State and Local Tax Burden and Cost-Adjusted Standard of Living — PUMA-Year Panel Evidence, 2010–2024 (replication package)

Replication package (data and code) for the paper: "The Effect of State and Local Tax Burden on Cost-Adjusted Standard of Living in U.S. Communities, 2010–2024" Thakur Singh, 2026. This deposit contains the reproducibility artifact for the paper's claims. It enables exact replication of every reported estimate from the constructed panel, without re-assembling the source data from federal statistical APIs. CONTENTS -------- • submission/extended_panel_v2.csv — the analytical panel used for all reported estimates: 35,038 PUMA-year observations, 2010–2024, all 50 states and DC. • submission/master_analytical_panel.csv — earlier panel build, retained for provenance. • submission/DATA_DICTIONARY.md — variable-level dictionary. • analysis/ — nine scripts: panel construction, the three nested specifications, non-parametric and robustness checks, state-clustered inference, the variable-exclusion evidence, and the heterogeneity analysis with wild cluster bootstrap. • submission/journal/ — output files from each analysis, as reported in the paper. • requirements.txt — pinned Python dependencies. • README.md — contents, run order, and data caveats. • MANIFEST.txt — SHA-256 checksum for every file. • CITATION.cff — citation metadata. REPRODUCIBILITY --------------- Scripts resolve their paths relative to their own location; keep analysis/ and submission/ side by side and no editing is required. Run order is in README.md. Bootstrap seed: 5910, with 199 Rademacher replications and the null imposed. Two scripts run bootstraps and take a few minutes; the rest complete in seconds. Estimation uses two-way (PUMA and year) fixed effects with standard errors clustered on state — the level at which the tax measure actually varies. RESULTS REPRODUCED ------------------ • H1 rejected: tax burden coefficient +0.0031 (p = 0.055), positive against the predicted negative sign. H1 is rejected on sign, not on significance. • H2 confirmed: the coefficient attenuates to +0.0015 (p = 0.336) once structural controls enter. • H3 confirmed: on the multi-dimensional welfare composite, +0.0048 (p = 0.813). • Clustering premise verified: the maximum within-state-year standard deviation of tax burden is 0.000000000000 across all 765 state-year cells, which is why state is the correct clustering level rather than PUMA. • Heterogeneity uniform on all four cuts (Census region and baseline income tercile, both outcomes). Wild cluster bootstrap p = 0.503 (Y1 x region) and 0.387 (Y2 x region), both well above the analytic values, showing analytic joint inference to be anti-conservative at 51 clusters. • Variable-exclusion evidence: r = 0.9553 between graduate-degree and bachelor's-plus attainment; adding the former lifts the retained control's within-demeaned VIF from 1.78 to 5.05, breaching the 4.0 threshold applied in the paper. • Variance decomposition: 77.9 percent of the variation in cost-adjusted income, and 93.1 percent for the welfare composite, lies between PUMAs rather than within them. DATA NOTES ---------- • Tax burden is a state-year measure and is invariant across the PUMAs within a state-year. This is a property of the source data and drives the clustering choice. • PUMA identifiers turn over with American Community Survey vintage revisions. Identifier continuity is about 52 percent across the 2011–2012 boundary and 61 percent across 2021–2022, and only 731 of 4,270 PUMAs are observed in all fifteen years. The panel is unbalanced by construction and the estimators treat it as such. • The 2020 observation uses ACS 5-year estimates; every other year uses 1-year estimates. • All inputs are publicly available U.S. federal statistics: Census Bureau American Community Survey; Census Bureau Annual Survey of State and Local Government Finances; Bureau of Economic Analysis; Bureau of Labor Statistics; USDA Economic Research Service. LICENSE ------- Code (analysis/*.py): Apache License 2.0, patent grant included — see LICENSE-CODE.txt. Data and documentation: Creative Commons Attribution 4.0 International (CC BY 4.0) — see LICENSE-DATA.txt. Copyright (C) 2026 The Authors. The underlying source statistics are works of the United States federal government and are in the public domain. The derived panel, the analysis code and the documentation in this deposit are the original contribution. CITATION -------- Cite this deposit using its concept DOI, which always resolves to the latest version: Singh, T. (2026). State and Local Tax Burden and Cost-Adjusted Standard of Living — PUMA-Year Panel Evidence, 2010–2024 (replication package) [Code and data]. Zenodo. https://doi.org/10.5281/zenodo.21612670