Impact of behavioral change communication on nutrition and women’s empowerment within the context of irrigation catchment areas in Myanmar: Evidence from a cluster randomized trial

Behavioral change communication on nutrition and gender equality consisting of bimonthly engagements with target households (alternating group-level meetings and household-level and individual coaching). The intervention will be implemented by selected and trained community facilitators by the Myanmar Institute of Gender Studies (MIGS).

2020-03-02

2021-04-30

1. Dietary diversity score for primary female and primary male within household (0-10). This is based on 10 food groups that are essential for nutrition: (i) grains and roots; (ii) beans/legumes; (iii) nuts and seeds; (iv) dairy; (v) meat and fish; (vi) egg; (vii) dark leafy vegetable; (viii) Vitamin-A-rich fruits and vegetables; (ix) other vegetables; (x) other fruits. This is measured by the count of food groups that respondents reported they consume in the last 24-hours before the survey.
2. Empowerment score of women (0-1), measured using WEAI methodology (see https://www.ifpri.org/project/weai). This is measured by the weighted adequacy in agency by women in the following domains: (i) input in productive decisions; (ii) control over use of income and outputs; (iii) autonomy in income; (iv) work balance; (v) access to and decisions on credit; (vi) Ownership of land and other assets; (vii) group membership; (viii) membership in influential groups; (ix) respect among household members; and (x) attitudes about domestic violence.
3. Women being empowered, which is a binary variable if women’s empowerment score is at least 0.75 cutoff.
4. Intrahousehold inequality score (-1 to 1), which is the male’s empowerment score minus female’s empowerment score within the household.
5. Gender parity, which is the binary variable (0/1) if men’s and women’s empowerment within the household are the same.

1. Dietary diversity score for primary female and primary male within household (0-10). This is based on 10 food groups that are essential for nutrition: (i) grains and roots; (ii) beans/legumes; (iii) nuts and seeds; (iv) dairy; (v) meat and fish; (vi) egg; (vii) dark leafy vegetable; (viii) Vitamin-A-rich fruits and vegetables; (ix) other vegetables; (x) other fruits. This is measured by the count of food groups that respondents reported they consume in the last 24-hours before the survey.
2. Empowerment score of women (0-1), measured using WEAI methodology (see https://www.ifpri.org/project/weai). This is measured by the weighted adequacy in agency by women in the following domains: (i) input in productive decisions; (ii) control over use of income and outputs; (iii) autonomy in income; (iv) work balance; (v) access to and decisions on credit; (vi) Ownership of land and other assets; (vii) group membership; (viii) membership in influential groups; (ix) respect among household members; and (x) attitudes about domestic violence.
3. Women being empowered, which is a binary variable if women’s empowerment score is at least 0.75 cutoff.
4. Intrahousehold inequality score (-1 to 1), which is the male’s empowerment score minus female’s empowerment score within the household.
5. Gender parity, which is the binary variable (0/1) if men’s and women’s empowerment within the household are the same.

1. Nutrition knowledge scores
2. Adoption of health and nutrition practices (dummy, count, score)
3. Protein deficiency
4. Micronutrient deficiency (iron, thiamin, vitamin A, calcium) (mean deficiency and percent deficiency)

Model and Hypothesis:
Given random assignment to the treatment, intention-to-treat effects are estimated by ordinary least squares, where the variable of interest is the indicator variable equal to one if the village was assigned to the treatment group (gender and nutrition BCC). The outcome can then be written as:
Y = a + b*T + e
Where Y is the outcome indicator (see above), T is the assignment to the Treatment (gender and nutrition BCC), and b measures the average effect of the Treatment (gender and nutrition BCC).
We test the null hypothesis (b=0). If rejected, we conclude that the intervention has significant effect to the magnitude of b.
We take advantage of the randomized experimental design and conduct an intent-to-treat (ITT) analysis using single-difference estimation with post-endline data. The randomized assignment and balance in baseline characteristics minimize concerns of bias in the single-difference treatment estimates.

Randomization method:
To minimize the risk of contamination, especially that the intervention is on training and communication, we use cluster randomized control trial, where the villages are the clusters. We have 30 villages sampled that are close to the pilot DY2 and DY5/6 in the pilot irrigation sites. The villages are pretty homogeneous in terms of livelihoods, proximity and involvement in irrigation activities, socioeconomic status, traditions, culture and practices related to farming, nutrition and gender. We anticipate minimal difference across villages. We anticipate strong information spillover within the village, and minimal spillover across villages, justifying clustering and randomization at the village level. From the list of 30 villages, we used a computer program (Stata) to select the villages and randomly assign 15 villages in treatment group and 15 villages in the control group. As the unit of randomization is the village, standard errors from the regression models will be clustered at the village level.

