Article by Bansi Malde (PEPA node, Institute for Fiscal Studies). This article appears in the Summer 2013 issue of MethodsNews newsletter (opens a .pdf file).
Studies evaluating health policies and interventions rarely consider how aspects of individual and household behaviour not directly targeted by the policy change so as to comply with it. Determining this lies at the core of understanding why a policy or intervention worked (or didn't) and moreover sheds light on whether a similar policy would be successful in other settings and contexts.
We investigate how households adjust their non-health behaviours so as to comply with behaviours encouraged by an intervention that provided mothers of young infants with information on child nutrition in Mchinji District in rural Malawi1. In particular, a local volunteer visited mothers of infants aged less than 6 months in their homes and encouraged them to exclusively breastfeed their infants until they were 6 months old, and further provided information and suggestions on nutritious foods for complementary feeding as the child got older. The intervention provided information only; no monetary resources or food were provided.
A cluster randomised control trial, the gold-standard evaluation method, was set up to evaluate the impacts of this intervention. We exploit the randomised control trial design to establish how the intervention influenced non-health and health behaviours and outcomes. As is well known, intervention impacts can be established by comparing average outcomes in clusters that received the intervention with those in clusters that didn't receive the intervention (this is also known as an intention-to-treat effect). A concern that often arises in randomised control trials is that the policy could spill-over to those in the control group, thus contaminating them and biasing the estimated intervention impact. This is particularly relevant in our case, since the policy being evaluated is an information intervention that can easily spread from treated to control clusters.
In order to minimise the possibility of such spill-overs, the intervention was assigned to clusters of villages, with buffer areas between adjacent clusters. Specifically, the rural parts of Mchinji District in Central Malawi were divided into 48 clusters, based on population density and geo-political borders. Within each of these clusters, the 3,000 individuals living closest to the geographical centre of the cluster were chosen to take part in the study. The remainder were in the buffer areas. 12 of the clusters were randomised to receive this infant feeding intervention, and 12 remained as controls2.
Though this design reduces the possibility of contamination of the control clusters, it reduces the number of clusters in the study. This influences the choice of method we use to determine the precision of our estimates (see M. Brewer, Methods News, Summer 2012).� When the number of clusters is small, many standard methods for assessing the precision of estimates are too lax, leading to false rejections of the null hypothesis of no effect. To overcome this, we use a bootstrap method3,4, which has been shown to work well in such situations. In order to focus on the most relevant non-health behaviours to consider, we build a simple economic model in which parents choose their own consumption and labour supply and their child's consumption. The model predicts that child and household consumption should increase in response to the intervention, and that adults should work more in order to pay for the additional consumption.
Our findings show that the intervention improved children's diets - in particular, they ate more proteins - and consequently their height (which is sensitive to nutrition) increased. Proteins are relatively expensive in this setting and are not generally grown by households themselves. Fathers increased their labour supply in order to fund the increased consumption: those in treated clusters were more likely to take on a second job and worked more hours relative to those in the control clusters. Moreover, the intervention benefitted the diets of children not directly targeted by the intervention.
These results shed light on why this intervention succeeded and on whether it is likely to succeed in other contexts and settings. Adults had spare capacity to work more, and had available employment opportunities that allowed them to work more and thereby fund better diets for their children.
This is a summary of the paper 'Household Responses to Child Nutrition: Experimental Evidence from Malawi' (IFS WP 12/07), which is joint work with Emla Fitzsimons, Alice Mesnard and Marcos Vera-Hernandez.
References
1 The intervention was set up by Mai Mwana, a research and development project established by researchers from the Institute of Child Health at UCL and Malawian paediatricians.
2 The remaining 24 clusters received a women's group intervention, which focused on improving reproductive health.
3 This is a statistical procedure which constructs a distribution for the test statistic by repeatedly drawing samples from the sample data.
4 Cameron, C., Gelbach, J., and Miller, D. (2008) Bootstrap-Based Improvements for Inference with Clustered Errors. Review of Economics and Statistics, 90: 414-427.