Programme Participation Intensity and Children’s Nutritional Status: Evidence from a Randomised Control Trial in Mozambique

Abstract

Agricultural interventions are thought to have the potential to improve nutrition, but little rigorous evidence is available about programmes that link the two. In this article, we study impacts of an integrated agricultural and nutritional biofortification project, the REU in Mozambique. We first provide evidence on dietary impacts of the programme and then examine impacts of the programme by participation intensity. Using OLS and IV techniques, we find that more intense participation in both project components led to larger impacts. The results therefore have important implications for refining the design of future projects attempting to link agricultural and nutrition interventions.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Acknowledgements

We want to thank Mary Arimond, Howarth Bouis, Dan Gilligan, Christine Hotz, Ricardo Labarta, Cornelia Loechl, Jan Low, J.V. Meenakshi and Bernardino Munhaua for contributions to the field work that were essential to this manuscript, and four anonymous reviewers whose comments have improved the manuscript.

Notes

1. In a related review, Dangour et al. (2013) search for evidence that agricultural price policies have affected nutritional outcomes, and find a paucity of evidence in general.

2. We do not study the marketing component of the REU here for several reasons. First, the marketing component of the intervention was somewhat focused on what were termed ‘medium-scale producers,’ who were a relatively small number of farmers attempting to produce larger amounts of OSP for market. Second, the remaining portion of the marketing component took a long while to materialise and was not uniformly implemented, even within the farmer groups selected for the intervention. As a consequence, the impact evaluation report found no large impacts on variables associated with the marketing component in either Mozambique or Uganda (de Brauw et al., 2010). We therefore focus on the agricultural and nutritional components here.

3. The REU was actually implemented in both Mozambique and Uganda using two intervention models in each country; the two models varied in intensity to measure variation in cost effectiveness. However, as average impacts did not differ by intervention model, here we treat farmers in all treated villages as the ‘treatment’ group and the remainder as the control group.

4. A primary reason that the REU conducted annual vine distributions is that there is only one primary agricultural season in Zambezia province. The remainder of the year is dry with very little rain.

5. In the more intensive intervention model, this training were largely repeated in each of the three years, whereas in the less intensive model, only the planting training occurred every year in conjunction with the vine distribution, whereas the other training took place in the first year.

6. We also leave out the agricultural promoter category, as none of the promoters showed up in our data set.

7. In the primary report on project impacts, de Brauw et al. (2010) carefully show that there is not much difference between average treatment effects that account for the overweighting of nutrition promoters in the data set and those that do not. In this article, since we are interested in measuring the difference in outcomes based on this intensity, we do not report weighted average treatment effects.

8. By definition, households in the control group did not participate in either component.

9. The intervention took place in 144 organisations in total.

10. Before the fieldwork occurred in all communities, staff informed the leaders of that village about the survey and compiled a list of households that were members of the primary community organisation that would be used as the organisation for the intervention. From that list of households, 25 households with children less than three years old were randomly selected from the list of community group, where five were meant as replacement households; in a few cases, the enumeration staff found that the community lists did not accurately indicate when households actually had children under three years old living in them; some children were slightly older and we drop dietary intake data for those children when it was collected.

11. Randomisation took place at a project meeting in Mozambique by selecting papers with village names on them from an urn.

12. Although the project collected data on dietary intakes on a panel of mothers and a repeated cross section of children aged 6–35 months old at endline, in this article we focus on outcomes among reference children.

13. By setting $\beta =$1 and moving $Y_{is0}$ to the ri. ght hand side, Equation (1)(1) $Y_{is1}={\alpha }_s+\beta Y_{is0}+\gamma P_{is}+{\delta }_1{A}_{is1}+{\delta }_2{A_{is1}}^2+\mu Z_{is0}+{\epsilon }_{is1}$(1) would be a classic difference-in-difference model. In the absence of correlation between $P_{is}$ nd $Y_{is0}$. Equation (1)(1) $Y_{is1}={\alpha }_s+\beta Y_{is0}+\gamma P_{is}+{\delta }_1{A}_{is1}+{\delta }_2{A_{is1}}^2+\mu Z_{is0}+{\epsilon }_{is1}$(1) is more general than a difference-in-difference estimator.

14. We find few significant coefficients among additional explanatory variables. Vitamin A knowledge at baseline is positively correlated with the density of vitamin A in the diet at endline; the age difference between mother and child is positively correlated with mean micronutrient density adequacy, and the distance to the village is negatively correlated with the dietary diversity score.

15. First stage estimates can be found in the Online Appendix, Table A1.