Experimental Evidence on the Impact of Timing of Unconditional Cash Transfers on Investment by Farmers

2026-07-15

2026-07-31

Investment in targeted agricultural activity (per unit area), Investment in other non-targeted agricultural activities (per unit area), investment in non-agricultural activities (per household member)

The main outcome variable is the increase in investment in targeted agricultural inputs per unit area (fertilizers). We focus on fertilizers as the primary measure for three reasons. First, fertilizer is the most widely used and easily quantifiable purchased input among smallholder farmers, making it a clean and comparable expenditure category (Duflo et al., 2011). Second, fertilizer purchase is a discrete, time-bound decision tied to the sowing calendar, making it particularly sensitive to the timing of liquidity inflows — unlike labor or land preparation, which are continuous and less easily attributable to a specific cash inflow. Third, fertilizer investment is the input category most directly targeted by PM-KISAN stated policy objective of enabling timely procurement of inputs (GoI, 2020), making it the most policy-relevant measure. Additionally, we collect data on a broader set of input purchases (eg: seeds, pesticides, hired labor, and equipment rental) during the study period, allowing us to check if the transfer is allocated to a non-targeted period. We also collect detailed household consumption data to capture how cash transfer funds are allocated to other general household needs.
The outcome measures will be collected through structured surveys (Questionnaire is provided in the supplementary material). Input purchases will be measured using itemized expenditure recall covering the full agricultural season, capturing quantities, unit prices, and timing of each purchase. This approach follows the measurement protocols used in comparable agricultural transfer evaluations (Aggarwal et al., 2024; Haushofer & Shapiro, 2016). Household consumption will be measured using monthly recall modules covering food and non-food expenditure categories, following standard practice in consumption measurement for developing-country settings (Beegle et al., 2012; De Weerdt et al., 2016). Data collection will commence immediately after the cash transfer intervention to minimize recall decay, with the endline survey timed to capture the full post-transfer expenditure period.
We acknowledge that self-reported expenditure/investment data are subject to recall decay, telescoping, and social desirability bias. Beegle et al. (2012) show that survey methodology choices can substantially affect consumption estimates, and De Weerdt et al. (2016) find that different approaches applied to the same households yield estimates differing by up to 30%. We mitigate these concerns by using itemized expenditure modules with category-specific recall periods and commencing data collection immediately after the intervention. Nonetheless, our primary outcome relies on self-reports, and treatment effect estimates should be interpreted with this measurement limitation in mind.

This study employs a randomized controlled trial (RCT) design, widely regarded as the most credible approach for establishing causal relationships (Banerjee & Duflo, 2009; Duflo et al., 2007). The experimental design relies on random assignment to generate exogenous variation in the timing of cash transfers, enabling causal identification of the effects of timing on investment allocation to targeted inputs. This approach isolates the causal effect of timing from confounding factors such as income seasonality, local price variation, and unobserved household characteristics.

Not available

Randomization will be conducted using a computer-generated random sequence, and the allocation list will be kept sealed until the point of transfer.

Assignment to treatment and control conditions will be carried out through random assignment at the household level.

No

1

320

160 treatment, 160 control

We conduct statistical power calculations assuming the main outcome is the investment on fertilizers per unit area. With a sample size of 320 households (160 per arm), a significance level of α = 0.05, and variance estimates drawn from comparable studies, the study is powered to detect a minimum effect size of approximately 0.31 standard deviations with 80% power. This effect size is consistent with existing evidence. Duflo et al. (2011) report effect sizes of 0.20–0.30 SD for timing-based fertilizer interventions among Kenyan farmers. Varshney et al. (2020) estimated a 36-percentage-point increase in the adoption of modern cultivars among PM-KISAN beneficiaries in Uttar Pradesh. Inclusion of baseline covariates through an ANCOVA specification (McKenzie, 2012) is expected to reduce residual variance by 20–30%, effectively lowering the MDE.