Self-Fulfilling Prophecies and the Iowa Caucus
Originally published on Blogger.
A few days ago Nate Silver made the following intriguing comments on the Iowa Caucus (emphasis added):
There are extremely strong incentives for supporters of Mrs. Bachmann, Mr. Santorum and Mr. Perry to behave tactically, throwing their weight behind whichever one appears to have the best chance of finishing in the top two. What that means is that if any of these candidates appear to have any momentum at all during the final week of the campaign, their support could grow quite quickly as other voters jump on the bandwagon.
This is also a case in which the polling may actually influence voter behavior. In particular, if one of these candidates does well in the highly influential Des Moines Register poll that should be published on New Year’s Eve or thereabouts, that candidate might be a pretty good bet to overperform polling as voters use that as a cue on caucus night to determine which one is most viable...
I’m not sure that this theory actually makes any sense... But it may not matter if the theory is true. If voters are looking for anything to break the logjam between these candidates, mere speculation that one of them has momentum could prove to be a self-fulfilling prophecy.
What's most interesting about this is the possibility that even a methodologically flawed or misleading poll, provided that it is given credence, could coordinate expectations on one of these three candidates and result in a surge of support.
In fact, this seems to be precisely what has happened. A CNN/Time poll covering the period December 21-27 revealed Santorum to be in third place with 16% of the vote. This was an outlier at the time, and was sharply criticized by Tom Jensen of PPP and by Nate himself for surveying only registered Republicans:
What’s wrong with using a list of Republican voters for a Republican caucus poll? The answer is that it’s extremely easy for independent and Democratic voters to register or re-register as Republicans at the caucus site. Historically, a fair number of independent voters do this.
According to entrance polls in Iowa in 2008, for instance, about 15 percent of participants in the Republican caucus identified themselves as independents or Democrats on the way into the caucus site... Most other pollsters are making some attempt to account for these voters. They are anticipating that the fraction of independents and Democrats will be at least as high as it was in 2008 if not a little higher, which would make sense since Republicans do not have a competitive Democratic caucus to compete with this year.
The recent Public Policy Polling survey, for instance, estimated that 24 percent of Iowa caucus participants are currently registered as independents or Democrats and will re-register as Republicans at the caucuses. This month’s Washington Post/ABC News poll put the fraction at 18 percent. There is room to debate what the right number is but it will certainly not be zero, as the CNN poll assumes.
Since few independents and Democrats are inclined to vote for Santorum, the CNN/Time poll very likely exaggerated the level of support he enjoyed at the time. But despite this, it contributed to expectations of a surge which seem to have become self-fulfilling. The Des Moines Register poll released last night confirms this, with Santorum rising sharply from 10% on the 27th all the way to 22% four days later. This survey, conducted by the highly regarded Ann Selzer, has historically been among the most reliable of Iowa polls.
Did a misleading poll based on an unsound sample shift expectations in such a manner as to fulfill it's own flawed forecast? Tom Jensen certainly appears to think so:
Selzer had Santorum at 9% Tu-W. We had him at 10% M-Tu. Surge quite possibly generated by CNN poll that was quite possibly wrong... If CNN had shown Perry at 15% and he got all the momentum stories, the buzz in Iowa might be all about him this weekend.
The CNN/Time poll may also have given Romney an expectational boost at the expense of Paul by excluding independents from the survey. As Tom Jensen noted in his response, Romney was ahead of Paul in the restricted sample of the PPP poll, but quite clearly behind overall on December 27. It's an interesting example of how a seemingly innocuous methodological decision on a single primary poll could end up having major ramifications for the direction of the country.
The mechanisms at work here have some broader implications. They reveal the potential value to candidates (or their supporters) of manipulating prices in prediction markets such as Intrade, which have come to be closely monitored indicators of candidate viability. And they appear in all sorts of other contexts, from sovereign debt crises to speculative currency attacks.
In fact, any borrower who has financed long-dated assets with short term liabilities needs to periodically roll over debt, and the willingness of investors to facilitate this depends on their beliefs about whether other investors will continue to facilitate it in the future. These expectations are subject to capricious change, often as a result of small and seemingly unimportant triggers. The Iowa caucus illustrates the phenomenon, and the Eurozone debt crisis demonstrates its broader relevance.