Why do “red” states that tend to vote Republican in presidential elections take more federal money than do Democratic-leaning, “blue” states? This surprising correlation between ideology and federal dependence has been often noted (see for instance here, here, and here). Indeed, this fact seems to be trotted out whenever we hear “what’s the matter with Kansas/Connecticut” arguments from the left/right, respectively. Are conservative states hypocritical and liberal states self-abnegating, or is there some deeper explanation?

First, let’s take a look at that correlation. In the chart below, I’ve plotted each state with federal grants to state and local governments in that state, as a percentage of personal income, for fiscal year 2007-8, on the Y axis, and percentage of the vote for Obama, McKinney, and Nader in the 2008 election on the X axis. The line through the points represents the least-squares line of best fit. As you can see, there does indeed appear to be a negative relationship between liberal ideology and acceptance of federal grants.

Is the correlation statistically significant? To see this, I ran a simple regression, estimating an equation of the form *Y = a + bX*, where *Y* is 2007-8 federal grants and *X* is state ideology as revealed in presidential voting behavior in 2008. The results are as follows:

------------------------------------------------------------------------------ Y: grants | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------- Liberalism | -.0671312 .0222499 -3.02 0.004 -.1118676 -.0223949 Intercept | 7.943979 1.160016 6.85 0.000 5.611611 10.27635 ------------------------------------------------------------------------------

So the answer is: Yes, the relationship is statistically significant. Every additional percentage point vote share for candidates of the left in the 2008 election is associated with a 0.07-percentage-point drop in federal grants as a percentage of personal income, plus or minus about 0.04 percentage points.

But might there be other factors that are correlated with ideology that would have an effect on federal grants? Two important ones stand out. First, smaller states should receive more grants as a percentage of their economy because the U.S. Senate overrepresents them. Second, states with more poverty might receive more federal grants, especially since TANF and (especially) Medicaid matching grants make up sizable proportions of state budgets. So next I run a multiple regression with the natural log of state population (since we might expect declining marginal effects of malapportionment on political power) and percentage of the state population living in households with incomes below the official poverty line. Here are the results:

----------------------------------------------------------------------------------- Y: grants | Coef. Std. Err. t P>|t| [95% Conf. Interval] ------------------+---------------------------------------------------------------- Log of Population | -.8330731 .1644707 -5.07 0.000 -1.164136 -.5020107 Poverty Rate | .3116748 .0612556 5.09 0.000 .1883737 .4349759 Liberalism | -.0118134 .0191028 -0.62 0.539 -.0502654 .0266386 Intercept | 13.74432 2.46959 5.57 0.000 8.773292 18.71535 -----------------------------------------------------------------------------------

Now ideology is no longer statistically significant (or anywhere close), and its substantive “effect” is also minimal. As expected, governments of larger states and less impoverished states receive less federal largesse. Conservative states are on average smaller and more impoverished, hence the spurious correlation we saw at first.

Is this the best we can do? After all, state and local governments receive federal grants for other programs. Education and highways come to mind. States with higher proportions of their population of school age and with lower population densities (probably approximating quite well the amount of transportation infrastructure needed per capita) will probably receive more federal grants. I next try including the proportion (0-1 scale) of the population between ages 5 and 22 and the natural log of population density. Here are the results.

----------------------------------------------------------------------------------- Y: grants | Coef. Std. Err. t P>|t| [95% Conf. Interval] ------------------+---------------------------------------------------------------- Log of Population | -.7441149 .2073956 -3.59 0.001 -1.162093 -.3261365 Poverty Rate | .3233147 .0623912 5.18 0.000 .1975735 .449056 Proportion 5-22 | 14.66201 14.64824 1.00 0.322 -14.85958 44.18361 Log of Density | -.1527231 .1749661 -0.87 0.387 -.5053442 .1998979 Liberalism | .0116293 .0244308 0.48 0.636 -.0376078 .0608663 Intercept | 8.12835 4.758223 1.71 0.095 -1.461218 17.71792 -----------------------------------------------------------------------------------

School-age population and population density have the expected partial correlations with grants, but they are not statistically significant at the 95% confidence level (still, they are more likely than not to have the expected association with grants). Liberalism remains insignificant. Because I have done other research finding that school-age population does associate with more state/local spending on education, and density does associate with less state/local spending on transportation, I’m going to keep them in the model for the next step.

