Reading this excellent post by Lynn Vavreck reminded me of a question that sometimes comes up (for example, in our recent q-and-a at the Larchmont public library): Who are those undecided voters? Are they thoughtful or clueless? John Dickerson characterizes them as thoughtful moderates, while liberal blogger Atrios characterizes them as “about politics at least, just idiots.” Conservative commenter Jim Geraghty characterizes undecided voters as “low-hanging fruit” for Romney, but our own John Sides estimates they will split about 50-50. (Here’s more from John on independent voters.)
I don’t have any data to add to this discussion but I thought it might help to give my general framework for thinking about vote choice, which is based on the latent-variable model for discrete-data regression (a standard idea in statistics and econometrics, explained, for example, in chapter 6 of my book with Jennifer).
The basic model is additive. You end up some continuous political attitude that is a positive number if you’re generally a Republican and negative if you’re a Democrat (remember, the positive numbers are on the right side of the number line). Call this number A. 0 is perfectly neutral, 0.2 is weakly Republican, -1.5 is fairly strongly Democratic, 4.4 is strongly Republican, and so forth. You start at 0 and then start adding and subtracting things: some number of points if each of your parents is a Republican or Democrat, some small amount based on your family’s economic status when you were growing up and your family’s economic situation now, add or subtract something based on what job you have, what the political situation was in your formative years (remember the story of the “Jimmy Carter Republican”), the political character of where you live, and, of course, current economic conditions and the efforts of the campaigns themselves. And some other things I’ve forgotten to mention. Add these all up and you get a number: if it’s zero, you’re equally likely to vote for either candidate (that is, if you vote at all), if it’s positive you’re more likely to vote R, if it’s negative you’re more likely to vote D. The larger the number is in absolute value, the more predictable your vote is.
Alongside all this is a multiplier—-call it B—-a positive number expressing your interest in politics: if the average value of this multiplier is 1, some people have low values near zero if they don’t care at all, other people are passionate about the political race and have high numbers.
Now you multiply A times B. If the product is not too large (between -1 and 1, say), you get an undecided or unpredictable voter, someone who could possibly vote for either candidate. (I’m ignoring one more complexity, which is that the “undecided” survey response can just reflect a certain way of answering the question and not a true unpredictability on the respondent’s part.)
So . . . the undecided or swingable voters are some mix of people who don’t know much or don’t care much about politics, along with a group of people who care and are informed but have a mix of views (for example, they could favor economic redistribution but have conservative views on social issues). As Delia and I found (see figure 8 of this article), more informed voters tend to have more coherent views on different political issues, so there is something to the idea that less informed people are overrepresented among political centrists.
The model I have described above is, I believe, the usual way that political scientists think about vote choice. For me, an advantage of this approach is that it gives an intuition underlying the general phenomenon of uniform swing (opinion and vote swings tend to happen everywhere, not just in some small demographic slice or some small group of states) and also gives a sense of the diversity of a group of people whose political leanings just happen to mostly cancel, leaving them in the center.