r.i.c.e.

This afternoon in Washington DC, IFPRI will release its new 2014-2015 Global Food Policy Report.  The report covers an impressive range of important topics.  Among them, is a chapter that I was honored to be asked to contribute, in collaboration with Lawrence Haddad: our chapter reviews the accumulating new evidence for the old observation that sanitation policy can be -- and should be -- part of nutrition policy.

Much of the new evidence that has emerged in the last year or so is from careful observational studies, such as Derek Headey and John Hoddinott's work about Nepal, or Audrie Lin and coauthors in Bangladesh.  But a new -- and surprising large -- set of intervention studies is also emerging.

One of the central points of our chapter is a reflection on a special challenge for intervention studies: you can only learn about the anthropometric consequences of a change in open defecation if you succeed in changing open defecation.  That is what the arrow diagram above is about: even if all you care about is the "second stage" effect of open defecation on child height, your ability to learn about it depends on your ability to achieve a "first stage" effect on open defecation.  Put bluntly, we would learn nothing about the effect of open defecation on child height from not changing open defecation.

This matters because we wouldn't expect the effect of open defecation on child height to be the same everywhere: unsurprisingly, the association between open defecation and child height is steeper where population density is greater, on average.  But open defecation is very dense in India, a place where changing "first stage" open defecation behavior has proven very difficult.  As a result, the evidence collectively available from everybody's intervention research may systematically miss the places where the effect of open defecation on nutrition will be most important.

One useful recent paper takes this "first stage" problem seriously.  Paul Gertler, Manisha Shah, Maria Laura Alzua, Lisa Cameron, Sebastian Martinez, and Sumeet Patil have combined data from three field experiments to construct an "Instrumental Variables" estimate of the effect of open defecation on child height.  Such an approach takes into consideration the first stage effect of an intervention on sanitation behavior. (We didn't include it in our IFPRI chapter because the paper wasn't out yet.)

Their average results combine three experiments, from India, Mali, and Indonesia.  Although we should keep in mind that we might actually expect the effect to be different in these different places, it is still important to note their summary estimate: a village switching from everybody defecating in the open to nobody defecating in the open would make children about 0.44 height-for-age points taller.  This is strikingly similar to the 0.45 effect size necessary for open defecation to completely statistically account for the India-Africa average child height gap. Of course, we shouldn't take any of these numbers too literally (for example, there are differences in population density and rural or urban settings), but the point is that their method uses randomized data to produce an effect that is big and important, and very closely matching what we see in population level data.

Yet, as our chapter concludes, the best solution to the problem of weak first stages is to get serious about learning how to change open defecation behavior in rural India!