Replication package for "Do State Higher-Education Cuts Pass Through to More Exposed Colleges? Evidence on Appropriation Incidence and Allocation Governance"

This replication package accompanies the manuscript "Do State Higher-Education Cuts Pass Through to More Exposed Colleges? Evidence on Appropriation Incidence and Allocation Governance". The study examines whether statewide higher-education appropriation cuts pass through differentially to public colleges that were more dependent on state appropriations before the fiscal shock. The main specifications compare more- and less-state-dependent public institutions within the same state-year, using institution fixed effects, state-by-year fixed effects, and sector-by-year fixed effects. The estimated coefficients should therefore be interpreted as differential incidence and operating responses across institutions with different baseline state-appropriation exposure, not as the total effect of state fiscal stress on all public institutions in a state. The main finding is that aggregate state higher-education cuts do not mechanically pass through as larger observed losses for more state-dependent public colleges. SHEF statewide appropriations and summed IPEDS public institution-level appropriations are strongly related in levels and moderately related in annual changes: in the analytic public sample, the correlation between log SHEF appropriations and log summed IPEDS state appropriations is 0.898, while the correlation in annual log changes is 0.615. Among 58 state-years with 10 percent SHEF cuts, 38 also show 10 percent summed IPEDS state-appropriation cuts. This validates that the measures are meaningfully related while also documenting nontrivial state-year mismatch. The institution-level incidence tests show little evidence of negative differential pass-through to high-exposure institutions. For a 10 percentage-point higher baseline exposure interacted with a 10 percentage-point aggregate cut, the estimated first-stage response is positive for total institution-level state appropriations, 1.14 percent with a standard error of 0.52, while the state-appropriation share of core revenue changes by -0.22 percentage points with a standard error of 0.21. Operating outcomes are also close to zero: staff FTE changes by 0.37 percent with a standard error of 0.57, 12-month enrollment by approximately 0.00 percent with a standard error of 0.46, and core expenses by -0.18 percent with a standard error of 0.43. The staff estimate remains similar in the state-sector-year fixed-effect specification, 0.32 percent with a standard error of 0.57. The interpretation is not that state fiscal stress is irrelevant. Rather, the results indicate that aggregate state higher-education cuts are not automatically transmitted to more state-dependent campuses in proportion to their baseline dependence. The evidence is consistent with allocation smoothing, institutional buffering, timing and accounting frictions, or state/system-level allocation rules that protect or redistribute resources across campuses. The package also includes threshold-event diagnostics and private-sector diagnostics showing why binary large-cut event designs can overstate a direct public-sector retrenchment interpretation. ## Replication Package The archive is organized as a professional replication package rather than a working repository. It contains executable code, clean analysis panels, raw-data staging instructions, generated manuscript tables, appendix tables, appendix figures, model estimate files, data checks, logs, checksums, and documentation. Main directories: - `code/`: replication scripts and model helper functions. - `data/clean/`: three clean parquet inputs used by the clean-data replication path. - `data/raw_schema/`: documentation for the raw-source staging structure. - `docs/`: data dictionary, source provenance, raw-data download and staging instructions, and table/figure crosswalk. - `outputs/`: regenerated estimates, data checks, tables, figures, and replication logs. The package has two replication paths. ### Path 1: Replication from Clean Data This is the primary reproducibility path. It starts from the bundled clean analysis panels and regenerates the manuscript tables, appendix tables, appendix figures, model-output CSV files, data checks, and smoke-test logs. ```bash python3 -m venv .venv . .venv/bin/activate pip install -r requirements.txt ./run_from_clean_data.sh ``` The clean-data path recalculates all reported tables and figures from the bundled data. It does not use copied ready-made tables as inputs. ### Path 2: Raw Public Data to Clean Panels The raw-data path is for users who want to reconstruct the clean analysis panels from public source data. The archive does not redistribute provider raw downloads. Instead, `docs/raw_data_download_instructions.md` gives detailed instructions for obtaining and staging SHEF and IPEDS source files in the required normalized schema. To inspect the raw-source validator: ```bash python3 -m venv .venv . .venv/bin/activate pip install -r requirements.txt ./run_from_raw_data.sh --help ``` After staging the required SHEF and IPEDS files, users can validate the raw-source schema and rebuild the clean panels: ```bash ./run_from_raw_data.sh --raw-dir data/raw --work-dir work/raw_validation --build-clean ./run_from_clean_data.sh ``` This two-path structure separates replication from clean data, which is fast and self-contained, from source-data reconstruction, which depends on separately downloaded public files and source-provider formats. ## Main Generated Outputs - Table 1: Analytic Samples, Fiscal Measures, and Exposure Support - Table 2: Reconciliation of SHEF and Summed IPEDS Appropriations - Table 3: First-Stage Incidence of Aggregate State Cuts - Table 4: Operating Responses to Aggregate State Cuts - Table 5: Scaled Effects and Meaningful Declines Ruled Out - Table 6: Robustness to Denominators, Aggregation, and COVID/HEERF Period - Table 7: Large-Cut Threshold Diagnostic - Table 8: Private-Sector Diagnostics The package also regenerates appendix diagnostics, including denominator sensitivity, state-sector exposure-bin aggregation, COVID-period robustness, lead/lag diagnostics, exposure-support diagnostics, pre-period balance, staff-data transition checks, exposure-definition robustness, leave-one-state-out estimates, leave-one-region-out estimates, and SHEF-IPEDS reconciliation figures. ## Keywords higher education finance; state appropriations; fiscal stress; IPEDS; SHEF; public colleges; allocation governance; pass-through; replication package; reproducibility ## Licenses Code is released under the MIT License. Documentation and generated tables/figures are released under CC BY 4.0 where allowed by source-data terms. Users remain responsible for complying with SHEF and IPEDS source-data terms when reconstructing panels from provider files.