Experimenting with Development Economics

A Review of Debunking the Stereotype of the Lazy Welfare Recipient: Evidence from Cash Transfer Programs World Wide

As I scrolled through Twitter a couple of months ago, I found a New York Times review of a recent economics paper that drew my interest. It caught my eye because the lead author is Abhijit Banerjee. Banerjee is all the co-author (with Esther Duflo) of what is probably my favorite economics book written for a general audience, Poor Economics: A Radical Rethinking of the Way to Fight Global Poverty.

I was not necessarily interested in the subject matter of this particular paper (though it turned out to be a fun read), as much I wanted excuse to write about Banerjee's work. But it's a been a while since read Poor Economics (and the other books I reference), so this seemed like a reasonable entry point. 

Experimenting with Development Economics

Poor Economics and much of Banerjee's other work into a discipline called development economics. This is the study of economics in developing countries and economic growth. When I took a class on this in undergrad, I assumed we would examine at large macro-economic trends designed to increase the GDP and build a stable economy. Instead, the vast majority of the class was focussed on micro-economics and the effects of individual interventions on household decision-making. These interventions are often looking at a wide range of impacts, rather than the immediate return on gdp; some are about health, others are about education. 

I was immediately interested and hoped to study this more in graduate school. I never really got the chance (though, in my last year was able to do a little bit of work thinking about climate change in developing countries). I did take the opportunity to read some books and papers about development economic.

There are two giants in the world of development economics today: Jeffery Sachs and William Easterly. And, these two are rivals. Sachs, a professor at Columbia,  is a huge proponent of large foreign-aid programs, detailed in his books The End of Poverty and Common Wealth. He is an advisor to the U.N on the Millenium Challenge goals. He is generally optimistic about the progress that has been made in developing nations towards pulling people out of extreme poverty, improving health, and increasing educational obtainment. He thinks that there is great potential to continue, with aid from developed nations. 

Easterly, a professor across the town at NYU,  has argued that many aid programs have failed, often leaving the countries in worse shape in his book The White Man's Burden. In particular, he is against debt relief. Instead, he believes only in small, targeted aid programs, tackling individual issues with a track record of success. He prefers a bottom up approach to aid. I recall his book as mostly skeptical and pessimistic about the ability to help raise people out of poverty, but perhaps realistic.

About seven or eight years ago, I binge-read their books. I recall getting utterly swept up in each of their arguments as I read their book, but as I reflected on them had a very hard-time reconciling the opposing arguments. So when I read a review of Poor Economics in The Economist a couple years ago telling me there were some economists with ideas that split the difference, I was interested. 

Two major insights from Banerjee and Duflo's book stood out to me.  First is a simple model for how wealth develops in a household, that offer some insights into the circumstances when interventions can help. This model is summarized by a graph presented in their book, shown below.

The x-axis is current income, the y-axis is income in the future. The idea is that households on the left side of the are trapped as their future income is likely to be less than their current. Households on the right can expect to learn more in the future. This simple sigmoid shaped curve is a big deal, because previously economists assumed the whole graph looks like the right side of this graph; future income would always be more than current income.* With this model, if you can find an intervention that gets a household past the inflection point, you've pulled them out of what Banerjee and Duflo call the "poverty trap".

The second point is, if some interventions can in certain circumstances can pull households out of the poverty trap, then we should try to figure out what those are. The best way to do this rigorously is to run experiments. It turns out, many interventions roll out over time or can't possibly reach everyone who needs assistance. If aid organizations  randomly assign the interventions in a way that facilitates inference, economists can measure the results. This measurement can be used to find interventions that are effective and can then be replicated elsewhere. Their big take away is that instead of being dogmatic about which types of interventions will work or won't work (as some may say both Sachs and Easterly are), we should just run experiments to learn!

