Comparing single versus double lump sum anticipatory cash transfers for drought in Afghanistan

NOTE: Data collection was complete prior to registration, but no data analysis examining treatment effects or descriptive information has been conducted.

Households will receive unconditional cash transfers valued at roughly $122 ahead of a predicted drought. Depending on the treatment assignment, households will either receive all of the cash upfront during the first window of action ("protect harvest") or they will receive half of it upfront ("protect harvest") and half of it at the second window of action ("mitigate lean season impacts"). All cash will be delivered in person to the primary household representative by IRC staff.

2025-09-01

2025-10-31

Food Consumption Score, Reduced Coping Strategies Index, Livelihoods Coping Strategies Index, asset ownership, household income sources and monthly income, monthly household expenditure

a. Asset ownership
i. Durable household assets (continuous – count): total number of assets self-reported by households from a preset list including radios, televisions, large household appliances, etc.
ii. Livestock and other productive assets (continuous – count): total number of livestock reported across categories (e.g., sheep, goat, cattle, chickens, etc.).
b. Food security and coping strategies
i. Food consumption score (FCS) (continuous + ordinal): composite score is calculated based on self-reported household consumption in the past 7 days for various food groups (e.g., meat, vegetables, pulses, etc.) and weighted according to pre-set nutritional values. Ordinal version involves categorizing the weighted scores into three categories: food secure, borderline insecure, and food insecure.
ii. Reduced coping strategies index (rCSI) (continuous + ordinal): composite score is calculated based on a severity weighting of the type of activity taken, in which more severe actions are given a heavier weight. Households self-report this measure for the last 7 days. The ordinal version classifies households into three categories: food secure, borderline insecure, and food insecure.
iii. Livelihoods coping strategies index (LCS) (continuous + ordinal): composite score is calculated based on a severity weighting of the type of activity taken, in which more severe actions are given a heavier weight. Households self-report this measure for the last 30 days. The ordinal version is constructed such that households are assigned a value based on the most severe action they take (ex., if a household takes both a lower-risk action and a medium-risk action, they will receive the classification associated with the medium-risk action).
c. Household income sources and income-generating activities
i. Primary and secondary livelihood activities (binary): self-reported indicators as to what the households’ primary and secondary sources of income are.
ii. Income from agropastoral activities (continuous): self-report 30-day household income aggregated across any household member(s) who were engaged in agropastoral activities such as farming and livestock rearing.
iii. Income from non-agropastoral activities (continuous): self-report 30-day income aggregated across any household member(s) who was engaged in non-agropastoral activities such as small trade or wage employment.
d. Monthly household expenditure (continuous): self-report 30 day expenditure across categories such as food, non-food household items, clothing, communications, transport, education, social gatherings, rent, debt repayment, and savings.

Climate risk reduction experiences and behaviors, anticipatory cash transfer spending, subjective well-being

a. Climate risk reduction experiences and behaviors
i. Actions taken before, during, and after drought (binary): a list of self-reported actions that households may have taken to prepare or respond to their most recent experience of drought.
ii. Previous experiences of drought and their impacts (baseline only)
b. Anticipatory cash transfer spending (endline only): a binary set of indicators in which households self-report spending any of their anticipatory cash transfers on key categories such as purchasing food, purchasing livestock or productive assets, taking pre-emptive actions, saving, etc.
c. Subjective well-being
i. Cantril’s ladder of life satisfaction (continuous) – measured through an 11-point self-report scale
ii. Bandwidth depletion (interval) – measured through a 4-point self-report measure developed by the MAGNET initiative

We selected Badghis province purposively based on long-range forecasts indicating potential drought risk within the next 3-6 months. Within Badghis, we purposively selected a district based on forecasted risk, existing vulnerabilities including livelihood risk to drought, and operational feasibility. Within the selected district, we constructed a list of communities from which we then randomly assigned communities to either the single or double lump sum groups. All households within those communities would receive the same treatment. Balance checks will be conducted to ensure randomization worked, and cluster-robust standard errors will be used to account for community-level effects.

Not available

Randomization was conducted on a computer in R by the Principal Investigator.

Community

Yes

36 communities: 18 single lump sum; 18 double lump sum.

n = 540 observations at the household level, with an even split between the two groups and disproportionate representation by community to ensure equal sample sizes within clusters.

n = 540 observations at the household level, with an even split between the two groups.

We used the WebPower package and its functions for cluster randomized controlled trials with two arms in R for all calculations. To confirm that we would likely be overpowered if we surveyed the estimated 3,300 clients in the program, we calculated the power assuming n = 3,300 across j = 55 clusters (communities), alpha = 0.05, effect size = 0.4 standard deviations (based on recent studies), and intra-class correlation of 0.15, and testing a two-sided hypothesis. We found that our beta would be 0.95 (or 95% powered). We then calculated the minimum sample size we would need in order to be able to detect effect sizes of at least 0.4 standard deviations in our primary outcomes. Assuming that we would have about j = 55 clusters (communities), an intra-class correlation of 0.15, alpha = 0.05, power = 0.80, and testing a two-sided hypothesis, we calculated the minimum sample size per cluster to be 7 households, or a total of 385 households. In the event that we only reach 40 communities, all else the same, we would need to have a minimum of 20 households per cluster, or 800 households to be adequately powered to detect effects. The IRC Afghanistan country program confirmed that a minimum of ten households per community will participate in the program. With this in mind, we calculated the minimum detectable effect for n = 10 per cluster across j = 55 clusters (communities) at 80% power, alpha = 0.05, and testing a two-sided hypothesis. We could reasonably detect an effect size of 0.37 should it exist.