Analysis Version Log

The empirical pipeline ran 36 analytical models across three phases, organized as 22 separate R projects (T0–T22). All analyses were run on R 4.4.0 (Apple Silicon) with set.seed(20260301).

Phase 1 — Exploratory (T0–T9, 13 models)

Descriptive statistics, exploratory factor analysis, hierarchical regression, latent profile analysis, fuzzy-set QCA, longitudinal comparison, structural topic modelling, multiple correspondence analysis, importance–performance analysis, panel random-effects regression, propensity score matching, latent transition analysis.

Phase 2 — Causal and structural (T10–T21, 12 models)

Staggered difference-in-differences, hierarchical regression, mediation, sequential mediation, size moderation with Johnson–Neyman intervals, 2×2 ANOVA, moderated mediation, TOE path SEM, ordinal logistic regression, four-class latent profile analysis, fsQCA combined with DID, cross-lagged panel model.

Phase 3 — Nonlinear response surface (T22, 11 models)

Bivariate response surface models (combo1–combo5) and multivariate response surface models (A1, A2, B1, B2, C1, C2) under dual TOE and Training-Transfer framings.

Toolchain

Component Version
R 4.4.0 (Apple Silicon)
rsm 0.10.6
RSA 0.10.6
QCA 3.23
boot 1.3-30
lavaan
Seed 20260301

Inference uses HC3 robust standard errors (Breusch–Pagan p<.001 in 8 of 11 response surface models) and 5,000-sample bootstrap confidence intervals.

WarningReproducibility note

All Draft v1 estimates were produced under the toolchain above. The pending revision requires re-running the response surface pipeline against a different dependent variable. The toolchain must be pinned or re-verified before that run, otherwise differences caused by the DV change cannot be separated from differences caused by environment drift.