|
Field
Intervention (Public)
|
Before
Context: Worldwide, 650 to 670 million people live in extreme poverty. Malawi is one of the poorest countries on Earth and 70% of the population lives in extreme poverty. Ultra-poor graduation programs (UPGs) are the closest development economics research has come to a silver bullet for ultra-poverty alleviation. The UN SDG 1.1 commits the world to "Eradicate extreme poverty by 2030". If SDG 1.1 is to be achieved, the development community needs to better understand how and why UPGs work, how well they scale, and how to make them more cost-effective.
Location: Within Malawi, Mangochi district at the Southern shore of Lake Malawi is among the poorest districts and the district government selected four subdistricts ("Traditional Authorities" or TAs Nankumba, Mponda, Chimwala, and Makanjila) for consideration as the study site for being currently under-served when it comes to poverty alleviation programming. TA Nankumba was selected among the four by the implementing NGO for its accessibility, both for program staff going to the area and for program participants to access urban markets and Lake Malawi for potential business ventures they may engage in.
Program and implementing partner: The implementing partner for this study, Yamba Malawi, is a Malawian and US non-profit founded in 2006 that has transitioned from a purely in-kind donation focus to a child-centered UPG modeled on the approach by the Bangladeshi NGO BRAC, but adapted to their specific ECD focus and the local context. At the household level, the all-female program participants receive monthly cash transfers of the equivalent of $20 for twelve months, business start-up grants of $125 at the end of a six-month weekly training in ECD and micro-entrepreneurship, a basic phone, support for the formation of savings groups, and fortnightly one-on-one coaching and monitoring visits. At the community level, the NGO supports community structures (community-based organizations and community-based childcare centers) financially and with trainings. After several iterations of refining the model, the NGO seeks to evaluate its impact with an openness for further-reaching research questions. The program evaluation that underlies this research study isolates the household component of the program. The control group will receive a basic phone as well, survey incentives (which are modest relative to the cash transfers of the UPG), and fortnightly monitoring visits that will share some of the coaching components of the treatment group. The community component lacks excludability and is extended to all households in the study area, also to secure local buy-in.
Sample: The implementing partner, Yamba Malawi, will collect a household census as well as community wealth rankings to determine program eligibility using the Simple Poverty Score Card. To be eligible for the program, households need to have a mother of a child under the age of 5, be classified as ultra-poor, not state any intentions of migrating in the foreseeable future, and the mother needs to be able to work.
Household clusters: We expect the study area to contain far more eligible households than the NGO has funds to serve. This provides an opportunity of strategically selecting the sample. Note first that in the local context, villages are fairly fluid concepts. Households routinely split off their village to form new villages and the area can be thought of as more or less continuously populated. The study therefore assigns all eligible households to clusters by proximity, regardless of village. It then selects 72 household clusters in a way that bolsters statistical power for research question (1).
Timeline: This is a multi-year program and project. The tentative timeline is to conduct a household census in August of 2023 and launch the first of three cohorts at the end of 2023 with cohorts 2 and 3 scheduled for 2024 and 2025.
|
After
Context: Worldwide, 650 to 670 million people live in extreme poverty. Malawi is one of the poorest countries on Earth and 70% of the population lives in extreme poverty. Ultra-poor graduation programs (UPGs) are the closest development economics research has come to a silver bullet for ultra-poverty alleviation. The UN SDG 1.1 commits the world to "Eradicate extreme poverty by 2030". If SDG 1.1 is to be achieved, the development community needs to better understand how and why UPGs work, how well they scale, and how to make them more cost-effective.
Location: Within Malawi, Mangochi district at the Southern shore of Lake Malawi is among the poorest districts and the district government selected four subdistricts ("Traditional Authorities" or TAs Nankumba, Mponda, Chimwala, and Makanjila) for consideration as the study site for being currently under-served when it comes to poverty alleviation programming. TA Nankumba was selected among the four by the implementing NGO for its accessibility, both for program staff going to the area and for program participants to access urban markets and Lake Malawi for potential business ventures they may engage in.
