# Extraction Report: Pabich 2018

## Source Crosswalk

- Thesis PDF: `pdf/paper.pdf`
- Text extraction: `out/text.txt`, `out/fulltext.md`
- Raw scenario datasets: `data/*.csv`
- Shared demographic/end files: `data/demo/625672.csv`, `data/end/255249.csv`
- Manual numeric analysis: `data/analiza_2025_bm.Rmd`
- Reproducible computation: `analysis/effect_sizes.qmd`

The thesis is treated as the source of truth for scenario wording, procedure, scale wording,
sample description, and moderator coding. The 2025 reanalysis and raw CSV exports are
treated as the source of truth for the extracted effect sizes.

## Extraction Decisions

- Included primary effects: three scenario-by-evidence-role high-vs-low contrasts where
  the subject of the target knowledge attribution is also the protagonist who has the
  evidence for the target proposition: Bank `Swiadectwa = Tak`, Biedronka `Swiadectwa = Tak`,
  and Orzeszki `Swiadectwa = Tak`.
- Removed the three `Swiadectwa = Nie` effects because the target knowledge subject is
  not the protagonist whose evidence is described, making evidence and evidence reliability
  underdetermined under the target-subject coding rule.
- Orzeszki skeptical pressure is coded to match the Sripada & Stanley pine-nuts cases:
  `skeptical_pressure: No`.
- The extracted effects follow the retained rows from the final table in `analiza_2025_bm.Rmd`.
- Effect computation: grouped low/high means, SDs, and Ns; SMD via
  `esc::esc_mean_sd`.
- Effect design is coded as between-subjects because, within a fixed evidence/self-ascription
  role (`Swiadectwa`), high and low target ratings come from different scenario versions.

## Issues For Review

### HUMAN_CHECK_PABICH_SAMPLE_N

The thesis reports 771 participants and 18 exclusions. The 2025 analysis produces
scenario-level cleaned counts of 263 Bank, 268 Biedronka, and 259 Orzeszki records after
scenario-specific filters. The YAML uses the thesis-level `n_final = 753` and the exact
effect-level group Ns from the 2025 analysis.

Required human answer: confirm that thesis-level `N = 753` should remain at study level.

## Validation Notes

The YAML uses thesis pages for stimuli and question-wording provenance. Dataset-derived
groups use `table_ref` values pointing to the 2025 Rmd and the reproducible QMD.