In part two of this miniseries (part one here), I bring together different indicators of libertarian political support at the state level in order to estimate the size of the libertarian bloc in each state. In addition, the method I use to bring these indicators together will tell us two things: 1) Is state-level libertarianism a valid concept, or are we just picking up random noise? In other words, can we discern consistent patterns in the data that are best interpreted as reflecting libertarian ideology in state populations? 2) Which of the different indicators of libertarianism is most and least reliable?
The first of these is an important point. State politics scholars have often studied “opinion liberalism-conservatism” in state electorates and have shown that popular ideology influences state policies. I am not aware of any previous study that has looked at opinion libertarianism-populism at the state level before (and nothing peer-reviewed at the national level either!). If we can find, in real data on mass political behavior, that opinion libertarianism can be distinguished from, say, conservatism, that will be a real advance, something no one has demonstrated before.
The second of the things I hope to learn from this exercise could also be important, as it may show us how to continue to measure the size of the libertarian constituency, even after unique events such as the Ron Paul candidacy have passed and even without state-level survey data.
The method I use is principal component analysis (PCA), which is related to the factor analysis often used by psychologists to explore the dimensions of human intellect or personality. PCA chooses the best linear representations of a combination of variables, that is, the “common elements” that minimize the sum of squared errors across variables. The procedure can tell us the dimensionality of a set of variables, that is, how the common variance among the variables in this set can be reduced to a smaller set of variables. These latter variables, the dimensions underlying the dataset as a whole, are completely uncorrelated with each other. The way PCA works is by extracting the first component, which explains the most variance, then choosing the next component based on how well it explains the remaining variance after the first component is taken – and so on.
To see whether libertarian constituency exists as a concept and is distinct from mere liberalism-conservatism, I run PCA on eight variables: the adjusted Ron Paul vote share described in my last post, the number of Ron Paul donors per state, Libertarian Party vote in the 2008 presidential election, the mean LP vote share in the 1996-2004 presidential elections, an indicator of citizen opinion liberalism based on survey data from Berry et al., the Berry et al. measure of government opinion liberalism (based on surveys of state legislators), the Democrat/Green/Nader/minor socialist party vote share in the 2008 presidential election, and the mean Democrat/Green/Nader/minor socialist party vote share in the 1996-2004 elections. I expect the first four variables to load onto a single component, to be interpreted as “size of libertarian constituency,” while the latter four variables should load onto another component, to be interpreted as “size of liberal constituency.”
Here are the results of the PCA:
. pca lp08 donors rpcons2 lpvote citi6006 inst6006 demgrvote demgr08 [aweight=lnpop]
(sum of wgt is 7.5674e+02)
(principal components; 8 components retained)
Component Eigenvalue Difference Proportion Cumulative
1 3.58679 1.63772 0.4483 0.4483
2 1.94907 1.01805 0.2436 0.6920
3 0.93102 0.39043 0.1164 0.8084
The results show that two and only two components can explain the covariance of these eight variables.
Variable | 1 2
lp08 | -0.26730 0.41529
donors | -0.19594 0.55189
rpcons2 | 0.02690 0.43189
lpvote | -0.22137 0.45861
citi6006 | 0.47376 0.16200
inst6006 | 0.40256 0.14078
demgrvote | 0.49388 0.11632
demgr08 | 0.45828 0.25789
And what do you know? The first component, liberalism, tracks the latter four variables, while the second component, libertarianism, tracks the first four. What we have here is evidence that libertarian constituency is a Real Thing, distinct from liberalism-conservatism, and can be discovered in election results and donation statistics. Again, by construction the PCA extracts totally uncorrelated components. As a result, all of the variables load at least a little bit onto all of the components. To get a cleaner indicator of libertarianism, we can just take the first four variables and run a PCA on them, and do the same on the last four variables to get a cleaner indicator of liberalism. Here are the results:
Variable | 1
lp08 | 0.52354
donors | 0.58063
rpcons2 | 0.29841
lpvote | 0.54747
Variable | 1
citi6006 | 0.51624
inst6006 | 0.43417
demgrvote | 0.52494
demgr08 | 0.51907
So what we see here is that Ron Paul’s primary vote share, although it contributes something to the libertarianism component, is not a very reliable indicator on its own. Encouragingly, LP vote share in different elections is a reliable indicator – that’s something we can use in the future. Both 2008 numbers and 1996-2004 numbers contribute equally. The reason for that is that these numbers jump around from election to election based on idiosyncratic factors. For the liberalism indicator, each variable contributes about as much, but the one variable least directly tied to citizen political behavior, government ideology, is the least reliable indicator on its own.
So…drumroll please… Which states have the biggest libertarian constituencies? I’ve chosen to present the results in a chart, plotting libertarianism against liberalism, which has been reversed so that higher values represent smaller liberal constituencies and – effectively – larger conservative ones. The chart thus maps perfectly onto the Nolan Chart conceptualization of the political spectrum. Click the graph below for a bigger image.
These data seem to pass the “smell test.” Idaho, Utah, Wyoming, Nebraska, Oklahoma, and Alaska are the most conservative states, while Vermont, Rhode Island, Massachusetts, Hawaii, Connecticut, and New York are the most liberal states. The states with the most libertarians are Montana, Alaska, New Hampshire, and Idaho, with Nevada, Indiana, Georgia, Wyoming, Washington, Oregon, Utah, California, and Colorado following. (The Georgia result is due in large part to Neal Boortz, incidentally, which becomes apparent if you track the presidential results over time.) There is a slight positive correlation between states with fewer liberals and those with more libertarians.
Note that New Hampshire’s result is affected by the migration due to the Free State Project. Since only about 600 people have moved, however, I don’t think the FSP has affected their position dramatically. However, in separate research, I have found that New Hampshire towns with each additional Free Stater who had moved in gave two more votes to Ron Paul in the 2008 primary. Thus, Free Staters have had influence beyond their numbers. However, they (we) still have a way to go to catch up to Montana(*).
Now that we know libertarians actually exist in the general population in measurable numbers (although we still can’t put absolute figures down – those who vote LP are surely a subset of actual libertarian and libertarian-leaning voters), do libertarians actually influence politics? Are they so minuscule that they have no effect, or – consistent with the hypothesis that LP and Ron Paul voters are just the thin end of the tail of an ideological distribution – do they actually influence the overall policy regime of a state? This question will be answered in Part 3 of the series.
(*)Montana’s a bit odd, because in 2008 the Constitution Party put Ron Paul on their ballot in Montana, and he got more than 2% of the vote. I counted half of that toward the libertarian vote.