I currently reside in the People’s Republic of New York, and with the election of a full slate of democrats to every office that matters, things have only started getting worse. Every day brings a new anti-freedom initiative which all usually sail through for the governor’s signature (779 new laws in 2019-20 so far, admittedly not all are terrible). As a result I’ve been exploring ways to rank various other states by freedom to get a short list of where to move to.

Freedom in the 50 States is a collaboration by William Ruger and Jason Sorens by way of the CATO institute to rank states by freedom. They use their own, often intricate weighting system to give each category a weight relative to the other categories, denominated in dollars. Most data is current up to 2016, so many recent changes to law have not been captured by these measures. There are several categories they used (Personal, Fiscal, Regulatory and Abortion1) and took a composite of the weighted scores to come up with the final rankings. Each category is broken down into various components, for example, Fiscal freedom rankings include measures for Government Consumption, State taxation, Local taxation, Government debt, Cash and Security Assets, Effective number of Jurisdictions,  and Government Employment. There are far too many categories to go into each category in detail, but I’d like to examine just a few of the Fiscal variables and weighting methodology.

It is somewhat obvious how taxes can affect one’s personal freedom, but why do does government spending have any effect? And by what method do the authors decide on a weight for these variables?

According to the information provided by Ruger and Sorens, government spending affects liberty thus:

Non-tax-funded government spending is still a freedom issue because rents-fueled government growth (via federal grants, mineral revenues) can still crowd out the private sector. There’s a large literature on size of government and economic growth. Bergh & Henrekson (2011) survey the literature and find a robust association of government spending with subsequent growth in rich countries: for every additional 1% of GDP in government spending, annual average growth declines by at least 0.05 percentage points. This is in addition to the effects of taxation. We look at the effects of a standard-deviation increase in government consumption and investment as a share of personal income over 10 years, assuming the 0.05-percentage-point relationship. We calculate the discounted foregone growth over 10 years assuming a social discount rate of 5%. (Using a finite time horizon is necessary to impose finiteness on the number, but endogenous growth theory also suggests that the growth rate benefit of any exogenous variable dissipates eventually when per capita income reaches a new steady state — this is likely to happen over the course of a business cycle.) Then we divide by two because government employment presumably captures some of the same effects that other studies find via government spending

The loss of freedom makes sense to me knowing that government spending can crowd out private investment and that government spending tends to be less efficient. I’m not doing a deep dive on each variable, so I will assume the 0.05% given by Ruger and Sorens accurately represents the research. The authors also note, that this effect is an additional cost, beyond taxation- so no adjustment to this variable is made for taxation. The use of a 5% social discount rate (SDR – considered the opportunity cost of money) is a bit arbitrary, but every number going into these calculations could be a field of study on its own. Small changes in this number will have a big effect on the ‘victim cost’ of the spending. Maybe a survey of what the SDR was used in the past ~10 years for various projects would have been helpful. A ten-year horizon is also pretty arbitrary for the business cycle. It isn’t clear why this was chosen. Counter intuitively, the higher the authors set the SDR and the time horizon, the lower the victim cost of this variable goes.

Victim cost is essentially a dollar value put on the loss of freedom. These are relatively straight forward to calculate for things already in dollar amounts (e.g. spending), but probably get increasingly arbitrary for say, gun rights or abortion rights. The victim cost of one Standard deviation change in spending was calculated as follows:

=(15209136976823-15133300000000)*(0.95^10)*D5*100/2

The difference of the first two numbers appear to be something like the 0.5% of total personal income of the USA. I wish the authors had coded a variable for this or other variables, because it is used many times throughout their calculations, and would have made their sheet much easier to update. That is how I would have set it up, but that is why I’m and engineer and they’re researchers2. The 0.95 in this calculation is presumably (1-SDR), and raised to the power of ten, to account for the ten-year horizon the authors chose. D5 is the cell in excel which held the standard deviation for the spending (a weighting factor for how much variation between the states). The division by 2 is explained in their note as a fudge factor to account for double counting in the government employment effect.

