From bdda475c8c252fdf22ab832fa9d0eac917384fc3 Mon Sep 17 00:00:00 2001 From: wassname <1103714+wassname@users.noreply.github.com> Date: Sun, 5 Jul 2026 12:32:39 +0800 Subject: [PATCH] MFV: add SOURCES.md audit trail (citations, URLs, transforms) for all 8 country rows Consolidates provenance that was only in terse CSV source-tags + commit bodies: per source, full citation + DOI/OSF URL + which table/figure + the exact data transformation (Care-collapse, Brazil affine bias-cal, Australia raw-recompute). Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com> --- .../data/human/mfv_country_factors_SOURCES.md | 95 +++++++++++++++++++ 1 file changed, 95 insertions(+) create mode 100644 src/tinymfv/data/human/mfv_country_factors_SOURCES.md diff --git a/src/tinymfv/data/human/mfv_country_factors_SOURCES.md b/src/tinymfv/data/human/mfv_country_factors_SOURCES.md new file mode 100644 index 0000000..01882a5 --- /dev/null +++ b/src/tinymfv/data/human/mfv_country_factors_SOURCES.md @@ -0,0 +1,95 @@ +# Sources & provenance: `mfv_country_factors.csv` + +Audit trail for the human country rows in the MFV value map. One entry per `source` +tag in the CSV. Each entry records: full citation, URL/DOI, exactly which table/figure +the numbers came from, and any data transformation applied (so a reviewer can reproduce +the row from the primary source). + +Author of these notes: Claude (pair-programming with wassname), not wassname. + +## Instrument & how the map uses these numbers + +All rows are the Clifford et al. (2015) Moral Foundations Vignettes (MFV): 2nd-person +"You see ..." vignettes rated for moral wrongness on a 1-5 scale, coded to six +foundations (care, fairness, liberty, authority, loyalty/ingroup, sanctity/purity). + +The value map (`scripts/plot_steer_showcase.py::read_human_mfv`) reads **only the `mean` +column**, then **z-scores each country across its six foundations** (ipsative). Because +z-scoring is affine-invariant, any per-country scale/offset (translation bias, digitizing +bias, differing scale anchors) cancels: the map shows *relative* foundation emphasis, not +a calibrated cross-country ranking. Absolute means/SDs are still stored honestly for any +non-map reuse. + +Caveat carried by two of the source papers (Jimenez-Leal, Hopp): the MFV shows +measurement non-invariance / differential item functioning across countries, so raw +between-country mean comparisons are a rough reference, not a validated ranking. + +## Sources + +### `JimenezLeal2025_LatAm` -- Argentina, Colombia, Peru, US +- Jimenez Leal, W., Carmona, G., Murray, S., & Amaya, S. (2025). Validation of the Moral + Foundation Vignettes in Latin America. *Collabra: Psychology*, 11(1), 128178. +- DOI: https://doi.org/10.1525/collabra.128178 (open access, CC BY) +- Numbers: per-country foundation means/SD/N from the paper's descriptive tables + (N = 1,650 across 3 Latin-American countries via polling agency, plus a US comparison). +- Transformation: none (means used as tabulated on the native 1-5 scale). + +### `Yamada2025_MFV-J` -- Japan +- Yamada, J., Nakawake, Y., & Suyama, M. (2026). Developing a Japanese version of the + Moral Foundations Vignettes (MFV-J). *The Japanese Journal of Psychology*. + (Tag says 2025 = preprint/advance-pub year; journal assigns 2026.) +- DOI: https://doi.org/10.4992/jjpsy.97.24228 (advance publication PDF, open access) +- Numbers: MFV-J foundation means/SD, N = 564, from the paper's descriptive table. +- Transformation: none. + +### `Hopp2024_DutchMFV` -- Netherlands +- Hopp, F. R., Jargow, B., Kouwen, E., & Bakker, B. N. (2024). The