The interaction between OGI budget and inspection_budget_share was previously
interpreted as 'diminishing returns/resource saturation.' Updated to acknowledge:
- budget_share varies only over time (same as budget), not across districts
- range is just 0.59-0.67 across 8 years (8 data points)
- cannot distinguish moderation mechanism from year-specific confounders
- H2 result is exploratory, not confirmatory
Updated cells: 17 (H2 header), 24 (Results H2 section), 28 (hypotheses table)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Insert Robustness Checks section before Hypotheses Summary
- Wild cluster bootstrap (B=999 Rademacher) addresses G=13 small-cluster problem
- Distributed lag model tests 1-year delayed budget effects (2017-2023 sample)
- Update Data & Methods to describe both robustness checks
- Update Results to reference bootstrap and distributed lag findings
- Fix duplicate robustness paragraph that was appended twice
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Markdown cell summarizing all 9 hypothesis tests with predictions,
key coefficients, and support status including footnotes for H2
diminishing-returns finding, H3 multicollinearity caveat, and H4
level-vs-slope distinction.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- SQL year filters: BETWEEN 2016 AND 2025
- is_budget_year flag: year >= 2024 (covers both 2024 budget estimate and
2025 where no budget data exists) — both excluded from regressions
- Data & Methods and Analysis cells updated to reflect 2016-2025 data range
with 2016-2023 as the regression sample
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Three markdown cells written in PAR/Policy Studies Journal style:
- Data and Methods: sources, sample, variables, model equations
- Analysis: analytical strategy and identification rationale
- Results: full findings with formatted tables for all four hypotheses
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
statsmodels uses z/P>|z| instead of t/P>|t| with cluster-robust SEs
in some versions. Auto-detect from the actual table columns.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
${workspaceFolder} is not expanded in python.defaultInterpreterPath;
use a relative path instead.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
VS Code workspace settings point to project-local .venv so the correct
Python interpreter is auto-selected on open.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