This seems to be in error. While it is admirable to avoid double counting, the government employment factor comes out to be 2% of the total score, while state spending is 8% of the total score. They also apply the same factor of 2 to the government employment score. This is incorrect in 2 ways. Because the effects have a large difference in magnitude, the government employment score cannot account for 50% of the government spending score. It would have been more accurate to leave off the government employment score entirely and count only the government spending score. Secondly, if you’ve already discounted one variable to avoid double counting, you shouldn’t subtract something from the confounding variable as well, unless you take that into account when reducing the first score.

I would have proceeded as follows: Calculate government spending score (~16%3), calculate government employment score (~4%3). Assume somewhere between 0-100% of the government employment score is already counted in the government spending score (would need to do a literature review to determine the value), lets be lazy and assume 50%. Subtract that from the government spending score, result is that the total sum of these two variables is 16+4-4/2 or 18%, rather than the 10% assumed by the authors. Even if we assumed 100% of the effect of government employment is captured in the spending variable, in no way should the sum of these two variables be less than 16+4-4, or 16%. In Ruger and Sorens’ analysis the sum of these variables is only 16/2+4/2 = 10%. Roughly 60% of what it should be (assuming their calculations are otherwise correct).

This seems like a large error, especially when many of the variables they consider have very small effects, on the order of 0.1%. Lastly, I think the factor of 100 in this calculation maybe incorrect. If an additional 1% of government spending effects only 0.05% of GDP growth that means the whole of government spending (~38% of GDP) accounts for a total decline in growth of 1.9% annually. Surely that is no small thing, but the state and local portion of that is ‘only’ about 0.8%. I find it hard to believe that a change in 0.8% of growth accounts for 8% of the total freedom lost to state and local government. Surely taking ~10% of my income is significantly worse than that- though I suppose it depends on the time horizon one uses and the relative deviations between the states.

Speaking of taxes, let us dive into the rationale of that variable. Should be straight forward, right?

The original index’s weight for tax burden assumed that all taxes take away freedom. But in fact some taxpayers consent to at least some of the taxes that they pay, as long as the taxes are legal and generally paid by others. Therefore, taxation is not wholly a violation of their freedom.

In this version, we take account of this fact of consent to some taxes. Let’s assume that the current tax burden in each state represents the ideal point of the median voter. Positive theories of democracy would suggest that this is as good a guess about where public opinion lies as any. Then 50% of voters would prefer a higher tax burden (and the services it would finance), and 50% would prefer a lower tax burden. Right away, we can slash the tax burden weight in half, because 50% of voters nationally would not see the taxes they currently pay as any diminution of their freedom at all. Now, this move assumes that the median-dollar taxpayer is the same as the median voter. That is unlikely to be the case. In fact, the median-dollar taxpayer is likely to be somewhat wealthier than the median voter and thus more ideologically conservative and more hostile to taxation. Thus, if anything, slashing tax burden in half on these grounds is somewhat too aggressive.

But we’re not done yet. Of the 50% of voters/taxpayers who would prefer a lower tax burden, most of them would not see all of the taxes they pay as a diminution of their freedom. That is, they would be fully willing to pay a lower tax burden that is greater than zero. To illustrate the logic, assume a normal probability density function over possible tax burdens. Fifty percent of the curve lies to the left or right of the mean of the tax burden distribution.

Now, what are the losses experienced by those who prefer a lower tax burden than what currently exists in their state? The loss curve will look like a mirror image of the left side of the normal density function. Those who want zero taxation will see all of income taxed away as a loss of freedom. Those who want taxation of 2.5% of income will see 3.1% of income taxed away (mean of actual taxation minus 2.5%) as a loss of freedom. And so on. Because the loss function is a mirror image of the probability density function, the area under the loss curve is also 0.5. So only 5.6%/2=2.8% of personal income, in total, is a loss to those who prefer lower taxation. We can divide tax burden’s weight by two again, or by four in total.

This should slightly understate the actual victim cost of taxation for 2 reasons. First, the median taxpayer is richer than the median voter and probably a little more anti-tax. Second, it assumes that taxes pay only for desired public services, not rents. To make up for these omissions, we multiply the weight by 1.1.

However, about 6% of state and local taxes were returned to taxpayers due to federal deductibility. So long as federal deductibility remains in place, the weight needs to be multiplied by 0.94.

Got it?

