"Indiana Jonesing" and Research in Cape Town

I watched Raiders of the Lost Ark for the first time when I was about nine years old. Immediately afterwards, I wanted to be a swashbuckling archaeologist when I grew up. And immediately after that, some killjoy destroyed my dream by pointing out how real archeology can be tedious. The real thrills, apparently, are in the anticipation of finding something interesting and the elucidation of an interesting story through more digging.

Doing health economics is kind of similar. You may be surprised to learn that health econ research is not always about the glamorous women, fast cars and exotic foreign locales. Instead, most of the time the joy comes from testing your theory with data, refining your model to incorporate new insights and tools, and growing intellectually with each additional project. Its more of a private excitement, often understood only by other like-minded nerds.

I have to say that the project I was part of in Cape Town was as close to the Indiana Jones model as I've ever been in my still young research life. A bit of background: the AIDS and Society Research Unit at the University of Cape Town has collected information on about 250 HIV+ individuals in the Khayelitsha township who have been taking ARVs for the last 5 years or so. The survey began when the majority of individuals had been on treatment for about a year or two and three waves have been collected since 2004. The survey questionnaire includes information on treatment adherence, stigma, labor force participation, and various other aspects of socio-economic life.

My colleague Brendan Maughan-Brown and I are working on a project that looks at how ARV patients and their families cope with the loss of social grants. Basically, individuals starting ARVs are often extremely sick (CD4 counts < 200). The South African Government provides such individuals with a monthly disability grant that pays roughly 2.5 times the median salary for a black family (almost all of the sampled individuals are black).

In the last few years, quite a few individuals have been losing these grants as they return to health. The legalese behind grant provision offers a potential explanation for this: individuals in the Western Cape with CD4 counts below 200 and who are unable to work are eligible for grants. When these grants come up for renewal, healthier individuals are no longer able to receive the grant.

This is how it should work in practice. However, with some intensive analysis of the data, we found that health and other parameters that should predict grant/no-grant status over time have no explanatory power. There is no correlation between health improvements (measure a variety of different ways) and labor force participation and the loss of a grant over time. Not only that, only individuals below a certain monthly income level should be receiving grants (i.e., there is a means test based on assets and income). However, people across the income distribution appear equally likely to have or lose a grant! Hmm...

At the same time, we see in the data that households may engage in a variety of practices (more to come on this) to cope with the loss of the grant: its a lot of money and, given that unemployment rates are so high in the area (~40%!), the transition from welfare grant to the labor force is fraught with lots of friction. Thus, the loss of a grant to the household is a very, very large negative shock and people engage in all sorts of coping mechanisms to smooth consumption.

However, we cannot say with certainty which direction the association with the outcomes goes: even controlling for unobserved fixed factors, is it the loss of the grant that precipitated a certain behavior, or vice versa (or are both correlated with some other behavior that changes over time)?

In order to get at this, we decided that we needed to learn more about how disability grants are awarded, especially given that we found virtually no evidence in the data that the official province rules for grant distribution were being followed. This is where it got really fun: we spent days meeting social workers, individuals on ARVs, academics and other involved sorts to try and understand the 'grant generation process,' the stuff thats really going on on the ground, legalese notwithstanding.

In essence, we were trying to uncover sources of variation that would allow us to statistically separate the effect of the loss of the grant on various outcomes from the reverse causality and omitted (time-varying) variable bias. In searching for this instrumental variable, we met all sorts of interesting people with really bizarre stories of how grants are provided in practice. We have a few extremely interesting leads that we plan to follow through. I'll definitely keep you posted on the specifics of this in a later post.

All that aside, the main point is really this: how amazing is it to hunt for statistical identification by interviewing highly interesting and engaging people in the real world, following up on a string of leads like journalists or detectives, and running away from gigantic boulders?

To quote Professor Allen Kelley: "Research is fun!"