I had been sitting on these for a while now: Study links rise in tuberculosis in post-communist countries to IMF loans. (See also SciAm.) The discussion within the forum is farily interesting as well. My first reaction was similar to Steve Kass but when I looked at the study I was fairly impressed. The effect sizes were large and it potentially had the implication that adding more covariates would do little to dampen the effect sizes.
1. I would have been more comfortable seeing some kind of Extreme Bounds Analysis like Levine and Renelt. It's hard not to come to the conclusion that the authors picked the specification that resulted in a large effect size otherwise.
2. I'm surprised they were not able to control for health expenditures directly. DOTs coverage, number of physicians per capita and government spending might have been affected by IMF programs but I think I'd prefer to see a direct control with health expenditure spending. (This is all part of sensitivity analysis anyway so they might as well have put it in.)
3. The problem with determining causality is that even though these indicators move in the direction that impacts TB mortality the analyst still has to separate out the effects of the IMF program and the effects without the IMF program. These countries went to the IMF because of some kind of fiscal or exchange rate crisis and even if they had not gone to the IMF it is hard to establish that the indicators would have remained unchanged. For instance, if country A decreased government expenditures it is unclear whether this can be attributed to the IMF program per se or whether it was a result of having experienced some kind of crisis. I don't think the authors did a good job establishing causality.
While they were able to make a comparison of countries with an IMF program versus those without an IMF program, the correct comparison is one of countries in crisis with an IMF program and countries in crisis without IMF program. So, if country A is in a downward trend and then has to resort to using the IMF, it is possible that some of the indicators would have been on a downward trend as well and it's hard to see how to separate out the two with a fixed effects regression. Also, the fact that there is a trend or persistence in these variables might indicate that there is some serial correlation in the errors.
4. I was also a little concerned with some of the quality of the data. I would have thought mortality would be very accurately measured but in some countries like the Czech Republic for instance (see the link to Steve Kass) there is literally no change annually which means mortality rate was identical every year. My first reaction is that some bureaucrat just entered the previous year's number for reporting purposes. But comments indicate that TB is closely monitored so I must say I'm a little perplexed because it seems like such a coincidence for the proportion of deaths to be the same every year. Then again, we have to deal with what we have for data.
5. Overall I had some doubts about the robustness of the results but I thought the findings were interesting all the same. I think it calls for deeper investigation via interactions and mediator/moderator effects. (Yes, someone please give me a grant for this!)
In any case, I sat on this for awhile thinking I could get some research out of it and here's a preliminary abstract:
Stuckler, King and Basu (2008) show that post communist countries who have been in IMF programs have experienced significantly worse tuberculosis outcomes. Their fndings are robust to inclusions of various covariates and the effect sizes large. This paper explores the relationship between IMF programs and tuberculosis mortality using longitudinal methods. Specifically, the errors from a fixed effects regression are assumed to have an autoregressive pattern over time. Once these errors are controlled the relationship between IMF programs and tuberculosis
outcomes is no longer as robust.
Note, I remain purposely vague and technical here just because I hate being wrong.