I have two problems with the arguments presented here for the rationale of dividing this variable by a factor of 4. First, ceding the theory about relative distributions of people wanting more or less taxes (I would think the distributions are far from being a normal distribution with exactly 50% wanting higher and 50% wanting lower taxes, and then that of those wanting lower taxes, the weighted average of their desires is 50% of the current level of taxation) – they are still losing freedom regardless of whether they want to lose that freedom or not. I also have a problem with the 1.1 factor here- you already accounting for the payment of rents when calculating loss of freedom to interest payments, which was its own variable! Now you *are* double counting something for sure. I assume the federal deductibility of state and local taxes has changed somewhat from when the authors wrote this, but some value less than one would still be applicable today, and I have no reason to doubt the 6% figure given.

For me, 100% of the taxes taken out of my paycheck are a loss of freedom, the benefits, if any, are actually in the government spending pile, not the taxation pile, and would be accounted for by what I receive, not by what I pay for. If what I paid for was at all equivalent to the benefit I got from government, the schools in my state would be top tier instead of bottom tier. Thus, if I was changing this sheet to match my preferences (0% government spending), I would remove this factor of four, the 1.1 as well as the 0.96 factors because none of those things apply to me. Obviously Ruger and Sorens want this to be applicable on a wide basis, so they felt they needed to account for the general population’s preferences somehow, but personally, there is just no basis for reducing the effect of taxation at all. For those that like 75% of their tax burden, I would say leave things as it is, I’m guessing most libertarians at least would prefer something much closer to zero. For those progressives out there, if you find that you really wish the state taxed you more, then you could even make this variable negative, since you’re losing your freedom to be taxed more. This would result in the higher taxing states being ranked ‘better’ than the low tax states.

There are tens of variables left to consider but I think I will leave things here for this discussion. Each variable is highly complex and can be the subject of multiple research papers, it is unfortunate that such haphazard scaling has been used on some variables. I still give credit to the authors for making everything so transparent and the reasoning explained, it makes it much easier to apply my own biases when coming up with my own rankings.

When I went through all the variables and adjusted as I found necessary, it didn’t make a huge impact on the relative rankings at either the top or the bottom of the list. Sorens and Ruger’s rankings of the top and bottom 5:

  1. Florida
  2. New Hampshire
  3. Indiana
  4. Colorado
  5. Nevada
    […]
  1. Vermont
  2. New Jersey
  3. California
  4. Hawaii
  5. New York

My rankings for top and bottom 5 (without abortion):

  1. Florida
  2. New Hampshire
  3. North Dakota
  4. Tennessee
  5. South Dakota
    […]
  1. New Jersey
  2. California
  3. Vermont
  4. Hawaii
  5. New York

Adding in the rankings for abortion (prolife) changes things quite a bit, because the freedom cost of abortion is huge for the prolife position (around 20% of the total weight for my analysis). This makes sense if you consider the value of a life to be fairly high. The result for me is:

  1. North Dakota
  2. Oklahoma
  3. Indiana
  4. Missouri
  5. Michigan

The bottom 5 didn’t change at all, which isn’t surprising exactly, but interesting to consider that progressivism seems to be an all-or-nothing deal, but social conservatism can be included with economic conservatism and deregulation, or not.

I’m sure at least a few Glibs will tinker with the data themselves- the excel spreadsheet for the data can be downloaded here. If you’re not keen on the calculations behind all this, but know that you value, say, gun rights more than anything else, you can go to the personalize page and tinker with the weightings there without messing around in excel. I’ll be interested to see everyone else’s ratings and critique of the methodology.

Will I be moving to Florida or North Dakota? Maybe. There are potential job opportunities in both states for me, but I’m not sure I could stand having only MikeS around to talk to or having to deal with Florida Man on a regular basis. Like any analysis tool, this can’t tell you the right answer, only give you more information.

  1. I believe this was not included in the final product presented online, because they have options for whether one finds abortion to be an abrogation of rights or a right in itself. Perceptive of the authors.
  2. It is quite good for a researcher to even share this much detail on their calculations, so I congratulate them on sharing so much so freely, even if I think the spreadsheet could have been setup more logically.
  3. The actual amounts for these variables are closer to 14.8% and 3.7% because they are in both the denominator and numerator, but I didn’t want to confuse the main dialogue by accounting for this.