In a couple of posts (here and here) last fall, I discussed the relationship between government size (usually measured by the ratio of government expenditures to GDP, or occasionally by public sector employment rates) and corruption. The main takeaway from the cross-country data is that, in apparent contradiction to the “big government causes corruption” hypothesis, government size is, if anything, negatively correlated with perceived corruption, as measured by the Corruption Perceptions Index (CPI) or similar sources. While that evidence does not decisively refute the claim that larger governments are more prone to corruption—the relevant studies have important limitations, and it’s at least possible that the result is due to reverse causation—it certainly seems to suggest that, when it comes to fighting corruption, too-small governments are probably a more significant problem than too-large governments.
Most of the research on the relationship between government size and corruption relies on international comparisons. But some work has performed single-country studies, attempting to identify the relationship between government size and corruption across sub-national jurisdiction. Some of this work reaches results that are largely consistent with the international research. For example, a recent analysis of 290 Swedish municipalities found that those municipalities with higher public expenditure levels had lower corruption, as reported in an anonymous survey of senior politicians and civil servants. But other research—particularly research on the United States—has found the opposite result: Within the U.S., when controlling for a number of other economic and demographic factors, states with larger public sectors seem to have higher corruption. What’s going on here?
Notably, the relevant research on the U.S. does not use perception indexes or surveys to measure corruption. Rather, this research takes advantage of the fact that the U.S. federal government (that is, the national government) often prosecutes state and local corruption. By making the key assumption that federal law enforcement effort is uniform across the country (or at least that any variation is not correlated with in-state corruption rates), researchers can treat per capita federal corruption conviction rates as a more objective measure of in-state corruption rates. In other words, if the above assumption holds, then higher federal conviction rates in a given state can be construed as evidence of more corruption in that state. Research using this approach, and attempting to control for other factors, has generally found that within the U.S., larger governments (measured by government spending and/or employment) is correlated with more per capita federal corruption convictions (see here, here, here, and here.)
Why does the correlation between government size and corruption look so different when looking across states within the U.S., as compared to looking across countries? One possibility, which would appeal to those who strongly believe that big government ought to be associated with more corruption, is that the U.S. studies are simply more reliable: The conviction-based corruption measure is more objective than subjective perception indexes, and the states are more similar, and therefore more comparable, than are countries.
It might also be the case that the U.S. government is already so large that the “corruptogenic” problems associated with excessively small governments—such as an underpaid, dysfunctional bureaucracy or massive wealth inequality—were not really problems (at least not in the time period covered by the studies). In other words, perhaps the relationship between government size and corruption is non-linear, with increases in government size past a certain point (which the U.S. has already reached) making the corruption problem worse. (Some evidence against this, however, is the fact that the negative relationship between government size and corruption in the cross-country data actually appears stronger for wealthier countries, which in turn tend to have larger governments.)
The above accounts accept the basic premise that the correlation between government size and corruption within the U.S. is, in fact, positive. But maybe it’s not. I can think of a number of explanations as to why the apparent relationship might be spurious. To me, the most interesting—and also the most persuasive—is that the allegedly objective measure of corruption (federal conviction rates) is in fact biased. Here’s the argument:
- Although we might like to imagine that U.S. federal prosecutors are politically neutral in deciding which public corruption cases to pursue, convincing research has found evidence of non-trivial partisan bias. In particular, public prosecutors in the (conservative Republican) George W. Bush administration were more likely to pursue charges against Democratic state and local officials than against fellow Republicans. This systematic empirical evidence is consistent with more anecdotal reports, in the wake of the outcry following the dismissal of a number of U.S. Attorneys in 2006, that several of these prosecutors were removed either for failing to pursue public corruption cases against Democrats, or for pursuing too aggressively public corruption cases against Republicans.
- All of the studies noted above, which find a correlation between government size (state-level public expenditure and/or public employment) use data primarily or exclusively from the 1980s. (The sample periods for these studies are 1977-1987, 1981-1984, or 1983-1987.) During this period (at least from 1981 onward), the White House was controlled by a (conservative) Republican President, Ronald Reagan, with the Department of Justice overseen by a hard-core Reagan loyalist, Attorney General Ed Meese.
- It’s highly plausible that in states controlled by (more left-leaning) Democrats, public spending and public employment were higher than in states controlled by (more conservative) Republicans. (This is a conjecture; I haven’t tried to examine the data to see whether it’s true. Obviously, if the evidence doesn’t bear out my supposition here, that cut strongly against the argument I’m spinning out here.)
- If all of the above are true, then the critical assumption undergirding the use of per capita federal corruption convictions as the measure of corruption would not hold. Instead, the results would be tainted by what the statistics jargon calls “omitted variable bias” (the “omitted variable” here being the partisan affiliation of the state and local officials). In short, even if corruption rates were the same in “bigger government” Democratic states and “smaller government” Republican states (or even if corruption rates were somewhat higher in the latter), the disproportionate targeting of Democratic public officials by a Republican Department of Justice would make per capita federal conviction rates higher in “big government” states than in “small government” states.
Is that the right explanation? It’s hard to say. One of the studies noted above (a 1992 study by Meier and Holbrook) did explicitly test for partisan targeting (by including a control variable for whether the state government was controlled by the opposite party as the White House); this analysis did indeed find evidence of partisan targeting in the Reagan years (though, interestingly, not in the Carter years). But including the partisan (non-)alignment between the White House and the state government did not eliminate the positive correlation between public sector employment rates and per capita federal corruption convictions. This would count as evidence against my conjecture that partisan bias in prosecution decisions explains why the U.S. state data looks so different from the international data. That said, the Meier & Holbrook did not include government expenditures as an explanatory variable–it focused on public sector employment, which means the result here might not really be comparable to the international results on government expenditures/GDP. And the result on government employment rates, though interesting, might suffer from a different sort of omitted variable bias: States with more public officials have more potential targets of federal prosecutions, even if individual officials’ probability of engaging in corruption is no higher than in states with fewer government employees (and even if the total volume of bribery and embezzlement, in dollar terms, is the same).
Unfortunately, the other studies in this vein–in particular, the studies that focus on public expenditures rather than public employment–don’t seem to test whether partisan bias in prosecutions explains some or all of the correlation between government size and corruption in the relevant sample period. But it seems like it easily would be fairly easy to do this. All one would need to do would be to re-run the regressions with a control for the partisan control of the state legislative and/or executive branches (as in the Meier & Holbrook study), or perhaps run a similar analysis again in a different time period (say, the Clinton years) to see if the results hold up. I don’t have the time to look into this at the moment. But perhaps some readers out there either know of other studies that have already done this, or have the time and expertise to do it themselves? I’d love to know whether this hypothesis is correct or not.