Direct Aid to Afghan Women

We are evaluating a direct digital payment system to deliver aid directly to vulnerable women in Afghanistan. The fundamental goal is to provide humanitarian assistance directly using a modality that is not easily captured by an authoritarian government hostile to western powers - the Taliban. The situation in Afghanistan is dire; the UNDP estimates that an astonishing 97% of Afghans are at risk of falling beneath the poverty line, fueling hunger, malnutrition, and increased migration borne of desperation. Our team has secured funding to provide transfers of 4,000 AFA every two weeks to around 2,400 Afghan women for two months. Beneficiaries will be randomized into two groups. The first group of about 1,200 participants will receive payments immediately, the remaining participants in the second group will start receiving payments after two months at which point payments to the first group will conclude. The study focuses on impacts on immediate food and humanitarian needs and on whether the payments are captured by the Taliban.

2022-11-06

2023-04-01

Our primary outcomes will be grouped in three broad categories: needs, informal taxation and economic/wellbeing outcomes. In terms of needs, we will focus on participants’ nutrition (e.g. in how many days have they skipped meals) and access to medicine. In terms of aid capture, we will measure whether the aid payments are (partly or completely) captured either by local government officials/community leaders or other members of the household. In terms of economic/wellbeing outcomes, we will focus on how participants assess their economic situation, their optimism and life satisfaction.

In our pre-analysis plan attached here, we detail the specific survey questions that will be used to construct the different primary outcomes and how exactly we will define these.

We will also conduct several descriptive exercises. These include questions related to how participants interacted with the digital payments app and the presence of any hurdles using it, how the aid payments were used, characterizing the population (e.g. whether they are credit constrained and whether they have any experience with mobile mobile or formal banking), and whether spillovers are likely to be a cause of concern in this setting. Moreover, we will compare the characteristics of our sample to a different sample of 5000 digitally-literate women recruited via digital channels that will receive smaller payments (non–experimentally).

We aim to recruit around 2400 vulnerable women across three cities in Afghanistan: Kabul, Mazar and Herat. Each of these women will get four semi-monthly 4000 AFA aid payments, paid directly to their digital wallets. These women will be then randomly assigned to one of two groups. The first group (treatment) will receive the four payments first, while the second group (control) will receive the four payments after the treatment group has finished receiving their aid payments. We will stratify the treatment assignment by neighbourhood and our main outcome variable.
We will have two additional experimental exercises. In one of these, we will randomize the timing of the follow-up calls to participants. This will generate exogenous variation in the amount of days that an individual has to wait to be called since receiving a payment, which will allow us to investigate how reporting varies with the time passed since receiving the aid payments. In the other exercise, we will assess the extent to which experimenter demands play a role in our setting. To do so, we will randomize whether participants receive a short message stating what we are expecting to see in our results (improvements in measures of need) when completing one round of follow-up surveys or not. We will then compare the results between “primed” and “non-primed” participants.

By computer using Stata’s randtreat command (see Carril, Stata Journal, 2017).

We will randomize at the individual level.

No

2409 women.

2409 women.

1208 women in the treatment group and 1201 women in the control group.

For the power calculations, we set a power of 80% and a significance level of 5%. We use Stata’s “power twomeans” command, using the mean and (residual) standard deviation of our main outcome variable (total tea and bread meals in the past seven days) at baseline. The mean and SD of this variable are 13.75 and 2.56, respectively. The residual standard deviation (after controlling for strata fixed effects) is 1.66. With these assumptions, we obtain a Minimum Detectable Effect of 0.1903. For comparison, using data from the third pilot, described above, and two rounds of follow-up surveys (after 1 and 2 payments, respectively), we found a decrease in the outcome variable of 2.48 when estimating the pooled regression in the “Main analysis” section of this PAP. Thus, we should be sufficiently powered to detect effects.

Pre-Analysis Plan