Replication data and code for "What a state-exposure design can and cannot show about the lapse of the Affordable Connectivity Program"

Replication archive, version 2. Accompanies the revised manuscript, which withdraws the single-year identification claim of version 1 after the design failed a placebo test on the 2022 and 2023 Household Pulse Survey years, and reports a stacked three-year triple difference as an exploratory estimate instead. The archive is in two layers. The first downloads the 2022, 2023 and 2024 Household Pulse Survey public-use files, the 2023 American Community Survey denominator tables, and the state and regional auxiliary series, and builds the two analysis inputs from them: a three-year state-by-slot panel of 1,020 cells for each of four household-composition groups, and the Phase 4.2 individual-level file behind the household comparisons. The second regenerates every number, table and figure in the paper from those inputs, and a final step checks each reported cell against what the paper prints. Version 2 adds the placebo years and the stacked design, the corrected wild cluster bootstrap (version 1 contained an implementation error that made the null distribution too narrow), the reference-period and window decompositions, the Summer EBT competing-exposure audit, and the time-varying state controls built from monthly state unemployment and regional food-at-home prices. Version 1 accompanied an earlier title and none of these. The raw survey files are not redistributed; the download scripts retrieve them from the Census Bureau, the Universal Service Administrative Company and FRED.