Did Peru’s CCT program halve its stunting rate?

On September 30, the Guardian ran several articles (see here, here, and an editorial here) linking the halving of Peru’s stunting rate (from 28 to 14% between mid-2000s and 2015) to its CCT program Juntos. Of course, it is great to hear that the share of stunted children in Peru declined dramatically over a short period. However, as I know that while CCT programs (conditional or not) have been successful in improving various outcomes including child health, the effect sizes are never this dramatic, I was curious to see whether the decline was part of a secular trend in Peru or actually could be attributed primarily to Juntos…

Juntos started in 2005. The first thing I wanted to see was whether there is a clear downward trend. This figure* shows data from DHS from 1992 until 2012: a 19 percentage-point (pp) decline from 37.1% to 18.1%, or about 1 pp per year. However, the decline between 1991-92 and 2004-06 was 8 pp in about 14 years, which accelerated to 11 pp in approximately seven years. So, that is some indication of acceleration of the rate of stunting decline that coincides with the introduction of Juntos.

But, of course, beware of before-after comparisons without a credible counterfactual. A quick search did not unearth any RCTs on the effects of Peru’s CCT program, but I found a couple of papers that use propensity score matching, which is useful not necessarily so much for exact impact estimates, but to take a peek at trends among non-beneficiaries, for whom there has been some attempt to make them look like Juntos beneficiaries. This briefing, by Escobal and Benites, using data from the Young Lives project shows that the height-for-age z-score (HAZ) among matched controls increased by 0.57 standard deviations (SD) between 2006 and 2009 – a huge increase in a short period. In comparison the treatment group had an increase of 0.71 SD. The numbers from 2002 to 2009 are more modest (0.22 vs. 0.33 SD), indicating a decrease in HAZ between 2002 and 2006. The bottom line here is that this evidence, as suggestive as it is, points to a strong secular trend in Peru for the decline in stunting.

A more recent paper that takes advantage of the Young Lives data by Sanchez, Melendez, and Behrman (2016) and uses pooled OLS with child fixed effects finds that there were no effects on HAZ or stunting, but there were large effects on severe stunting – on the order of more than 13 pp – for siblings of index children who were exposed to the program as children younger than three years old. The authors contrast their findings to Sanchez and Jaramillo (2012) who have similar findings during the first three years of the program (no impact on stunting, but on severe stunting) and point out that their control group comes from very poor districts, compared with other studies. One of those studies is Andersen et al. (2015), who also use matching and find no effects on HAZ, but find imprecise declines in stunting among those exposed to the program for longer than two years. Again, striking time trends among the non-treated are evident in Sanchez et al. (2016): the (severe) stunting rate among the index children in this comparison group declined from 53% (12%) to 28% (2%) between 2006 and 2009! The increase in HAZ during the same period (roughly 0.5 SD) is similar to that reported in Escobal and Benites (2012). It’s just that the gains among beneficiaries might have been larger… However, certainty is lacking here because (a) the estimates themselves for stunting are often not significant or only significant for sub-groups, and (b) matching often does not achieve any pre-treatment balance between the beneficiaries and the control group, so we have some doubts regarding identification.

A much more balanced piece comes from Alessandra Marini** in a blog post back in April 2016. Instead of causally linking the effects to the CCT program, she lists the three main factors that she thinks must have played a role. This points to more comprehensive nutrition initiatives (Peru was not a stranger to such programs before 2005, even if, perhaps, with less success), supply side improvements and results-based financing, along with help from the CCT program, by linking/harmonizing it with these surrounding measures.

It’s fair to say that we know Peru has done quite well in reducing stunting and severe stunting rates among its poor populations in the past 25 years, particularly since 2005, particularly among the poorest. We also know that Juntos started in 2005. While it is plausible that the CCT program contributed to the large and impressive gains cited in the Guardian, it seems unlikely that it is the main driver of Peru’s progress against malnutrition among children. The media can surely do a little better than before-after comparisons when making causal claims about program impacts…

On the bigger picture issue of the effect of cash transfers and stunting, there is evidence that they can help (see papers here). However, we know much less about whether the relative size of the income effect vs. the marginal effect that comes from the conditionality: it is again plausible that the conditions are effective – see this paper by Attanasio, Oppedisano, and Hernández (2015), which shows lower levels of preventive care visits and health outcomes among children excluded from the conditionality in a CCT program in Colombia. The question of “Cash or Condition” is still open for child health.

* The DHS might have moved from the CDC standards to the WHO standards in defining HAZ over time. If anyone has definitive word on when the change happened and how it affects over time comparability, please comment below.
**Correction: The blog post linked above is actually by Marini and Arias. Thanks for the readers for pointing it out quickly...