# F4-conf-01 — Extraction of treatment effects and uncertainty from two published tables

## Question
Extract, exactly as printed, specified point estimates, standard errors (SEs), 95% confidence-interval (CI) bounds and sample counts from (a) Armendariz (2026), arXiv:2609.27732v1 (DCR openings in Switzerland, staggered DiD, Tables 6–7) and (b) Li, Pearson-Merkowitz & Yokum (2026), arXiv:2606.14621v1 (Rhode Island RESEA RCT, Table 2). Attach a locator to each value, derive one CI from an SE and one SE from a CI, and identify the requested quantity that is not reported.

## Data and exclusions
Only the two extracts in `data/sources/` were used, plus `data/README.md` for provenance (line counts: README 10, arxiv-2609.27732v1.txt 134, arxiv-2606.14621v1.txt 58). Nothing was excluded. Where a value appears in more than one table (DCR 5 km ATT in Table 6 and Appendix Table A19; RESEA ATEs in Tables 2 and 3), the table named in the key definition (Table 6, Table 2) was used.

## Method
`out/analysis.py` parses the pipe-delimited tables. For regression tables it reads the line below each coefficient row as the SE line, then removes stars, `$` and thousands separators. It prints every matched source line so that each value can be checked by eye. It also runs consistency checks: stars against a normal approximation of coef/SE, the p-value implied by the derived SE, and observation counts against region counts. Derived quantities use the normal critical value 1.96.

## Results
| key | value | locator (abridged) |
|---|---|---|
| dcr_att_5km / se | -0.431** / 0.210 | 2609.27732v1, Table 6, "ATT", "5 km" |
| dcr_att_3km / se | -0.283 / 0.279 | Table 6, "ATT", "3 km" |
| dcr_obs_5km | 17250 (printed "17,250") | Table 6, "Observations", "5 km" |
| dcr_narcotics_att_3km / se | 0.549*** / 0.187 | Table 7, "Narcotics", "3 km" |
| dcr_fatal_att_1km / se | -1.207*** / 0.137 | Table 7, "Drug-related (fatal)", "1 km" |
| dcr_hiv_att_5km / se | -0.340 / 0.294 (no star) | Table 7, "HIV", "5 km" |
| dcr_hiv_att_5km_p_value | **null — not reported** | only stars are printed, no exact p-values |
| dcr_hiv_att_5km_significant_10pct | false | no star on the cell |
| dcr_att_5km CI (derived) | [-0.8426, -0.0194] | Table 6 inputs, derived |
| ri_wage_ate [95% CI] | 1153 [166, 2141] USD/yr | 2606.14621v1, Table 2, "Wage", "Average treatment effect estimate" |
| ri_wage_ate_se_reported | false | Table 2 prints p and CI, not an SE |
| ri_wage_ate_se_derived | 503.83 USD/yr | Table 2 CI, derived |
| ri_control_mean_wage | 33043 USD/yr | Table 2, "Wage", "Control" |
| ri_reemployment_ate_pp | 1.47 pp | Table 2, "Reemployment", ATE |
| ri_weeks_ui_ate [95% CI] | -1.99 [-2.24, -1.75] weeks | Table 2, "Weeks on UI", ATE |

Note: JSON cannot store trailing zeros. In `results.json`, 0.210 is stored as `0.21` and -0.340 as `-0.34`. The printed strings are 0.210 and -0.340, and the numeric values are the same.

The full locator strings are in `results.json["locators"]`.

## Uncertainty
- **DCR study (arXiv:2609.27732v1):** every uncertainty measure is a **standard error** clustered at the MedStat region level, printed in parentheses below each coefficient, with significance stars. The source prints no CIs or exact p-values for the table cells, so the HIV 5 km p-value is reported as null. The **derived 95% CI** for the Table 6 5 km ATT is −0.431 ± 1.96 × 0.210 = [−0.8426, −0.0194] log points. It assumes the estimator is asymptotically normal and that the printed (rounded) SE is the right SE. With critical value 2.0 it would be [−0.851, −0.011].
- **RESEA study (arXiv:2606.14621v1):** the uncertainties in Table 2 are **95% confidence intervals** plus two-sided p-values; no SEs are printed. The **derived SE** for the wage ATE is (2141 − 166)/(2 × 1.96) = 503.83 USD/yr. This assumes the CI is a symmetric normal-based interval. With N = 23,549, a t critical value would make no practical difference. As a check, 1153/503.83 gives z = 2.29 and a two-sided p of 0.022, which matches the printed p = 0.022. The construction of the CI is described only in the paper's main text, which is not in the extract.
- All printed stars in DCR Tables 6, 7 and A19 match a normal approximation of coef/SE.

## Limitations and open issues
- **There is an internal inconsistency in the DCR source.** The 5 km CS-DR ATT (−0.431**) has SE 0.210 in Table 6 but SE 0.218 in Appendix Table A19. The Table 6 value is used, as the key specifies. With 0.218, the derived CI would be [−0.8583, −0.0037].
- **The control-region counts in the DCR source disagree.** Tables 6, 7 and A19 state 608 never-treated control regions, but the A19 note says 640. The printed observation counts are 16,175, 16,875 and 17,250. Each equals 25 × (640 + treated regions), which fits 640 controls rather than 608. This check assumes a 25-year panel (1998–2022), which is my inference. The value 17250 is reported as printed, without correction.
- **Some DCR text does not match the tables.** The abstract says fatal hospitalisations fall "by 35%". The Table 7 5 km fatal ATT of −0.354 log points corresponds to about −29.8%, and Section 6.1 itself says about 30%. The text also calls the HIV estimates "close to zero", while the 5 km cell is −0.340 (not significant). None of this affects the extracted keys.
- **Two minor points in the RESEA source:**
  - The results paragraph prints the Weeks-on-UI CI as magnitudes, [1.75, 2.24]. Table 2 prints the signed interval [−2.24, −1.75], which is the one used.
  - The wage CI is very slightly asymmetric about the estimate (987 below, 988 above), which is consistent with rounding.
- Overall the task is complete. The one quantity that is not available (the HIV 5 km exact p-value) is reported as null because the source does not print it, as the task anticipates.