Government Size and Economic Prosperity

Recently Ed Dolan had a post on his blog on Medium arguing that small government was not in fact conducive to prosperity (https://medium.com/@dolanecon/does-the-government-that-governs-least-really-govern-best-142106728cba), as libertarians often argue. Specifically, he shows that the size of government is actually negatively correlated with measures of prosperity, freedom, and quality of government; on the other hand, quality of governance was positively associated with per capita GDP. I came across this post, I should note, when Tyler Cowen linked to it on his blog as evidence for “state capacity libertarianism.”


But there’s a problem with Dolan’s analysis, or at least the interpretation he provides of it with respect to the role of government size in economic prosperity. It may seem pedantic to repeat the old canard that ‘correlation does not equal causation,’ but the problem is more than that: there is an equally plausible explanation for the correlation he observes in which causation runs in the opposite direction. That is, it is plausible that prosperity leads to the enlargement of the state, rather than vice versa. In fact, it may be the case that state enlargement has a negative effect on economic growth, but economic growth itself has a larger positive effect on the size of the government, leading to a positive correlation. This is precisely what I argue is the case, and it’s a major flaw in Cowen’s case for ‘state capacity libertarianism.’ I think ‘state capacity’ is more a consequence than a cause of prosperity. A big government, with a generous welfare state and extensive provision of public services, is a societal luxury good. When societies get rich, they can afford to buy these things, but they do not, in and of themselves, promote growth, and in fact more likely than not impede it. The argument against small government or in favor of more ‘capacious’ government in order to bring about greater prosperity, I contend, is akin to someone arguing that, from the correlation between luxury car ownership and wealth, we can conclude that buying a luxury car makes one wealthier.

Hence, looking at mere correlation is plainly the wrong way to assess the effect of size of government on economic growth.  A couple other way came to mind to get a better idea of what the real relationship is between government size and economic growth, so I downloaded some World Bank data on per capita GDP, per capita GDP growth, and government spending as a % of GDP for all recorded nations going back to 1960 (as far back as the data goes). Here’s what I found in my own little rather cursory analysis.

First, I ran a linear regression for each country in the data set using both per capita GDP and government spending (again, as a % of GDP) as regressors, which I’ll hence forth just call G, and per capita GDP growth rate as the dependent variable. If G really does have a positive effect on per capita GDP, then we should expect countries with higher values of G to also have a higher rate of per capita GDP growth, after controlling for current per capita GDP (which we would expect to have a negative relationship with growth, since more developed countries tend to grow more slowly). Instead, I found the opposite tends to be the case:

Here I’m plotting the regression coefficients against the p-values for G as a predictor of per capita GDP growth rate (also controlling for current per capita GDP) across all 218 countries in the data set.
Next, I looked at whether increases in G tended to predict an increase or decrease in the rate of economic growth, controlling for per capita GDP.

This time, we’re looking at the coefficient and p-value for % change in G relative to the previous year on per capita GDP growth rate.
Given that individual countries often didn’t have very much data, I tried this using each ‘country-year’ in the data set as an independent data point. Plotting current year per capita GDP growth against G for all 60 years across all countries gives us the following plot.

For what it’s worth, the coefficient on G is -0.14, and is, according to the Pearson test, was statistically significant with a P-value of pretty much 0 (or as close to 0 as is you can get in R, which I’m using to do these computations). Of course, the plot looks a bit iffy, and may not pass an ‘eyeball test.’ The concentration of data points makes it hard to see if there’s a trend, and in light of the recommendations of the American Statistical Association, perhaps we ought not trust p-values.

Given that, visually speaking, the trend looks questionable, I decided to look at what the relationship was between change in G and per capita GDP growth rate up to years in the future. I then plotted the coefficient of G on per capita GDP growth rate n years into the future:

The size of each point is the negative square root of the log of its p-value (basically, the bigger the point, the more significant) Red points have p-values below 0.01. What we see is that the coefficient of G on per capita GDP growth appears negative and significant at first, then declines in magnitude (and significance) until it converges to 0 about 4 years after the increase in G. It seems pretty unlikely to me that, if there were no relationship between G and per capita GDP growth, we would see such a trend by chance.

Finally, I did something similar with per capita GDP growth rate as the independent variable to see how well it predicted growth in G in the future up to 10 years.

The coefficients are all positive and half have p-values below 0.01. In other words, increases in per capita GDP tend to predict future higher in G, while increases in G tend to predict lower increases in per capita GDP.

Overall, what my cursory analysis suggests is that, if anything, an increase in G has a negative effect on per capita GDP, not a positive one. Now, plenty of caveats apply: this is macoreconomic data after all, and there are myriad possible confounders, and the trends are not consistently negative across all countries. But this is macroeconomics data, so beggars can’t be choosers, and it is clear that the more data we have on a country, the more likely it is to have a negative relationship between annual change in G and per capita GDP growth, once we control for current per capita GDP. What trends we do see seem unlikely to be explained by chance alone. The correlation between G and per capita GDP is thus most likely, as I hypothesized early, due to the effect prosperity has on government spending, not the other way around, and given what data we have, more likely than not, government size itself does indeed have a negative effect on economic growth.

The data I used here (government spending as % of GDP,  per capita GDP, and per capita growth as % of GDP, all by country going back to 1960) are from the world bank: https://data.worldbank.org/

Comments

Popular posts from this blog

How Do We Prepare for an Epidemic without Bankrupting Ourselves?

Did Peter Turchin predict 2020?

How "Political" Should the Pandemic Be?