Articles and Analysis


Primary Polling Primer: Timing

Topics: 2008 , The 2004 Race , The 2008 Race

Onward with my Primary Polling Primer. In the first two installments, we talked about turnout, or rather, how national polls tend to sample a broader population than actually participates in the presidential primaries and caucuses. Today, I want to turn to something even more important to the "accuracy" of those national vote preference polls: Timing.

The presidential primaries and caucuses are a dynamic process. Unlike virtually every other election that pollsters ask about, the selection of presidential nominees does not occur on a single election day. Rather, the nominating process consists of a series of statewide primaries and caucuses that plays out over the first few months of every presidential election year.

Here is the critical point: A few early primaries, especially the Iowa caucuses and the New Hampshire primary, play a huge role in influencing voter preferences in the states whose primaries and caucuses follow.

Go back and read that last sentence again, because the basic idea is hard to overstate, especially in years (like 2008) without an incumbent president seeking reelection. Moreover, since the outcome of the Iowa caucuses and New Hampshire primary tend to reshuffle voter preferences, they can render the "standings" of earlier horse race polling more or less moot.

Over the next few posts of this series, I will try to provide some data to demonstrate the influence of those events and how they often cause a significant reshuffling of voter preferences.

For today, consider the last Democratic nomination battle in 2004, which produced one of the most dramatic shifts. Fortunately, for those of us who obsess about such things, we have an incredible resource in the National Annenberg Election Study (NAES), which conducted a rolling tracking survey from October 2003 through November 2004. Kate Kenski, now an assistant professor at the University of Arizona, complied over 7,000 interviews conducted among respondents who were planning to vote in the Democratic presidential primaries and caucuses in 2004 and prepared the following graphic:


As the chart makes clear, the Iowa caucuses and New Hampshire primary turned the 2004 Democratic contest upside-down. Just before Iowa, according to the ANES data, only 9% of those planning to vote in the Democratic primaries or caucuses nationally supported Kerry. Kerry shot up after his Iowa caucus victory, approaching 50% in the week before New Hampshire and hitting 68% by the end of February. At the same time, Howard Dean's support collapsed, falling from 31% in early January to near zero by March, while John Edwards, who finished a surprisingly strong second in Iowa, saw his national support grow from 5% to 25%.

Kenski's analysis (which she presented at the 2004 meeting of the American Political Science Association) indicates that the impact of Iowa and New Hampshire that year had less to do with increased name recognition and more with changing voter evaluations of the candidates. Nationally, Kerry experienced a name recognition gain of roughly ten percentage points nationally after Iowa, but he was already known to more than 80% of Democratic primary voters beforehand. Kenski's analysis showed more dramatic shifts in voter ratings of Kerry and Dean. In other words, voters who already knew the candidates changed their minds about them after Iowa and New Hampshire. The NAES surveys also showed a dramatic shift in perceptions of Kerry and Dean's chances, respectively, of winning the Democratic nomination that neatly paralleled shifts in voter preference.

The 2004 Democratic race - as illustrated by the NAES data - is arguably one of the most dramatic examples of a phenomenon typical in the presidential primary season. Those who win early contests get a boost that helps them in subsequent contests. Those who study primary elections may disagree about the reasons for that "momentum" (see this review just last Sunday by the Boston Globe's Drake Bennett), but my sense is that it is more than just a blind "bandwagon" effect. Here are two reasons the early primaries shake up the race:

First, they effectively winnow the field. While news accounts may emphasize the large field of candidates before the early primaries (with an emphasis on front-runners), the coverage afterwards focuses far more intensely on the early winners. The result is that also-ran candidates either drop out entirely, or effectively drop out of site, a process that simplifies voter choices. If a voter had been torn between two candidates, and one of those candidates wins an early primary while the other finishes far back in the pack, that voter's decision gets much easier.

Second, and probably far more important, the horse race nature of the coverage leads voters to more positive evaluations of the winners and more negative evaluations of the losers. Media accounts portray the winners and their campaigns as competent and able, while the losers look hapless and faltering. Which set of characteristics would you want in a president? Not surprisingly, voters readily make the connection between winning and competence.

Either way, the implication is that the current horse-race preference numbers are not particularly meaningful as predictors of the outcome. An early loss is not necessarily fatal. Many early front runners (Reagan, Mondale, Clinton, both Bushes) have lost early primaries and bounced back to win their nominations, but the early primaries almost always change voter preferences.

So what effect will the especially "front-loaded" primary calendar have in 2008? I'll take that up in the next installment.


Chris S.:

I was wondering if you might address the issue of state primary polls vs. national primary polls. That NAES graph seems to show that, just before the '04 Iowa caucus, none of the candidates really had any momentum in one direction or the other. The standings in mid-Jan. were virtually unchanged from the standings in mid-Dec. It was only after Iowa that the standings were shaken up. Therefore, if one had only been following the *national* primary polls, one would have been extremely surprised by the extent to which Iowa shook up the race.

However, if memory serves, that was definitely *not* the case in the IA and NH polls. As I recall, Dean was already starting to slip in both IA and NH by December 2003. I just dug up these Zogby polls:



which show big gains in the Iowa polls for both Kerry and Edwards between December '03 and January '04. The polls grossly underestimated the margin by which Kerry and Edwards ultimately defeated Dean and Gephardt in Iowa, but the trendline was pretty clear.

So I guess the question is, should we be paying more attention to the Iowa and NH polls than we are to national polls? (even if the Iowa caucus is extremely difficult to poll)


Hmmm ...

Zogby shows Dean sprinting into the lead for NH by late August: http://www.zogby.com/news/ReadNews.dbm?ID=729

Also, Zogby shows that Lieberman was still winning SC in July 2003...

Zogby also shows that Dean was tied with Lieberman and Gephardt as of mid-July, just after Q2 fundraising results.


Again, very different methodologies I'm sure, but some people picked up on the Dean spike much earlier than that graph.


Dan Guild:

FYI - I have done an anaysis of how National Polls change after NH. In general, a win in NH is worth about 25 points (I have looked at only Democrats), though it matters whether they were a front runner are not.

My work is here:


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.