# Extraction report: feltzzarpentine2010knowmattersless

## Summary
- Paper key: `feltzzarpentine2010knowmattersless`
- Pipeline outputs used (preferred sources):
  - Extraction index/status: `papers/feltzzarpentine2010knowmattersless/out/bundle.json`
  - Main text (structure/context): `papers/feltzzarpentine2010knowmattersless/out/fulltext.md`
  - Extracted table(s): `papers/feltzzarpentine2010knowmattersless/out/tables/tabula_stream_p6_t1.csv` (Stanley’s predictions; not used for effect sizes)
- Output YAML: `papers/feltzzarpentine2010knowmattersless/feltzzarpentine2010knowmattersless.yaml`
- Effect sizes computed in: `papers/feltzzarpentine2010knowmattersless/analysis/effect_sizes.qmd` (rendered to `papers/feltzzarpentine2010knowmattersless/analysis/effect_sizes.html`)
- Added missing low–high pairings:
  - Study 1: `s1_e3` (Low Attributer–High Subject Stakes vs Low Stakes)
  - Study 2: `s2_e2` (Minimal High Stakes vs Attributer)

## Manual interventions / non-pipeline sources (required disclosure)
- The pipeline text outputs did not include complete endnote content (esp. Notes [14]–[16]) and the Appendix stimulus wording with full prompts/scales.
- I therefore consulted the ingested PDF (`papers/feltzzarpentine2010knowmattersless/pdf/paper.pdf`) via `pdftotext` to extract:
  - Appendix II–V stimulus wording + exact question prompts/scale anchors (used for `measures` + moderator coding provenance).
  - Notes [14] and [16] (used for Experiments 3–4 group means/SDs/t-tests).

## Derived quantities / inference notes
- Split group Ns are not always explicitly reported; where needed they were inferred from reported independent-samples t-test dfs and total N statements:
  - Study 1 (Experiment 1): solved group Ns across the four conditions using df values in Notes [6]–[8] plus total N=152.
  - Study 2 (Experiment 2): solved group Ns across the three conditions using df values in Notes [10]–[11] plus total N=119.
  - Study 4 (Experiment 4): derived n_high=n_low=70 from total N=140 and cross-comparison dfs in Notes [16].
- The response scale is reversed relative to most other extractions (1 = Strongly Agree, 7 = Strongly Disagree). For comparability, effect sizes were computed after reverse-scoring the reported means (8 − M), which flips the sign of `d` while leaving `v` unchanged. The YAML retains the reported means/SDs but stores the reverse-scored direction in `effect_size.d`.
- Study 3 (Experiment 3) has a minor consistency issue: the experiment text reports 83 participants, but Notes [14] report t(80) for the key contrast (suggesting N≈82 if df=n1+n2−2). I inferred n_high=42 and n_low=40 from the df pattern across Notes [14] and [16] and flagged `s3_e1` as `needs_review: true`.

## Notes / potential limitations
- Study 1’s High vs Low stakes contrast is confounded with skeptical pressure and utterance polarity (High Stakes includes “banks do change their hours” and uses an “I don’t know...” utterance that the authors reverse-scored; see Notes [4]). This is documented in `s1_e1.contrast.other_notes` and moderator coding.
- Study 1’s `s1_e3` contrast is additionally confounded with attribution type (third-person attribution) and includes the skeptical-pressure cue; `s2_e2` is also confounded with attribution type (third-person) vs stakes. Both are documented at the effect level in YAML.

## Validation / rendering
- YAML schema validation: `papers/feltzzarpentine2010knowmattersless/feltzzarpentine2010knowmattersless.yaml` validates against `docs/stakes_meta_schema.json` (via `jsonschema`).
- Effect-size computations rendered: `quarto render papers/feltzzarpentine2010knowmattersless/analysis/effect_sizes.qmd`.
- HTML extraction report: re-generate `papers/feltzzarpentine2010knowmattersless/feltzzarpentine2010knowmattersless.html` via `invoke render-extraction feltzzarpentine2010knowmattersless` (or the equivalent `meta-extract render-extraction`).