The old paper focused on explaining declines in underweight rates over the period 1970-1996 for 63 countries. The update (supported by the Transform Nutrition consortium) focuses on the period 1970-2012 for 116 countries and switches attention to explaining declines in stunting.
We use an unbalanced panel estimation method to estimate two models:
- stunting explained by underlying determinants (calorie supply, the percent of calories from non-staples, women's secondary education, the ratio of female to male life expectancy to pick up gender inequalities in health and other inputs, safe water access and improved sanitation access). Sanitation and percent calories from non-staples are new additions to the regressors from 2000.
- stunting explained by basic determinants (GDP per capita, governance variables, and demographic indicators relating to rural/urban and age structure of the population)
We test for the endogeneity of all explanatory variables to make sure we are getting and independent effect of the right hand side variables on the dependent variable. Where we detect endogeneity we use instrumental variables to (try to) manage it.
We find:
- A reassuring validation of the UNICEF food-care-health environment model. For the 1970-2012 period and across all 116 countries, the contributions of our sets of variable proxies for food, care and health environment each approximate to one third. For example, calorie supply per capita and the percent of calories from non-staples together account for 1/3 of all the declines in stunting observed (we can explain nearly all of the declines).
- We find some regional variations. The broad results hold for the regions, but we do find some notable variations. For South Asia, the gender equity variable is twice as powerful in reducing stunting as in the other countries. For South America, food supply is stronger in reducing stunting than in other regions and in Sub Saharan Africa the percent of calories from non staples has no effect on stunting.
- We found no differences between the pre MDG and MDG periods. There is little difference between the results for the years up to 2000 compared to the years 2000-2012. In many ways this is encouraging. It suggests that increased inputs to stunting reduction (and the MDGs are widely thought to have resulted in greater investment in the social sectors) are showing no decline in their ability to reduce stunting (and we do allow for non-linearities in the regressions).
- Do priorities change if we use stunting or underweight? Put another way, are the results different for underweight? Underweight is the MDG indicator but stunting is likely to be a key post MDG indicator. We find that there are some differences, but they are not huge. The main difference is that safe water access is a more important driver of underweight, but this does not shift the focus of priorities too much because the scope for increasing safe water coverage remains lower than for the other determinants (such as sanitation or female secondary education enrolment).
- What are the priorities for the future? Any prioritisation exercise would be most useful at the country level, but we have too few observations per country to do this (although it is increasingly possible with accumulating waves of DHS and MICS survey data). In the paper, priorities are determined by two things: (a) how strong the underlying determinant has been in the past at driving down stunting and (b) the scope for further declines in the underlying determinant. On this basis we conclude that the priorities for South Asia are: access to sanitation, diversity of food supply and gender equality. For Sub Saharan Africa we conclude that the priorities are: access to sanitation, women’s secondary education and gender equality.
- Does governance matter for stunting reduction? There are too few countries for which we have nutrition governance indicators (see HANCI), so we use general ICRG governance indicators from the PRS Group. When entered one by one, we find that most of the 5 governance variables are positively associated with a decline in stunting--even after including, GDP per capita, demographics and country fixed effects. But when we include the governance variables together they are insignificant. Most interestingly they seem to work through the access to safe water indicator. More work is needed here to unpack these effects, but they are the first time we have seen these types of variables included in a big picture examination of stunting declines
So, some interesting results that speak to a number of big questions. But the study has a number of flaws (it is currently in review at a journal).
First, there is no variable for strength of/access to health system. We searched high and low and dismissed proxy after proxy. Some were available for a subset of countries and years, but they described use, not access, and are self reported. We tried to minimise the inclusion of these kinds of variables for conceptual consistency and econometric reasons. We do have secondary school enrolment of girls and this is a choice variable but it is available for all years and countries and is not self reported. If we did have such a health system variable it would surely show up as an important contributor to stunting reduction. Our results do not imply health systems are not important, but they do highlight a key data gap.
Second, we have no cost data. We can say that it requires a 9.7% increase in access to safe water to generate a one percentage point reduction in stunting rates and that it requires an 8% increase in access to improved sanitation to do the same, but we cannot compare the relative costs of increasing the underlying determinants by these amounts. This makes the priority setting less useful for policymakers.
Third, the study is a cross-country regression. The pendulum has swung from "these are all the rage" in the 90s's to "these are worthless" in this decade. The truth is surely in-between. These kinds of analyses give us useful big picture views which are needed every now and then, but they are not fine grained enough and do not have the causal power we would like to see to represent any more than a rough guide to the landscape.
Fourth, the study is easily misinterpreted. Does it mean nutrition specific and nutrition sensitive interventions are unimportant? Absolutely not! What the study does say is that the underlying determinants (a big resource flow, which nutrition sensitive interventions are chiseled from) must not be overlooked in the fight against stunting. Does it mean that health systems are not important? Of course not, micro studies show how important health systems are. Does it mean income growth will solve stunting? No. The long term elasticities are around 0.6 (i.e. a 10% increase in GDP/capita is associated with a 6% decline in stunting rate) and the short term ones are around 0.2. Income growth is very important, but it is not nearly enough.
The final flaw is a common enough one--the past is an imperfect guide to the future. Priorities in nutrition need to be constantly challenged not driven solely by researchers. They depend on capacity, political windows as well as evidence. Studies like these can help, but they are only one input into the constant press for progressive change.
1 comment:
Thanks to Lisa and Lawrence for this important analysis on a widely used indicator to describe nutrition problems and evaluate and monitor programmes to address the issues. In the CGIAR, UK and US Government along with the UN, SUN etc. the prevalence of stunting and numbers of stunted children forms the basis for results based management and evaluation metrics. WHO: 40% reduction in the number of children stunted. Feed the Future to reduce prevent stunting among children under age 5 by 20%. Did the authors examine the implications of their analysis on these often ambitious targets around nutrition specific and sometimes sensitive interventions? In terms of attribution?
Post a Comment