Program and implementing partner: The implementing partner for this study, Yamba Malawi, is a Malawian and US non-profit founded in 2006 that has transitioned from a purely in-kind donation focus to a child-centered UPG modeled on the approach by the Bangladeshi NGO BRAC, but adapted to their specific ECD focus and the local context. At the household level, the all-female program participants receive monthly cash transfers of the equivalent of $20 for six months, business start-up grants of $300 at the end of a six-month weekly training in ECD and micro-entrepreneurship, a basic phone, support for the formation of savings groups, and fortnightly one-on-one coaching and monitoring visits. At the community level, the NGO supports community structures (community-based organizations and community-based childcare centers) financially and with trainings. After several iterations of refining the model, the NGO seeks to evaluate its impact with an openness for further-reaching research questions. The program evaluation that underlies this research study isolates the household component of the program. The control group will receive a basic phone as well, survey incentives (which are modest relative to the cash transfers of the UPG), and fortnightly monitoring visits that will share some of the coaching components of the treatment group. The community component lacks excludability and is extended to all households in the study area, also to secure local buy-in.
Sample: The implementing partner, Yamba Malawi, will collect a household census as well as community wealth rankings to determine program eligibility using the Simple Poverty Score Card. To be eligible for the program, households need to have a mother of a child under the age of 5, be classified as ultra-poor, not state any intentions of migrating in the foreseeable future, and the mother needs to be able to work.
Household clusters: We expect the study area to contain far more eligible households than the NGO has funds to serve. This provides an opportunity of strategically selecting the sample. Note first that in the local context, villages are fairly fluid concepts. Households routinely split off their village to form new villages and the area can be thought of as more or less continuously populated. The study therefore assigns all eligible households to clusters by proximity, regardless of village. It then selects 72 household clusters in a way that bolsters statistical power for research question (1).
Timeline: This is a multi-year program and project. The tentative timeline is to conduct a household census in August of 2023 and launch the first of three cohorts at the end of 2023 with cohorts 2 and 3 scheduled for 2024 and 2025.
|
|
Field
Randomization Method
|
Before
By a computer
|
After
Treatment will be assigned at random by a computer following the two-level procedure described below. In this process, baseline balance is improved in two ways:
1) Stratification/blocking on geography: As recommended by Bruhn & McKenzie (2009), I stratify study area is split into field facilitator (FF) areas, each containing three clusters by proximity. One of the three clusters each is assigned to one of the three treatment saturations, thus stratifying the randomization by geography (and by proxy implicitly on characteristics such as proximity to markets, occupation in farming or fishing, religion, tribe etc. in as much as these tend to group together).
2) Rerandomization: In order to balance the sample on a number of key outcome and mechanism variables, I first run the below treatment assignment algorithm 100,000 times (seeding each with the counter of its iteration). I then calculate baseline balance under each potential treatment assignment and discard assignments that lead to a statistically significant imbalance in any variable at the 10% significance threshold (Morgan, Rubin, 2012). In other words, I enforce a "visually clean" balance table. I then randomly pick one of the permissible treatment assignments. As pointed out in both Bruhn & McKenzie (2009) and Morgan & Rubin (2012), classical inference that does not take this rerandomization into account would now on average (though not in every single case) be too conservative, leading to confidence intervals on the average treatment effect with overly wide coverage and too few rejections of null hypotheses. A Fisher (1935) exact test has the correct coverage. Traditionally, researchers using rerandomization methods have done so somewhat ad hoc, rerandomizing until the balance table was satisfactory. My approach is more transparent. Given R simulated assignments and k baseline balance table variables, the likelihood of discarding an assignment at the 10% significance level is q = 1 - (1 - 0.1)^k and I can therefore use the remaining R * (1-q) simulated assignments for the randomization inference (Young, 2019).
|
|
Field
Randomization Unit
|
Before
Two levels of randomization: Household clusters are randomly assigned to high saturation (everybody treated), low saturation (half treated randomly at the household level), or pure control (nobody treated). Household clusters are assigned to field facilitators (FFs) where each FF is assigned one of each type of clusters. FF clusters are also assigned randomly to start as part of cohorts 1, 2, or 3. Half of treated household clusters are assigned to receive one extra lesson where they are encouraged to coordinate their business ideas with each other. At the individual level, I also randomize the order in which households are prompted to take certain preventive steps to avoid various shocks.
|
After
Three levels of randomization: Triplets of household clusters (each served by one field facilitator) are randomly assigned to cohorts 1, 2, or 3. Household clusters are randomly assigned to high saturation (everybody treated), low saturation (half treated randomly at the household level), or pure control (nobody treated). Half of treated household clusters are assigned to receive one extra lesson where they are encouraged to coordinate their business ideas with each other while the other treated clusters have a review session in that same time. At the individual level, I also randomize at the household level the order in which households are prompted to take certain preventive steps to avoid various shocks.
|