The paper

So with that background in mind, I checked out this paper. Banerjee worked with three other economists to investigate a common criticism of income-redistribution programs (think of welfare in the U.S.): these programs decrease the incentive to work. If low-income people are given income, what incentive would they have to work to get it? On the other hand, if you believe in something like the poverty trap, maybe the little bit on income can put these individuals in a position to work more and ultimately receive more income. For example, maybe the cash-transfer can serve to provide some capital to begin business, or allow a woman to afford child care to leave the home and work more. So to find out, they looked at seven different income-redistribution programs; and each was run as experiment.

The authors look at seven income redistribution programs, each implemented by a government (not an aid organization) in Honduras, Indonesia, Morocco, Mexico, Nicaragua, and the Philippines. And they have some strict criteria for the programs they include. Each program had be a true experiment with a control group that received no income benefit. The studies had to include both males and females. The programs could be an unconditional cash transfer or conditioned on the recipient following some behavior (such as enrolling their child in school). Many of these experiment were not randomized on an individual level, but on a village or town level (the program randomly chooses which town will receive the benefit, and then only individuals in the chosen town are eligible). So the authors only looked at studies with 40 of separate clusters of the benefit (this allows facilitates a statistical technique to calculate confidence intervals on parameters when observations are clustered rather than truly independent).

The seven programs also have some important difference between them. They began at different times, the earliest took place between 1998 and 1999, while the latest spanned 2009 to 2011. The largest population a program affected over over 18,000 people; the smallest affected around 1,400 (though for the study, the weighted each program equally). The largest cash-transfer accounted for nearly 20 percent of household consumption, while the smallest was 4 percent. 

The authors take each of the individual programs and group them into a common data set. They look at all working age adults, regardless of whether or not they are part of the labor force. To measure work, they create two variables: a binary variable of whether an individual worked in the last week (from when the survey was taken) and a count of the hours worked in the last week (putting individuals who did not work at zero). For their definition of labor, they include work within the household (frequently a family owned farm) or work for an outside employer.

Then, the analysis is pretty simple. They ran a regression, with the dependent variable as the measure of work, and the independent variable as whether the individual received the cash transfer, and some controls (such as gender, age, and fixed effects of the cluster). Strictly speaking the controls aren't necessary, as these are results from true experiments. Their inclusion adds statistical precision to the estimate. For the five studies that they had before and after measurements, they also test a difference-in-difference approach.

They results are pretty clear. They see very small point estimates for effect of the cash-transfer programs on either hours worked or whether an individual worked, none of which are statistically significant. One might be afraid that the lack of statistical significance comes from insufficient sample size, but because the point estimates are so close to zero the authors are confident that is not the case. They break down the analysis separately for within household and outside household work, and for gender, No effect. The results seem unambiguous, wealth transfer programs do not effect the incentive to work. 

Now, there are a couple caveats. They are only looking at developing countries. There is no reason to assume this analysis holds for developed nations. I also noticed the transfers included in the study are relatively small. Even 20 percent of consumption is not enough of a cash transfer to that I would think it allow individuals to truly exit the labor market. 

In fact, this points to the one criticism that I have of the paper. Instead of the thinking about the problem as just a binary treatment (did an individual receive a transfer at all), I would design this as an elasticity. That is, for every increase in the cash transfer, there is probably some effect on the willingness to work. Instead of measuring whether or not they received the transfer in the regression, use the quantity of the transfer received. And it might not be linear. For instance smaller cash transfers might be the type of thing that allows individuals enough slack to try something risky, or get some assistance with childcare (so there would be a positive impact), but really large transfer might eliminate the need to work all together (so there would be negative impact). However, this study I envision probably requires far richer data than was available. 

Anyway, this was a neat a paper, that took advantage of the many experiments that are currently being run in development economics, and dispelled a common myth about cash-transfer programs (of a certain size)

* Come to think of it, it doesn't seem like a coincidence a lot economist previously assumed strictly increasing first derivitives and deacreasing second derivitives in many problems, and that the logic totally changes when that asusmption is relaxed. It's very similar to prospect theory, as I discuss here