Some work on political corruption

I just finished reading three new papers dealing with the problem of political corruption in developing countries. Chris Blattman mentioned the first two by Ferraz and Finan here. While all are interesting, I think they are evidence that the fad search for "natural experiments" is getting out of control:

A paper by Ferraz and Finan looks at how the motivation of reelection can decrease corruption by elected officials. They use an interesting identification strategy: term limits in Brazil mean local politicians can only serve 2 terms. When a politician is in their second term, they can't get reelected. The authors find that mayors in their first term appropriate 27 percent less than those in their second term.

While they use some interesting techniques in the paper, their measure of corruption seems very suspect to me. From the authors:

Based on our readings of the [government corruption] reports, we codified the irregularities listed into several categories of corruption. For the purpose of coding, we define political corruption to be any irregularity associated with fraud in procurements, diversion of public funds, and over-invoicing. Specifically, we define a procurement to be irregular if: i) a required procurement was not executed; ii) the minimum number of bids was not attained; iii) there was evidence of fraud in the procurement process (e.g. use of bids from non-existing firms). We categorize diversion of public funds as any expenditure without proof of purchase or provision and/or direct evidence of diversion provided by the CGU. Finally, we define over-invoicing as any evidence that public goods and services were bought for a value above the market price.

While this seems like a good list of the things corrupt officials do, a person doing these things is not necessarily corrupt. I suspect the authors are also measuring incompetence and laziness, both of which I think would actually increase over a politicians tenure and as the prospect of reelection decreases.

Ferraz and Finan have another technically interesting paper on the pay of politicians and how this impacts performance. They find that higher wages increase quality of legislators. Again, they use an interesting identification strategy: local legislators in Brazil face salary caps according to municipal population. They argue this creates an exogenous variation in wages.

My problem with this paper involves a technical problem with using instrumental variables. We have to believe a potential IV meets the exclusion restriction. That is, the IV is itself not explaining what is going on, either through direct or indirect effects. Natural phenomenon work well for this since people can't affect things like rain or large scale land gradients.

Policy instruments don't always work so well. In this case, I think they're a major problem. The population of a municipality is certainly not exogenous to the qualities of the municipality. I would expect municipality population to be highly correlated with the level of education, suaveness and general cooperativeness of people in the municipality, and so the legislators that come from that municipality. All of these effects are directly related to how the authors measure the performance of legislators "as the number of bills submitted and the number of bills approved by the legislators in 2005". The IV is probably itself affecting the outcome, and so fails the exclusion restriction.

Finally, Manacorda, Miguel and Vigorito have a paper on how government transfers affect political support (I previously blogged about the paper here). They find that households that recieved a grant are 21 to 28 percentage points more likely to support the government.

The first problem is that the assignment of treatment and control is certainly not random, and there doesn't seem to be much evidence for their claim that "because assignment to the program near the threshold was nearly “as good as random”, we are able to circumvent the problems of reverse causality". They give no evidence that would make me believe that selection at the threshold was "as good as random".

I also find the follow-up survey question problematic. They ask people: “In relation to the previous government, do you believe that the current government is worse (-1), the same (0), better (+1)?”. That's not as bad as it could be, but why would we believe that people would answer this honestly, especially given that these people just got money from the government and, even though it was a university collecting the data, people know the government will look at it.

The biggest problem here though is that the authors don't have a baseline measure of political support before individuals recieved the program, so they can't be sure if the effect they find is really an increase in support of those funded, or a decrease in support from those not funded.

The authors address this on page 21, but don't have a strong argument against this interpretation. They use another dataset to predict support for the government absent an intervention and find that "the predicted support for the government is remarkably similar to the follow-up survey among ineligible households". Without an understanding of the change in opinion, this is very circumspect evidence. To put it another way, if predicted opinion is so good, why even use the quasi-experimental design they have?