Independent" always declines on election day. This is true for both national exit polls and my own. In a self-administered XPoll questionnaire it is an option by necessity.
2006 was typical of past years:
10/27-29 IL Trib poll: Dems 42%, Reps 25%, Ind/else 33%
Illinois XPoll: Dems 46%, Reps 31%, Ind/else 23%

As I said above, Party ID fluctuates across time anyway. So there must be something about the act of voting, the act of committing to a candidate in a general or a party in primaries that affects self-perception.



Mike Rappeport:

One thing Ann Selzer's results show is the advantage of deep thought based on experience (in her case with Iowa and the Iowa caucuses) over charisma, resources expended, etc, in really having enough understanding of a situation to do a good job. Does anyone think (or Ann claim) that she could do a similarly outstanding job in other states? I doubt it.

I would suggest that what underlies Ann's results ought to give some guidance (but clearly is mostly ignored) about both the Democratic and Republican candidates and the record of the Bush administration (see Rumsfeld/Chaney vs Petraeus, and/or inexperience vs. Katrina).



Read Dick Bennett on the American Research Group site. Yes, Seltzer's #s were good - except she miscalculated the independents;
they were closer to 24% not 45%.
Obama won because he picked up 8% in 2nd choice votes in areas where Dodd et ala couldn't muster 15% to pass the arcane viability rule. In MI and NV and FL et ala,
there won't be any 2nd choice votes to bail out Obama who trails badly in almost every state in the union.



Dick Bennett tries unsuccessfully to explain why he was so wrong. ooooooooh noooo, He actually tries to viciously criticize the DMR poll which was very accurate instead of explaining why his poll was so off. One thing we learn, the ARG poll is trash. Please stop including it in the pollster average. Read Dick(with emphasis) indefensible explanation, no one should buy it http://americanresearchgroup.com/


Mark Lindeman:

jonny, as to Selzer and independents (note that Mark B. did make that point), check out especially Nick Panagakis' comment above. Selzer may well have had too many independents, but it is hard to know how to compare the pre-election percentage to the entrance poll percentage.

Actually, ARG provides an interesting angle on this. I wondered where your 8% figure came from: "If the results from the Register's poll are adjusted based on party composition of the entrance poll, the Register's final poll results would be Obama 30%, Clinton 29%, and Edwards 24%. Using the final caucus results that include only the viable candidates, Obama picked up 8 percentage points of second choices (thanks in large part to the Richardson campaign)...." I'm a trusting guy, but I don't think Obama netted 7 points over Clinton when reallocating maybe 15 points' worth of non-viable candidate support. I think some people who would have shown up as independents in the Iowa Poll called themselves Democrats in the entrance poll -- and that's why ARG's adjustment comes out so oddly.

Remember, the same entrance poll that ARG is using to hack up Selzer's numbers showed Obama with over a 6-point lead over Clinton. How does that square with Obama's win being due to reallocated second choices?

I'm not going to rant about the ARG poll, but Bennett's analysis is a head-scratcher.



I talked to 6 people who were at different precincts. This is what happened in 5 of them (the 6th was tied all the way through). Obama had over half of the people there on the first vote. Hillary was second in 4 and Edwards in the other. The people who's candidate was not viable for the most part went to Edwards.
The main reason was to not let Clinton to get second. The Obama crowd was overwhelming (in one precinct, there were 267 Obama on the first vote out of 502 people - and this precinct only awarded 7 delegates), so they decided to go somewhere. A lot of the voters (I can only speak of the ones at these precincts) move was against Clinton, not for Edwards.


Jeff Winchell:

I don't get your math about Obama picking up 8 points due to 2nd choice votes.

My calculations using the entrance poll data vs. state delegates (admittedly not an apples to apples comparison)

57% Edwards
21% Clinton
20% Obama

I haven't done the calc on county delegates vs. entrance poll yet, so maybe it will shift a bit.


Not sure where else to put this comment, but are there no new polls in SC and elsewhere to report? I'm really curious how Iowa has changed undecideds and particularly black voters in South Carolina. Thanks



Yes, what about South Carolina and Florida? I'd like to see the post-Iowa GOP #s in those states.


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.