Balance at baseline:
We checked the balance or equality of means between the treatment and control using various individual, household and village-level covariates using Wald test. We used age, marital status and education level of respondents; measure of land area owned, housing status, electricity, wage employment, remittances, and household size at the household level.
Moreover, while we tried to minimize the risk on contamination and information spillover between treatment and control villages, it is still possible to occur. We will control for the distance between the household to the nearest border of the nearest treatment village to capture spillover. In the follow-up survey, we will also ask for detailed interactions of the respondents to give some sense of information sharing and spillover especially between treatment and control village that are near to each other. If any, the effect (b) can be interpreted as the lower bound of the true effect.
Lastly, while we do not foresee other projects or activities that would affect the treatment and control villages differently within the intervention period, it may be possible. At the mid-line and end-line surveys, we will ask for any projects or activities implemented in the villages, and we will control for them in the regression analysis.

Randomization done in the office by a computer

village

Yes

30

900 households

15 villages and 450 households assigned to treatment

The list of Treatment villages are: North Yamar irrigation site (Pale and Yinmarbin townships): Village tract Village Kyay Nin Nar Daw Mauk Kyae Tha Hmyar Kyay Tha Myar East Kyae Tha Hmyar Khin Ma = min ma anauk Poke Par Poke Par Mon Kyaing Mon Kyaing Kan Gyi Ale Thi Tinn Min Kan Gyi Min Kan Gyi Min Kan Gyi Min Kan Tike Lel Ngauk Lel Ngauk Sinthe (Tatkone township): Village tract Village Kyaung Kaing Aung Thar Kyaung Kaing Yway Su Oke Shit Kone Oke Shit Kone Kyay Chaung HTa Yan Kar Ma Gyi Pin Pyaw Ma Gyi Pin Yae Twin Phyu Minimum sample size: We used clustersampsi in Stata to compute for the minimum number of clusters needed and the number of sample per cluster. For dietary diversity score, the mean is 4.70, and standard deviation is 1.32 among surveyed respondents in the baseline survey. Using complete randomization, minimum sample size per treatment arm would have been 111 with 80% power and 5% level of significance. Given the randomization done at village level, the sample size has to be inflated for intraclass correlation (ICC), which is 1.92. With this design effect inflator, the sample size of 270 respondents in 9 villages in each treatment arm can detect a minimum of 0.5 increase in the dietary score. Some nutrition BCC projects were able to achieve this impact, and discussions with nutrition experts in Myanmar indicate that this target is feasible. For women’ empowerment score, the mean is 0.61 and the standard deviation is 0.13 among surveyed female respondents in the baseline survey. Using complete randomization, minimum sample size per treatment arm would have been 75 with 80% power and 5% level of significance. Given the randomization done at village level, the sample size has to be inflated for intraclass correlation (ICC), which is 3.06. With this design effect inflator, the sample size of 270 respondents in 9 villages in each treatment arm can detect a minimum of 10% increase in the empowerment score to 0.67. Preliminary analysis of the ANGeL project in Bangladesh show this impact is achieved in the context of that project (Quisumbing et al 2019). This is feasible in our study site, with 37% of the villages achieving this at baseline. For women being empowered (dummy), the mean is 0.21 and the standard deviation is 0.41 among surveyed households in the baseline survey. Using complete randomization, minimum sample size per treatment arm would have been 207 with 80% power and 5% level of significance. Given the randomization done at village level, the sample size has to be inflated for intraclass correlation (ICC), which is 1.89. With this design effect inflator, the sample size of 420 respondents in 14 villages in each treatment arm can detect a minimum of 27% increase in the proportion of households achieving parity to 0.43. Preliminary analysis of the ANGeL project in Bangladesh show this impact is achieved in the context of that project (Quisumbing et al 2019). This is feasible in our study site, with 43% of the villages achieving this level at baseline. For intrahousehold inequality score, the mean is 0.12 and the standard deviation is 0.17 among surveyed households in the baseline survey. Using complete randomization, minimum sample size per treatment arm would have been 127 with 80% power and 5% level of significance. Given the randomization done at village level, the sample size has to be inflated for intraclass correlation (ICC), which is 1.89. With this design effect inflator, the sample size of 270 respondents in 9 villages in each treatment arm can detect a minimum of 50% decrease in the intrahousehold inequality score to 0.06. This is feasible in our study site, with 10% of the villages achieving this level at baseline. For gender parity, the mean is 0.33 and the standard deviation is 0.48 among surveyed households in the baseline survey. Using complete randomization, minimum sample size per treatment arm would have been 288 with 80% power and 5% level of significance. Given the randomization done at village level, the sample size has to be inflated for intraclass correlation (ICC), which is 1.40. With this design effect inflator, the sample size of 450 respondents in 15 villages in each treatment arm can detect a minimum of 33% increase in the proportion of households achieving parity to 0.45. This level is achieved in 20% of the villages in the study site at baseline. The baseline survey covers 30 villages, 29-32 households randomly selected per village.

Pre-Analysis Plan