In the next step, I look at the nonlinear association between ideology and grants. Why would I expect a nonlinear association? If very conservative and very liberal states tend to produce very “safe” senators (and possibly lots of safe House representatives), then these congresspeople should have significant bargaining power in the vote-trading negotiations that go on in Congress. Other congresspeople will trust them to be around to deliver on the promises they make – or to suffer the retaliation due if they fail to follow through on those promises. By contrast, senators and – perhaps, on average, House representatives – in swing states might be less safe and less likely to be able to make good deals to bring home bacon. Therefore, we should probably expect both very conservative and very liberal states to enjoy more federal grants than centrist states.

To test this hypothesis, I’ve run a fractional polynomial regression that finds the best fit of grants to ideology in two degrees, controlling for everything else. The numbers are too complex to explain in the space of a blog post, but I’ll post a figure that displays the results more effectively. The Y axis in this figure represents federal grants *if every state were identical in every respect except ideology*. In other words, the plotted federal grants numbers are hypothetical numbers devised after controlling for everything we’ve looked at other than ideology. On the X axis we have the same ideology variable as before.

The shaded bars represent the 95% confidence interval. What we find somewhat supports my hypothesis, except that extremely conservative states seem to be schizophrenic. Some take the federal bacon gladly (Louisiana, grants 9.4% of personal income in FY 2008), while others take surprisingly little given their social conditions (Idaho, grants 4.2% of personal income in FY 2008). Extremely liberal states cluster very tightly together with high federal grants relative to what social conditions would demand. My new hypothesis, given these results, is as follows. Safe representatives can bring home more federal bacon, but state governments in conservative states don’t necessarily want to accept these grants if it means enacting higher state government spending (and, probably, taxes). Governments of liberal states face no such cognitive dissonance.

In significant digits, how precise are the measurements on each of your axes.

Let me ask the same question a different way. What is the smallest number of grants any state received and are there any fractional grants?

I ask this because it speaks to the precision of the analysis. Your table suggests a precision of seven significant digits. I would suggest that this level of precision is likely false. Precision to seven significant digits is only possible if the least accurate quantity you utilized was accurate to seven significant digits.

I theorize, but don’t know for sure, that the least accurate quantity would be the number of grants per state. Precision to seven significant digits would be possible if there were fractional grants measurable to seven significant digits or if the smallest absolute number of grants any state received was in excess of 999,999.

Grants are measured as the number of dollars given by the federal government in a fiscal year, divided by total state personal income in dollars (times 100 to create a percentage). So there are lots and lots of significant digits, but of course the data probably aren’t accurate to the dollar for every state. The smallest grant amount any state received in FY 2008 was $1,342,511,000. But still – your broader point is true that the coefficient estimates shouldn’t be reported to so many digits. If this were a scholarly article, I would clean the tables up, but I just did a data dump out of the statistical software I used.

I am realizing now that I just don’t get how you came up with the y axis. Can you explain it in a little more detail.

In the first chart, the Y axis is federal grants as a % of personal income. In the second chart, it’s federal grants as a % of personal income once you control for population, poverty, school-age population, and density.

Thanks

Jason,

Any thoughts on trying to explain the difference between the conservative states that eat at the pig trough and those that don’t? Is it just a matter of poverty and state size, or could you find another explanatory variable (such as some of the rankings of state freedom and political culture you’ve already blogged about)?

That’s a very good question. A first glance at the data suggests there is definitely something regional going on, with southern states taking more and western states less. But it’s worth a more formal analysis.

In terms of numbers, one is dealing with a fairly small sample. Especially when you start talking about a handful of southern states. A more anecdotal explanation may be more apt than a statistical analysis.

Congressional delegations from southern states are typically more solidly Republican than from other states with Republican majorities. Those delegations could be more cooperative both intra and inter delegation wise, thus more horse trading and grant getting. It may be as simple as that. That small handful of skilled horse traders can skew your sample. I am merely curious; if you exclude them, what happens to the statistical analysis?

Well, I took these suggestions on and ran a quick analysis with a dummy variable for ex-Confederate states added. That variable wasn’t even close to statistical significance, so it looks as if Southern states don’t, all else equal, take more federal grants. The states that take the most grants, compared to what you would expect (the highest positive residuals), are, in order: Louisiana, Alaska, and Wyoming. These are all energy-rich states, which seems too unlikely to be coincidence. I know the federal government subsidizes oil & gas production, but why would that lead to more

intergovernmentalgrants? Curious.Nevada, Oklahoma, and Idaho take far less in federal grants than you would expect.