Adolescents Girls, Agency and Appetite: Evidence from a Food Voucher Experiment in High Schools in Indonesia

Classes of female adolescents aged 12-16 are randomized into three types of vouchers. Participants will receive the same voucher twice a week over a three week period.

2024-08-12

2024-08-29

Healthy meal choice

This binary outcome variable is based on each student's choice in a controlled free meal choice between a healthy vs. a relatively unhealthy option.

-Healthy foods or drinks consumed per day
-Unhealthy foods or drinks consumed per day

Consumption frequencies are collected via students' food diaries (featuring a list of typical meal/drink items that are classified into healthy vs. unhealthy by our nutrition co-PI). The diary is collected 6 times during the intervention period. We will use the number of healthy as well as the number of unhealthy meals/drinks consumed per day as dependent variables.

Classes are randomly allocated to the three experimental voucher conditions. We will collect baseline survey and controlled meal choices, as well as two rounds of follow-up controlled meal choices. We will also collect body height and weight at baseline, as well as weights at follow ups. We will also collect food diaries during the intervention period.

Not available

Randomization is done in office using Stata 18.5 based on the baseline data we collected. We initially registered an individual randomization stratified by class. After collecting the baseline data, we observed a low ICC of 0.01046 at class level, which allowed us for a cluster randomization (MDEs of individual and clustered randomization are qualitatively similar). We thus updated our power calculations below and randomized 70 classes (instead of individuals) into control, treatment 1 and treatment 2 stratified by school. This change in randomization approach is meant to better guard against spillovers and make administration of vouchers easier. After randomizing classes into three groups, we checked balance on the meal choice (primary outcome), as well as age, BMI for age, weight, height and class size. We detected no significant differences at conventional levels.

We randomize classes stratified by school.

Yes

70 classes located in 5 schools as included in baseline data.

1011 students as included in baseline data.

The targeted 1011 baseline participants in 70 classes in 5 schools are allocated to the three groups in a roughly balanced way (stratified by schools): 328 students (23 classes) in control group, 337 students (23 classes) in treatment 1 (restricted voucher) and 346 students (24 classes) in treatment 2 (voucher with choice).

Using our baseline data we performed an updated power calculation for a clustered randomization (instead of an individual one as planned). Taking the baseline healthy choice of 0.2275, an average of 14 kids per class, 23 classes in control and treatment each, we use the following command in STATA: power twoproportions 0.2275, m1(14) m2(14) rho(0.01052) power(0.8) k1(23) k2(23). We are powered for a MDE of 0.1055. Inclusion of covariates including baseline meal choice may further increase power.