This post is the first part of a Nate Silver-esque miniseries of posts reporting the results of statistical analysis on a macropolitics topic: the size of the “liberty constituency” in each state. Essentially, what I’m trying to estimate here is the relative percentage of the voting population in each state that would consistently prefer libertarian candidates. It’s similar to what David Boaz and David Kirby have done to estimate the “libertarian vote” nationally, but the main differences are that a) I am ranking states, not giving an absolute percentage for the nation as a whole; and b) the numbers are based on actual voting and donation behavior, rather than responses to questions about issue positions.
Readers should be careful not to interpret these results as giving a ranking of the “most libertarian states.” Any such designation would have to be based on an examination of the entire ideological distribution of voters. We cannot assume identical distributions in each state. To take an extreme example, imagine a state composed of 20% hardcore anarcho-capitalists and 80% stark raving Hitler lovers. Would this be a more or less libertarian state than one comprised of 15% moderate libertarians, 15% populists, 35% conservatives, and 35% liberals? Probably less. I’m only measuring the proportions of libertarians in each state.
The three indicators I will use are: vote percentage for libertarian candidates in the 2008 presidential general election (Bob Barr, Ron Paul in Louisiana only (where he was on the ballot), and George Phillies in New Hampshire (where he was on the ballot)); per capita donors to the Ron Paul presidential campaign (from ronpaulgraphs.com); and “adjusted” percentage vote for Ron Paul in the 2008 presidential primaries. Of course, many if not most libertarians did not vote for or donate to any of these candidates. However, the size of the libertarian constituency in each state should correlate strongly with the percentage of voters that did. That’s all we need to come up with a relative ranking of states on size of libertarian constituency.
The first step I want to take is to adjust Ron Paul’s 2008 primary results for state institutional context. Some states have caucuses or conventions rather than primaries, and of course these elections took place at different points in the electoral cycle. Ron Paul did much better in caucuses and conventions than primaries, because his supporters were particularly motivated compared to the rest of the Republican field. He also did better when turnout was lower. Two states that held conventions, Hawaii and Wyoming, do not have results available. If a state held both caucuses/conventions and a primary, I use the primary results.
I took the log of Ron Paul’s percentage of the vote in each state (plus D.C.) and regressed it on an estimate of turnout (total votes cast divided by population – an ideal denominator would be registered voters, but that would be difficult to acquire for all 50 states, and it should make very little difference to the results), a dummy variable for caucus/convention, a dummy variable for whether the election was held after McCain clinched, and the log of the number of candidates in the race. (Taking the log of the dependent variable is necessary to make it impossible for predicted vote share to fall below zero and to ensure normality. I also tested for heteroskedasticity in this regression and found no evidence of it.) These are the results:
lnrp | Coef. Std. Err. t P>|t| [95% Conf. Interval]
turnout | -.0070837 .0221658 -0.32 0.751 -.051756 .0375885
caucus | 1.060498 .1955968 5.42 0.000 .6662991 1.454698
clinched | .6133622 .1627105 3.77 0.000 .2854407 .9412838
lncand | -.2069483 .1333964 -1.55 0.128 -.475791 .0618944
_cons | 1.999169 .2588205 7.72 0.000 1.477551 2.520788
Controlling for everything else, turnout actually does not predict Ron Paul’s vote share, but the results demonstrate that Paul did much better in caucuses than primaries and after McCain had clinched – and perhaps when the number of candidates on the ballot was smaller, although this result is not quite statistically significant. These last two results suggest that Paul was a protest vote for some people, and/or that some rather pro-Paul voters ended up going for one of the other candidates when it might have made a difference, and an agreeable alternative candidate was in the race (for instance, some libertarians supported Fred Thompson and Rudy Giuliani).
Now that we have estimated the effects of electoral institutions, we can adjust Ron Paul’s vote shares in each state accordingly and come up with a prediction of just how “pro-Ron Paul” each state was. Let us assume that every state had the exact same electoral institutions: primary not caucus, pre-clinching, with 5 candidates in the race, and a turnout of 6.27%. These are the median values on each variable. An “average” state (right on the regression line) would be predicted to give Ron Paul 5.06% of the vote under these conditions. We can add to this each state’s residual from the regression above (and convert out of logarithms) to get the percentage of the vote that Ron Paul would have won in that state under these conditions.
Here are the results:
|District of Columbia||7.824208|
New Hampshire and Idaho were the most pro-Ron Paul states, while Mississippi was the least. These results give us some insight into the composition of the Republican Party in each state. States with a more “establishment” bent, especially those in the South, gave fewer votes to Ron Paul, while states with more of an anti-Washington bent gave him more votes. Ron Paul’s good score in the District of Columbia helps demonstrate my point about ideological distributions. D.C. is a hostile place to libertarianism overall, but there is a small contingent of very politically aware libertarians there, and they made a noticeable mark on the (tiny) Republican primary there.
Of course, this is just one of three indicators I will use to compile an aggregate measure of size of the liberty constituency in each state. If there are some quirks in these data (I am surprised by how low Colorado scored), they should drop out when combined with other, independent measures of the concept. I will discuss how that can be done in Part 2 of the series.