Four preprints on Zenodo benchmark AnalyVa's engines against R 4.4.0, SmartPLS 4, and the 31 NIST StRD certified datasets. Every dataset, script, and output file used in these studies is public.
by Abdelouahd Bouzar & Khaoula El Idrissi · July 2026
Descriptive statistics, correlations, regression, ANOVA, and factor analysis benchmarked on iris, mtcars, simulated survey data (nine studies total), plus the 31 certified reference datasets published by the U.S. National Institute of Standards and Technology (NIST StRD). Results reproduce R's output to numerical precision.
Three moderation-model specifications benchmarked on the canonical corporate reputation dataset from the Hair et al. PLS-SEM textbook (n = 344): a simple four-construct reflective model, an extended moderation model with a multi-item moderator, and a multiple-interaction model. Absolute differences in path coefficients stay below .010 across all three models; outer loadings match to within .006.
Two benchmarks in one paper. CB-SEM: the Wheaton alienation model (AMOS Example 6, N = 932), a classical identification-and-fit test. PLS-SEM: a ten-construct burnout model (N = 592). AnalyVa reproduces SmartPLS 4 numerics and model-fit indices in both cases.
Five studies in one paper. CFA and configural MGA on the Holzinger–Swineford dataset. Latent growth curve modelling. Gaussian copula endogeneity correction in both OLS and PLS-SEM. Each result cross-checked against SmartPLS 4.
The four preprints above are hosted on Zenodo and are citable immediately via their DOIs. Peer-reviewed journal placement is planned for a later round; candidate venues include SoftwareX, Journal of Statistical Software, JOSS, Behavior Research Methods, and PLOS ONE.
Using these results in your own work? See how to cite AnalyVa.