In Freedom in the 50 States, we present some statistical results on the association between the three dimensions of freedom — fiscal, regulatory, and personal — and “net interstate migration,” that is, the number of movers into a state from other states minus the number of movers from a state to other states, divided by initial population. We found that all three dimensions are positively associated with net in-migration, usually statistically significantly so. Moreover, the substantive importance of the associations is large. A half-point increase in each of the three dimensions, measured in 2001, is associated with between two and five percentage points more in-migration from 2000 to 2011, as a percentage of 2000 population.
The results seem to imply that Americans value freedom and are willing to vote with their feet for it. Of course, some freedoms are not very plausibly related to migration. Tobacco, alcohol, and gambling laws can be evaded through travel or the black market. It seems unlikely that very many people at all will move from New York simply because of the high cigarette taxes. There are cheaper alternatives. And some freedoms with high symbolic importance, like eminent domain reform or legalization of sodomy (prior to 2003), are unlikely to drive anyone to move, simply because so few people are likely to suffer from their denial. Sodomy laws were almost never enforced, and eminent domain for private gain is rather rare even where totally unregulated.
But some other freedoms are plausibly related to migration. People definitely consider tax burden in their choice of a new home. Business regulation can dampen job opportunities, and people tend to move where the jobs are. Medical cannabis users move where their medicine is legal; gun enthusiasts move where their lifestyle is respected; same-sex couples move where they have legal rights; home-schooling parents move where they can educate with less state control.
In this blog post, I explore some other ways of testing whether the relationship between freedom and migration is causal. The first technique is something I call “matched-neighbors analysis.” The independent variables here, including freedom, represent the value of the variable for the given state minus the average value for its neighbors (technically, the weighted-average value, where the weights are the neighboring states’ populations — I’ve also tried using a pure average, with nearly identical results). This procedure is called “spatial differencing.” So the notion here is that states that are freer than their neighbors will be more likely to see net in-migration. Let’s see if that’s true.
First, some specs: regressions include all 50 states (unlike the results with just the Lower 48 included in the F50S study), all independent variables are standardized to mean zero and standard deviation one (so that the coefficient estimates represent the effect of a standard-deviation change in each variable), and the dependent variable, net migration, is measured over 2000-2012 instead of just 2000-2011 as in the original study. Here are the results:
The p values are in parentheses. The different equations reported here use different control variables: WRLURI (an index of land-use regulation), state CPI (cost of living), accommodations GDP, capital per worker, and state real income growth from 2000 to 2007 (latest available date). In all these equations, fiscal and regulatory policy are strongly statistically significant and positive. Personal freedom is positive but never quite statistically significant. Its coefficient is also quite a bit smaller than those on fiscal and regulatory policy. The coefficient on regulatory policy implies that a standard-deviation increase in regulatory policy score in 2001 is associated with a two-percentage-point increase in net migration rate over 2000 to 2012. Note that CPI is strongly correlated with fiscal freedom and land-use regulation, so models including the CPI may actually underestimate the total effect of fiscal and regulatory freedom on migration. (Land-use regulation on its own is not significant, perhaps because it is endogenous: land-use regulation drives away future in-migration, but states where people expect more in-migration are more likely to have strict land-use regulation to begin with.)
The second technique is spatial regression analysis, where “spatial lags” of the dependent and independent variables are included in a maximum-likelihood estimation, a so-called “spatial Durbin model” (Lesage and Pace 2009). The spatial weights matrix is the same as that above (weights are neighboring states’ population shares). The idea here is to test whether neighboring states’ freedom levels are associated with migration to a state. If freedom causes migration, state i’s neighboring states’ freedom levels should be negatively associated with migration to state i. Because of the small N and the doubling of the number of variables, statistical significance might well prove elusive in these models. (There is a “degrees of freedom” problem.) Here are the results:
Once the state CPI is included in the estimations, the results are mostly as predicted. Own state’s fiscal, regulatory, and (maybe) personal freedom is positively associated with net in-migration, while neighboring states’ freedoms are negatively associated with net in-migration. However, the results on personal freedom are again weak. Also, the spatial lag of fiscal freedom is not quite statistically significant at conventional levels. We can be most confident that neighboring states’ regulatory freedom is harmful for own state’s migration. Overall, the results seem to buttress a causal interpretation of fiscal and regulatory freedom’s association with in-migration.
Finally, I look at state-to-state (directed-dyadic) migration. The dependent variable here is a logarithmic transformation of the proportion of state i’s initial population that moved from state i to state j: Y = ln((mig/pop)/(1-(mig/pop))). State-to-state migration numbers come from the IRS and cover the years 2000-2010. It correlates with the Census Bureau monadic data over the same years at r=0.98, when summed to the state level. Consistent with other work using state-to-state data, I exclude Hawaii and Alaska (when I include them, they are extreme outliers).
I expect Y to be a positive function of state j’s freedom levels minus state i’s freedom levels and a negative function of state j’s to state i’s cost of living ratio. In addition, we have to control for the log of distance between the centroids of two states, since states that are far apart will send fewer migrants to each other, and for the sizes (populations) of the source and destination states. The estimates end up being a bit more unstable in these models than in the monadic ones, since state populations correlate with other features of those states, including freedom levels. Still, the basic results pop out pretty clearly. Here they are:
When CPI is excluded, the only freedom variable that is statistically significant is fiscal policy. When it is included, none of the freedom variables is individually statistically significant. However, I have a line in this table showing the p value on a test that the summed coefficients on the freedom variables amount to zero. This null hypothesis is consistently strongly rejected. Indeed, had I included overall freedom in these regressions, it would have been statistically significant. So we can conclude that it is difficult to apportion credit for driving migration to just one dimension of freedom, but we can be pretty confident that all freedom variables, taken together, associate positively with state-to-state migration, even controlling for the cost-of-living channel. Moreover, the sum of these coefficients comes close to the absolute size of the CPI coefficient and is much larger than the coefficient on accommodations GDP.
In summary, all of these results suggest that the relationship between at least economic freedom and interstate migration is causal. If some omitted variable were driving the results we report in the book, then we would not expect neighboring states’ freedoms to be negatively associated with own-state migration–but they (probably) are. The monadic, 50-state results suggest that personal freedom, at least over the entire 2000-2012 period, is likely not as important for migration as economic freedom, but the state-to-state results bring personal freedom back in as a potentially important predictor of migration. Indeed, it is almost statistically significant on its own in the final model above. Part of the reason for the difference might be that amenity-driven migration declined significantly after the 2007 recession in favor of economic migration or even just staying home and riding things out (especially for those with underwater mortgages). Indeed, the state-to-state migration results show little support for “moving to nice weather.” As we accumulate a couple more years of data, it will be possible to run models on 2000-2007 migration and 2007-2013 or 2014 migration separately. It is likely that amenities played a larger role in the earlier period.