Dutch moral foundations + stimulus database. *Judgment and Decision Making*, 19, e10. +- DOI: https://doi.org/10.1017/jdm.2024.5 (open access, CC BY). OSF: https://osf.io/9gnza/ +- Numbers: foundation means + 95% CIs from the paper's Table 1 (N = 586 Dutch crowdworkers, + 120 translated MFVs). Per-foundation N varies by item allocation (that's why the CSV N + differs per row). +- Transformation: **Care collapsed** from the paper's split physical-care (4.09) + + emotional-care (3.53) into a single care = 3.81 (mean of the two). Other foundations + taken as tabulated. SD/SE/CI from the paper's reported CIs. + +### `Marques2020_BrazilMFV_fig3digitized_affinecal` -- Brazil +- Marques, L. M., et al. (2020). Translation and validation of the Moral Foundations + Vignettes for the Portuguese language in a Brazilian sample. *Judgment and Decision + Making*. +- URL: http://journal.sjdm.org/19/190809a/jdm190809a.html + PDF: http://journal.sjdm.org/19/190809a/jdm190809a.pdf +- Numbers: the paper tabulates no per-foundation means, so means were **digitized from + Figure 3** (Brazil series) with WebPlotDigitizer by wassname. N = 494 (paper). +- Transformations (two, both by Claude): + 1. **Care collapsed** from digitized Care-E + Care-P into one care value. + 2. **Affine bias-correction** of all digitized means to two paper-stated Purity anchors + (Brazil purity 3.45, Clifford US purity 3.85; paper text): `true = 0.9155*digitized + + 0.228`. Reproduces both anchors; Brazil purity lands exactly 3.45. SDs recovered + from the digitized 95% CI whiskers (Fig 3 caption): `sd = halfwidth/1.96 * sqrt(494)`; + each whisker pair's midpoint matches its mean to <=0.005 (validated). SDs scaled by + the affine slope. Provably map-neutral (max |dz| = 0.017, rounding only). +- Digitized source figure staged at `docs/digitize/brazil_fig3_page-08.png`. + +### `Crone2021_AusUndergrad_MFV90raw` -- Australia +- Crone, D. L., Rhee, J. J., & Laham, S. M. (2021). Developing brief versions of the Moral + Foundations Vignettes using a genetic algorithm-based approach. *Behavior Research + Methods*, 53(3), 1179-1187. +- DOI: https://doi.org/10.3758/s13428-020-01489-y . OSF: https://osf.io/cmwpv/ + (component "Data" = https://osf.io/nv4ty/ , file `dat_rep.sav`). +- Sample: 756 Australian undergraduates (complete cases). The paper's other sample is + 580 US MTurk workers (`dat_amt`), NOT used here -- that would duplicate the US row. +- Numbers: computed by Claude from the **raw participant ratings** in `dat_rep.sav`, the + full 90-item Clifford MFV (1-5 wrongness), using the author's own item->foundation map + from `mfv_abbreviation.Rmd` (Care = MFV 1-27 [physical+emotional+other], Fairness 28-39, + Liberty 40-50, Authority 51-64, Loyalty/Ingroup 65-80, Sanctity/Purity 81-90). + Per foundation: mean over participants of each participant's item-mean; SD = between- + participant SD (ddof=1); SE = SD/sqrt(N); CI = mean +/- 1.96*SE. Complete-case exclusion + (all 90 items present) reproduces the paper's N = 756 exactly. +- Transformation: **NOT the genetic-algorithm-abbreviated subset.** The paper's headline + contribution is a brief MFV chosen by GAabbreviate; we deliberately used the *full* + 90-item ratings so Australia is comparable to the other (full-instrument) country rows. + Care spans all 27 care items (no collapse needed; it's the natural mean). +- Reproduce: OSF files fetched via `https://osf.io/download/4psfc/` (dat_rep.sav); + read with `pyreadstat.read_sav`. (The OSF "Download as zip" gave an empty archive -- + fetch files individually.)