[Update as of 16 November 2015: this post has been edited to accurately reflect GiveDirectly’s role in the study. They did not design the randomization scheme and had not seen the results before the paper was released. Paul Niehaus from GiveDirectly has also informed me that future research with their clients will no longer involve randomization at the household level, and that they are offering cash transfers to the comparison group from the earlier study.]
Is it ethical to give cash transfers to some poor people while their equally poor neighbors get nothing? Johannes Haushofer, James Reisinger and Jeremy Shapiro just released a new study of this program design with GiveDirectly. They found that people who received nothing were less happy than they’d been before the program started. Anke Hoeffler has taken them to task for this, essentially arguing that these negative effects are so predictable that it’s unethical to study them, no matter how clever the research design.
As a basic point of research ethics, I agree that a study should never be designed to knowingly decrease participants’ wellbeing. However, this wasn’t the goal of this project. The new paper builds on earlier work by Haushofer & Shapiro (2013), which randomized access to cash transfers at the household level in order to examine both direct effects (on recipient households) and spillover effects (on their non-recipient neighbors). Ex ante, the expectation was that the neighbors might still benefit from the program, as recipient households shared resources through informal insurance networks. While the prospect that the program would create jealousy or unhappiness among the neighbors might seem obvious in one sense, other studies have indicated that transparent eligibility criteria for cash transfers can mitigate unhappiness among non-recipients, so it wasn’t clear that household-level randomization would make non-recipients either economically or psychologically worse off. Clinical equipose still applied.
As it turned out, the 2013 study found that the spillover group didn’t see any economic benefit from the program. This might explain part of the 2015 findings – people might be unhappy not only because their neighbors received cash grants, but also because they failed to share their resources. Finding a negative outcome doesn’t make the research design inherently unethical, but researchers now have an additional datapoint to consider when thinking about how future studies of cash transfer programs might be designed.
Given this new evidence, could there ever be a good reason to distribute cash transfers to fewer than 100% of eligible poor people in a town (as a long-term program design rather than a short-term research project)? It’s not clear to me that there is, although not for the reasons that Hoeffler points out. Most countries haven’t got the funding to offer a basic income guarantee to all citizens, so distributional questions are an important aspect of program design. Say a program only has the funding to provide a certain level of grants to 50% of eligible people. Should the grants be allocated to all the poor people in 50% of towns to avoid conflict among neighbors, perhaps at the cost of conflict between towns (c.f. Bates 1974)? Or might it be preferable to reduce the value of the grants by 50% and provide them to all eligible people in each town, with the possibility that this makes the amount too low to make a meaningful difference? Note that this latter solution is a formal version of what Haushofer & Shapiro believed might happen anyway – if recipient households are assumed to share their large grants with their neighbors, it’s simply a less efficient way to give everyone a small grant. It would be fascinating to see a study comparing these two targeting schemes.