Credit the published sources: Newman et al. (1990), Williams & Newman (1982), Dukes et al. (1995), and the 1988 worked example (#105, #106, #109)

[JUhalt]

Fixes #105, #106, and #109. The sources were read in full, and a second reader checked every page, quotation, and number against the PDFs before implementation. An independent reviewer then checked the branch, and all 12 of its findings were fixed or resolved.

#105: the published origin of the unified model

• The model is published. It is Newman, Benz, and Williams's (1990, p. 98) model, which they call a "psuedo-analysis [sic] of covariance". solomonR's coding has the same column space as their equation 7. Their worked example reproduces: slope .55264, within-groups SS 62.39, F = 21.01, and F = 0.21 for the interaction, against the printed .22. test-glm-newman1990.R pins these values.
• Labels changed.
  • The methods guide, ?fit_solomon_glm, the GLM vignette, the worked example, and the README now call it a published Solomon proposal with contemporary inference.
  • They give the restriction as printed (p. 98) and the authors' caution about zero coding.
  • They list solomonR's additions: named contrasts with intervals, HC3 and CR2, Satterthwaite df, GLM families, offsets, marginal_solomon(), covariates, several-treatment designs, and permutation and equivalence tests.
• Reports. report_solomon() cites Newman et al. (1990) alongside Lin (2013) wherever the pretest adjustment is used. The key is not in the core registry, because it is cited only when a pretest is used.
• History. Williams and Newman (1982) fit the whole design in one six-group model, a published answer to Campbell and Stanley's "no single procedure" (p. 25). The history article now has 1982 and 1990 sections, with a timeline that does not overlap. The citing article and the coverage article include both works. Williams and Newman (1982) is not cited by report_solomon(), because its six-group model is not what fit_solomon_glm() fits; the decision is recorded on [v1.0] Credit the published origin of the unified model (Newman, Benz, & Williams, 1990; Williams & Newman, 1982) #105.

#106: Dukes, Ullman, and Stein (1995)

• A precedent, cited. The SEM article, the methods guide, the coverage table, and the references now cite it as the published latent-variable analysis of a Solomon design.
• What is shared: a multiple-group mean structure, loadings tested before latent means are compared, and an equal pretest-to-posttest path.
• What differs:
  • three separate two-group models;
  • no interaction test;
  • loadings tested but intercepts imposed;
  • classroom means.
• What solomonR adds: one four-group model with identified latent means ([v0.3] Identify the latent Solomon mean structure #16), and configural, metric, and scalar tests ([v0.6] SEM and latent-variable article; resolve lavaan warnings #55).
• The "reactive pretest" finding is described as more consistent with a pretest main effect in one indicator than with sensitization, as the verified reading found.
• Planned SEM features taken from it are recorded under [v1.0] No experimental functions at 1.0: implement each one to its literature and meet its validation rule #117; see the comment on [v1.0] Cite Dukes, Ullman, and Stein (1995), the published latent-variable Solomon analysis #106.

#109: the Walton Braver and Braver (1988) worked example

• New data set. waltonbraver1988 holds Table 3 (p. 153), with data-raw code, a help page of known results with a page for every caveat, and COPYRIGHTS and pkgdown entries.
• Tests. Its test reproduces Tests A, D, E, H, and I, the pretest main effect, the 1988 decision path, and the power remark. It also checks the Sawilowsky and Markman counterexample (1988 ERIC version):
  • Tests E and H reproduce, and the printed 1200.04 is a misprint (the error SS is 1148.14);
  • their Test I is reconstructed from their z values.
• Vignette. classic-solomon.Rmd gains a "Published worked examples" section. The pre-registered validation article is unchanged.
• Roadmap. The validation item now says what is reproduced.

Checks

• Full test suite: 2,885 expectations, 0 failures.
• tools/check-references.R: 553 entries match 102 canonical references.
• The spelling check is clean, and all 8 changed articles render.

References

Dukes, R. L., Ullman, J. B., & Stein, J. A. (1995). An evaluation of D.A.R.E. (Drug Abuse Resistance Education), using a Solomon four-group design with latent variables. Evaluation Review, 19(4), 409–435. https://doi.org/10.1177/0193841X9501900404

Walton Braver, M. C., & Braver, S. L. (1988). Statistical treatment of the Solomon four-group design: A meta-analytic approach. Psychological Bulletin, 104(1), 150–154. https://doi.org/10.1037/0033-2909.104.1.150

The full references for Newman et al. (1990), Williams and Newman (1982), and Sawilowsky and Markman (1988) are in vignettes/articles/references.Rmd on this branch.

🤖 Generated with Claude Code