Elucidating Avenues for Corruption: Micronutrient Fortification Strategies in India’s Mid-day Meals Program

2014-10-27

2015-04-30

1. take-up of fortification schemes.
2. nutrition content and quality of meals.
3. satisfaction with the schools meals among the recipients of this public program (i.e., both children and parents).
4. haemoglobin levels
5. height for age
6. weight for age
7. weight for height
8. mid-upper arm circumference
9. school attendance.
10. test scores.
11. cognitive test scores.

This evaluation will compare methods to fortify school meals in 300 primary schools in Keonjhar District, Orissa, with a micronutrient mix. Students are between the ages of 6 and 13. Schools in this district are given rice and funds for purchasing ingredients for the meals. Headmasters have bank accounts for schools funds and typically have full control of these funds. We propose four treatment arms (see the table below). As described in detail below, the different arms vary in the extent of participation by school officials and other local meals providers (most commonly self-help groups).

Treatment Arms:

T0 Control. Standard midday meal scheme with no additional intervention

T1A Centralized midday meal scheme (unfortified). The centralized kitchen model was designed by Naandi to minimize costs and provide uniform meals. The model is also designed to prevent individual meal providers from pocketing funds.

T1B Centralized fortified midday meal scheme. This model is similar to T1A, but Naandi will fortify the meals with the micronutrient mix.

T2 Standard midday meal scheme with fortification. In this treatment, individuals who traditionally provide school meals (e.g. the headmasters and teachers, and a local cook) will receive training on the nutrition standards used in the Naandi-fortified meals and receive the micronutrient mix with which to fortify the school meals.

T3 High intensity monitoring (cross randomization). Lastly, we will vary the intensity with which we monitor the school meals in the other four treatment arms (high and low intensity) to examine the impact of monitoring on leakages and service delivery.

Done in office.

The unit of randomization is a school.

While all grade 1-5 children in the schools will receive the assigned treatment, we have randomly chosen a sample of children from each school to survey.

Yes

300 schools

4500 pupils in the endline survey (due to financing constraints, we may have to reduce it to approximately 3000 pupils)

75 schools in each of the four treatment arms.

A. Power calculation: Power to Detect Differences in Take-up In these power calculations, we first focus on power to detect differences between a treatment arm and the control schools (where we expect zero take-up). We calculate a minimum detectable effect (MDE) of 2.19 mg/kg for zinc and 25.13 mcg/100g for vitamin A. These calculations assume an intra-cluster correlation of 0.2 and a standard deviation of 6.18 mg/kg for zinc and 70.91 mcg/100g for vitamin A. We estimated these standard deviations by collecting 2 food samples each from 4 schools, fortifying one of each pair with the mix and sending all 8 samples for laboratory testing. This exercise also gave us a sense of the difference in the amount of zinc and vitamin A we should expect between a fortified meal and an unfortified meal: The mean difference in nutrient content between the fortified samples and the non-fortified samples was 61.14 mg/kg for zinc and 219.43 mcg/100g for vitamin A, 95% and 46% of the amount of each nutrient in the mix. However, we have since changed the mix due to pressure from the government; with the new mix, we expect an effect of approximately 37 mg/kg for zinc and 107 mcg/100g for vitamin A, respectively. With 100% take-up, the MDE is still substantially below the expected effect size. In fact, we would be able to detect a significant difference in zinc content even if take-up was as low as 6% and in vitamin A content if take-up was as low as 24%. Based on our previous experience with the schools during the pilot, we expect much higher levels of take-up. We next test whether we have sufficient power to detect differences between multiple treatment arms. Here we assumed a higher standard deviation based on our pilot test results. With this higher variation, we should be able to detect a 4.35 mg/kg difference in zinc content and 50.25 mcg/100g in vitamin A if the difference in take-up was 100%. This large difference in take-up is unlikely, but we should still be able to detect a difference as long as the difference in take-up is higher than 12 percentage points for zinc and 47 percentage points for vitamin A. B. Power to Detect Differences in Haemoglobin We next turn to our ability to detect differences in haemoglobin. With a sample size of approximately 14 students per school surveyed, our MDE (Intention to Treat) is approximately 0.20 g/dL. Even with a take-up rate of 50% (during our pilot interventions we experienced even higher take-up rates), we can detect an effect (Treatment on the Treated) of 0.40 g/dL. Due to financing constraints, we may have to reduce our endline sample to 10.5 surveyed students per school; in this event, our MDE (Intention to Treat) will be approximately 0.22 g/dL. Even with a take-up rate of 50%, we will be able to detect an effect (Treatment on the Treated) of 0.44 g/dL. We compared our MDE to expected effect sizes we drew from the literature, specifically, from two studies that looked at the impact of providing a micronutrient mix that did not contain iron (Fawzi et al. 2007 and Mehta et al. 2011). The mean effect size across the two studies was 0.62 g/dL, higher than our MDE estimates.