India's Jan Aushadhi–Suvidha program sells sanitary napkins at ₹1 per pad through a national network of government-sponsored generic pharmacies (Jan Aushadhi Kendras). We link administrative records on the universe of Kendras, their locations, opening dates, and owner characteristics, to nearly half a million women in two National Family Health Survey rounds, and estimate the adoption effect with the doubly robust difference-in-differences estimator of Sant'Anna and Zhao (2020), tracing dynamics with Callaway and Sant'Anna (2021).
Subsidized availability raises sanitary napkin use by 9.6 percentage points, a 16% gain over a 61% baseline. Adoption jumps in the year subsidized pads arrive, with no movement beforehand. A falsification test is the design's built-in placebo: Kendra openings in the decade before subsidized pads existed produce no adoption response, so the ₹1 product, not pharmacy presence per se, is the operative channel.
The response is far from uniform. Effects are larger where a Kendra opens amid existing health and retail infrastructure, and at woman-owned outlets, and smallest in the least served areas. That ordering argues against physical access as the binding constraint and is consistent with the price, information, and stigma channels doing the work.
Adoption builds over the first six to eight months, consistent with information diffusion and adoption at successive menstrual cycles rather than an immediate jump.
Kendra entry before any ₹1 pad existed moves adoption by ≈ 0. The jump is specific to the subsidized product.
The effect survives parallel-trends violations up to three-quarters of the largest pre-period deviation (Rambachan–Roth breakdown at M̄ ≈ 0.75).
Dropping all nine states with their own napkin schemes still yields +5.3pp (p < 0.01). No single state drives the result.
Two earlier Shiny applications map the planner's selections across subdistricts. The first shows the base efficiency and equity trade-off. The second accounts for the availability of waste disposal, which shifts the optimal locations.
These applications were built on an earlier version of the estimates and are kept for the maps rather than the numbers. Where they disagree with this page or the paper, the paper is correct. They are hosted on a free tier and may take up to a minute to start.
Because adoption responds most strongly where Kendras co-locate with existing infrastructure, a planner maximizing adoption per rupee would put new outlets where deprivation is smallest. The paper formalizes the tension as a constrained siting problem for the planned expansion. The planner picks K subdistricts to maximize
where nj is women of menstruating age, τ̂j the predicted adoption gain, dj a healthcare-desert indicator, and λ the equity weight. Because the objective is additive, the solution ranks subdistricts by sj(λ) = nj[(1−λ)τ̂j + λdj] and takes the top K. Move λ to trace the frontier.
CSV header: name,women,tau,desert
tau ∈ [0,1] predicted gain · desert ∈ {0,1}
Candidate units shown are illustrative, generated to match the paper's heterogeneity pattern (predicted gains largest where infrastructure is densest, and negatively associated with the desert indicator). Load a CSV to run the planner on real subdistrict estimates. Adoption predictions apply the paper's subgroup effects; they are not causal guarantees for any specific site.