# F4-conf-01 — Treatment effects and uncertainty extracted from two published tables

## Question
Extract, exactly as printed, specified point estimates, standard errors (SEs), 95% confidence-interval (CI) bounds and counts from (i) Armendariz (2026), arXiv:2609.27732v1 (staggered DiD of drug consumption room openings, Tables 6 and 7) and (ii) Li, Pearson-Merkowitz & Yokum (2026), arXiv:2606.14621v1 (RESEA RCT, Table 2). Each value gets a locator. I also derive a 95% CI from a printed SE and an SE from a printed CI, and I flag any requested quantity that the source does not print. Nothing is re-estimated.

## Data and exclusions
The only inputs were the two source extracts and `data/README.md`. Line counts are in `n_input_rows`: 134, 58 and 10 lines. The SHA-256 of each extract matches the README. No rows were excluded (`out/exclusions.csv`). All 25 requested items were assessed (`n_analysed` = 25). Six input issues are recorded in `data_issues`. The most relevant one: the same 5 km Callaway–Sant'Anna ATT of -0.431 has SE 0.210 in Table 6 but SE 0.218 in Appendix Table A19. Following the key definitions, I used Table 6. The other issues are:
- the number of control regions is 608 in some places and 640 in the A19 note;
- the abstract gives a 35% fall in fatal hospitalisations, while the body text gives about 30%;
- the RESEA text gives the weeks-on-UI CI as unsigned magnitudes, while the table gives signed values;
- the text has a "p ¡ 0.001" rendering defect;
- the extract editor re-set RESEA Table 2.

None of these issues changes any extracted value.

## Method
`out/analysis.py` parses the pipe-separated table lines, finds cells by row and column label, and prints each matched source line. It strips stars, `$`, thousands separators and parentheses, and keeps the sign. Values printed without decimals are stored as integers. JSON does not keep trailing zeros, so 0.210 is stored as 0.21 and -0.340 as -0.34. These are numerically identical, and the printed strings appear in the script's output. For the derived quantities:
- **CI from SE:** ATT ± 1.96 × SE (`derivation_notes.dcr_att_5km_ci`).
- **SE from CI:** (upper − lower) / (2 × 1.96) (`derivation_notes.ri_wage_ate_se`).

No randomness is involved (`random_seed` = null).

## Results
**DCR study (log points)**

| Item | Table | Value | SE |
|---|---|---|---|
| ATT, 5 km | 6 | `dcr_att_5km` = -0.431 (\*\*) | `dcr_att_5km_se` = 0.210 |
| ATT, 3 km | 6 | `dcr_att_3km` = -0.283 | `dcr_att_3km_se` = 0.279 |
| Narcotics ATT, 3 km | 7 | `dcr_narcotics_att_3km` = 0.549 (\*\*\*) | `dcr_narcotics_att_3km_se` = 0.187 |
| Fatal drug-related ATT, 1 km | 7 | `dcr_fatal_att_1km` = -1.207 (\*\*\*) | `dcr_fatal_att_1km_se` = 0.137 |
| HIV ATT, 5 km | 7 | `dcr_hiv_att_5km` = -0.340 | `dcr_hiv_att_5km_se` = 0.294 |

- Observations at 5 km: `dcr_obs_5km` = 17250 region-years (Table 6).
- The HIV 5 km ATT has no star, so it is not marked significant at 10% (`dcr_hiv_att_5km_significant_10pct` = false).
- Its exact p-value is **not reported** in the extract, which prints only stars (`dcr_hiv_att_5km_p_value` = null).
- Derived 95% CI for the Table 6 5 km ATT: [`dcr_att_5km_ci_low` = -0.8426, `dcr_att_5km_ci_high` = -0.0194].

**RESEA study, Table 2**
- Wage ATE: `ri_wage_ate` = 1153 USD/yr, with printed 95% CI [`ri_wage_ate_ci_low` = 166, `ri_wage_ate_ci_high` = 2141].
- No SE is printed for the wage ATE (`ri_wage_ate_se_reported` = false). The SE derived from the CI is `ri_wage_ate_se_derived` = 503.83 USD/yr.
- Control-group mean wage: `ri_control_mean_wage` = 33043 USD/yr.
- Reemployment ATE: `ri_reemployment_ate_pp` = 1.47 pp.
- Weeks-on-UI ATE: `ri_weeks_ui_ate` = -1.99 weeks, with CI [`ri_weeks_ui_ate_ci_low` = -2.24, `ri_weeks_ui_ate_ci_high` = -1.75].

## Uncertainty
- **DCR (Tables 6 and 7):** uncertainty is printed as **standard errors** clustered at the MedStat region level, in parentheses. Significance stars mark p<0.01, p<0.05 and p<0.10. No CIs and no exact p-values are printed for the overall ATTs.
- **RESEA (Table 2):** uncertainty is printed as **95% confidence intervals** with two-sided p-values. No standard errors are printed.
- **What the derived quantities assume:** both assume an approximately normal (Wald) estimator with critical value 1.96. The SE derivation also assumes the printed CI is symmetric. It is nearly so: 987 below and 988 above 1153. With N = 23,549, a t critical value would be practically identical.
- **Check on the wage SE:** the implied z = 1153/503.83 ≈ 2.29 gives p ≈ 0.022, matching the printed p (`out/state/checks.md`).
- **Sensitivity of the DCR CI:** the derived CI depends on using the Table 6 SE. With the A19 SE of 0.218 the interval would be wider. That alternative is not reported as a result.

## Limitations and open issues
- I could not produce an exact p-value for the HIV 5 km ATT (null). Computing one would need a distributional assumption, and the task asks for the printed value.
- The two DCR tables disagree on the 5 km ATT SE (0.210 vs 0.218). The extract does not explain why; I used Table 6 as instructed.
- The derived CI and SE are approximations under the conventions stated above, not values from the sources.
- The RESEA Table 2 layout was re-set by the extract editor. I relied on the editor saying the values are verbatim. As a cross-check, Table 3 repeats the wage ATE cell identically.
- I judged the task complete (`feasibility` = "complete"). The null p-value is the expected answer for a quantity the source does not report, not a failure.