Enabling Sustainable Graduation out of Poverty for the Extreme Poor in Malawi: The Role of Gender Targeting and Couples Training.

The research is designed to address the gap on the role of gender in Graduation program outcomes. The research will estimate the impact of three distinct versions of the Graduation Model on gender empowerment and household welfare outcomes. The three proposed versions tested are as follows: 1) the graduation program targeted at women; 2) the graduation program targeted at men, and; 3) the graduation program targeted at women but with an additional couples’ empowerment training called ‘Transforming Gender & Power Relations’ that is rolled out through 12 monthly group sessions. Each intervention will be delivered to 600 households randomly selected in Mangochi and Nsanje, with 300 households in each district receiving each version of the proposed program. There will be an additional 750 households in each district who will act as control households.

The Graduation intervention in Malawi is a 18-month intense, multifaceted intervention but the research project will be spread over five years to allow for the estimation of medium term effects. The Randomized Control Trial (RCT) will rely on three rounds of a quantitative survey per cohort. A baseline (Q2 2018/19), midline (Q3 2020/21) and an endline (Q3 2021/22) survey will be conducted to capture key welfare measures, as well as important measures of empowerment, IPV and spousal cohesion. The quantitative research will be complemented by qualitative research.

2018-11-01

2021-05-31

- Economic outcomes: income, savings, likelihood of being employed, time use, likelihood of having a business and business revenue, livestock ownership;
- Well-being outcomes: well-being index, health, food security, consumption, education, social standing, future expectations;
- Gender outcomes: intra-household decision making, empowerment, gender norms, relationship quality and intimate partner violence

Through the use of a poverty index, our implementing partner, Concern Worldwide, identified the two districts with the highest poverty score, Mangochi and Nsanje, for implementing the Graduation program in. Within this, they selected sub-districts (VDCs) that they would implement the program in. The pilot intervention will be evaluated using a clustered randomized controlled trial (RCT). The randomisation will be conducted at the village level, where Concern Worldwide will be implementing their Graduation program in.

Treatment villages are randomly allocated to one of three treatment options: 1) the graduation programme targeted at women; 2) the graduation program targeted at men, and; 3) the graduation program targeted at women but with an additional couples’ empowerment training called ‘Transforming Gender & Power Relations’. In treatment village, 12 eligible households will be randomly selected to receive treatment.

In order to be eligible, the households had to meet the following criteria:
• The household had to be classified as poor through a wealth ranking or a proxy means test.
• The household head is married and his spouse lives within the household
• Both head and spouse are willing and able to participate in an Income Generating Activity.
• Both head and spouse reside at the household at least 60% of the year.

* Both village and household randomisation were done on a computer using Stata, a statistical software.

We identified 200 villages, 100 per district, based on each village having at least 20 eligible households in the village. We then randomized across villages, assigning 50 villages as pure control villages and the remaining as treatment villages. After our baseline, we did our next round of randomization, which involved 3 steps. First, we randomized the treatment villages into the three treatment arms. Second, in treatment villages, we randomized 12 households to receive treatment, and the remaining 6 as control households. Third, we randomized all treatment and control villages into cohort 1 and cohort 2, with the first treatment cohort starting with the program in November 2018, and the cohort 2 starting in 2019.

Yes

200 villages

3300 households: 12 households in control villages, 12 treatment households and 6 control households in treatment villages

- 600 control households in 50 control villages
- 600 treatment households plus 300 control households in 50 treatment 1 villages: Graduation targeted to the female spouse
- 600 treatment households plus 300 control households in 50 treatment 2 villages: Graduation targeted to the male spouse
- 600 treatment households plus 300 control households in 50 treatment 3 villages: Graduation targeted to the female spouse with couples training.

In order to identify the sample size required for this evaluation, we performed a number of power calculations. When research costs and implementation costs come from the same budget, we can maximize power (given an overall cost) by allocating a larger sample to the control group than the equal split. To understand the intuition, starting from the equal split, allocating units from treatment to control will result in a overall cost reduction but also loss of power because of the imbalance. However, the latter could be more than offset if we use the part of the cost reduction to boost total sample size. But of course, this will have a limit. As the loss of power increases more than linearly with the imbalance, the loss of power might not be offset if the sample is already highly unbalanced (McConnell and Vera-Hernandez, 2015). This structure allows us to detect a minimum detectable effect size (MDES) of 0.25 standard deviation for continuous variables (which equates to a 10% increase in consumption) when comparing treatments. Consumption data is taken from the recent LSMS Malawi survey and for the poor only. As we are most interested in the effect differential effect of the different treatments, this power calculation is of first order importance. The cluster RCT proposed here allows for the detection of a MDES of 0.16 standard deviations (6% change in consumption) for each individual treatment arm to the control. This comparison tells us what is the impact of an individual treatment compared with the counterfactual of no intervention. For gender empowerment measures, we focus on the typical LSMS question “Decision on use of wife's cash earnings: joint or only wife”, which has a baseline value of 75.7%. The first order comparison between treatment groups, allows us to detect a MDES of 10-percentage points. If the structure of the data allows we will endeavor to use an ANCOVA model rather than difference in difference methodology to help improve power.