Pre-Iowa

On the day before the Iowa caucuses, I sent out the following e-mail to friends, family and co-workers:

I may have mentioned to some of you that I am continually frustrated by the media's coverage of public opinion polling. Almost invariably, the media will report the latest poll as if it actually contains totally new information from the last poll ("So and So Widens Lead" or "Candidate Slips"). This, in general, is a poor interpretation of polls and is an especially poor interpretation when the changes have been small (as they have been in the past two weeks).

Instead, I think it's much better to think of all these polls of near simultaneous snapshots of a moving target. Thought the latest one may give you a slightly different perspective from the first one, if you really want to get a good idea of what the thing looks like, it's best to put all the snapshots together. Because of the nature of public opinion polling (polls with bigger samples have less error than small samples), an ordinary average isn't a good method for combining these polls because not all polls have the same error (as measured by the Standard Error, or more popularly, the Margin of Error). Luckily, as long as you have the sample sizes for all these polls, you can reasonably create an average, weighted by original sample size – meaning polls with bigger samples will be treated as slightly more accurate than polls with lower sample sizes. This method has the added benefit of allowing us to attach a new margin of error to our estimates.

All that said, looking at every publicly available poll (see pollster.com) in the final two weeks of the campaign before January 3rd and combining those polls as described above, we get the following results


DEMS (MOE = 1%)

Clinton – 28.8%
Obama – 27.13%
Edwards – 25.55%
Richardson – 5.22%
Biden – 4.55%
Other/DK – 8.76%



GOP (MOE = 1%)

Huckabee – 28.76%
Romney – 26.68%
McCain – 12.25%
Thompson – 10.08%
Paul – 7.34%
Giuliani – 6.89%
Other/DK – 8.01%


Notice that the Margins of Error are substantially lower than for most polls (which usually have MOEs on the order of 4 or even 5%).


On the GOP side, I would wager that this is a really good estimate of how things will turn out this evening. For the Dems, however, because of the crazy caucus rules, things could turn out quite differently. We can try a little data jujitsu to make some guesses though. Assuming that most of the 18% of voters who are currently leaning toward someone other than the big 3 will have to change to their affiliation to one of the top candidates, the question becomes, "How will these voters break down?" Most of the recent polling does not ask about second choices, but 4 of them do, and they all suggest that Edwards is the most popular second choice (to varying degrees). Using a similar weighted average method on these four polls gives us the estimate that if forced to choose between Clinton, Obama and Edwards as a second choice about half would choose Edwards, and the others would split pretty evenly for Clinton and Obama. If that's what happens, then the final results would look more like this:


Edwards – 33.5%
Clinton – 33%
Obama – 31%
Biden – 1%
Richardson – 1%

One other possibility. Recent reports have suggested that Richardson and Biden are encouraging their supporters to caucus for Obama, should they not be viable and Kucinich has already officially gone ahead and done this. This could cut into Edwards' lead as second choice candidate. It's hard to say how much this might matter, but if second-choicers break 40% to Edwards, 35% to Obama and 25% to Clinton (instead of 50, 25, 25), the outcome would be:

Clinton - 33%
Obama - 32.75%
Edwards - 32%

A bigger swing to Obama, however, would put him over the top.

One final wrinkle: for the Dems, Iowa doesn't report the vote total, only the State Delegate total which comes out of each precinct. Geographic quirks could skew the results (similar to the potential for the electoral college to skew results).