vera-data-nominal-generating

name	vera-data-nominal-generating
description	Server-side extension that completes the full analysis pipeline for nominal (unordered multi-class) outcome variables after vera-data-nominal-reviewing has run. Adds remaining association tests (Chi-square/Fisher's for categorical, ANOVA/Kruskal-Wallis for continuous predictors, pairwise class comparisons), subgroup analysis with interaction tests, full modeling (multinomial logistic regression with RRR, LDA, CART, Random Forest, LightGBM multi-class), confusion matrices, unified variable importance (0-100), and cross-method insight synthesis. Generates manuscript-ready methods.md and results.md with formatted tables, publication-quality figures, and references.bib. Applies output variation and code style variation for natural, non-repetitive output. Triggered after vera-data-nominal-reviewing completes and its PART 0–2 artifacts are present (see ../../../../CROSS-SKILL- INTERFACE.md).
allowed-tools	Read, Bash, Write, Edit, Grep, Glob

Nominal Outcome — Full Analysis & Manuscript Generation

Table of Contents

• Workflow
• Additional Inputs
• Output Structure
• Key References (read before generation)
• Reporting Standards
• Multinomial-Logit Assumption: IIA
• Cross-Skill Interface

Open-source skill. Read reference/specs/output-variation-protocol.md before every generation — apply all variation layers.

Workflow

Continues from where vera-data-nominal-reviewing stopped (PART 0-2 done).

Step	Responsibility	Executor	Document	Input	Output
Additional tests	Run Additional Tests	Main Agent	workflow/step04-run-additional-tests.md	Prior step output	PART 3 code + prose
Subgroup	Analyze Subgroups	Main Agent	workflow/step05-analyze-subgroups.md	Prior step output	PART 4 code + prose
Modeling	Fit Models	Main Agent	workflow/step06-fit-models.md	Prior step output	PART 5 code + prose
Comparison	Compare Models	Main Agent	workflow/step07-compare-models.md	Prior step output	PART 6 code + prose
Manuscript	Generate Manuscript	Main Agent	workflow/step08-generate-manuscript.md	Prior step output	methods.md + results.md

Additional Inputs

Collect if not already provided:

• Target discipline (for reporting conventions)
• Target journal or style (APA 7th, STROBE, etc.)
• Research question / hypothesis
• Subgroup variable (if subgroup analysis desired)
• Reference category for multinomial logistic regression

Output Structure

output/
├── methods.md
├── results.md
├── tables/             ← Markdown + CSV per table
├── figures/            ← PNGs, 300 DPI
├── references.bib
├── code.R              ← Style-varied
└── code.py             ← Style-varied

Key References (read before generation)

File	Purpose
reference/specs/output-variation-protocol.md	Output quality variation layers
reference/specs/code-style-variation.md	Seven-dimension code style diversity
reference/patterns/sentence-bank.md	4-6 phrasings per result type
reference/rules/reporting-standards.md	Hard rules for statistical reporting

Reporting Standards

Same as vera-data-nominal-reviewing, plus:

• RRR: "RRR = X.XX, 95% CI [X.XX, X.XX]" — relative to stated reference category
• Always state which category is the reference for multinomial logistic
• Confusion matrix: report overall accuracy + per-class precision/recall (in-sample caveat if N small)
• LDA: report discriminant function loadings, Wilks' lambda, canonical correlations
• Chi-square: report observed vs expected, Cramer's V with interpretation
• Coefficients: class-specific log-odds with SE; convert to RRR for interpretation
• Tree-based with small N: frame as "exploratory"; never claim predictive validity

Multinomial-Logit Assumption: IIA

Multinomial logistic regression assumes Independence of Irrelevant Alternatives (IIA) — the relative odds of choosing category A vs B should not change when category C is added to or removed from the choice set. IIA is often violated when alternatives are close substitutes (e.g., "bus" vs "metro" in a transport-mode study).

• Test IIA with the Hausman–McFadden test (mlogit::hmftest in R; manual implementation in Python). Report χ², df, p.
• If IIA is violated (p < .05) or theoretically implausible, report the violation and consider: nested logit (ordered substitutability), multinomial probit (no IIA), or a generalized estimating equations approach. Do NOT silently proceed.
• Document explicitly in Methods: "We assessed IIA using the Hausman–McFadden test. [Result.]"
• The tree-based / LDA analyses in Path A do NOT make the IIA assumption and serve as a robustness check when IIA is suspect.

Cross-Skill Interface

Method Unit Contract:
├── code_r           → .R script (style-varied)
├── code_python      → .py script (style-varied)
├── methods_md       → methods.md (varied structure)
├── results_md       → results.md (varied phrasing)
├── tables/          → Markdown + CSV
├── figures/         → PNGs 300 DPI (varied layout)
├── references_bib   → .bib with cited references
└── comparison       → cross-method narrative (in results.md)

Invoked directly after vera-data-nominal-reviewing or orchestrated by vera-data-application-pipelining.