Reproducibility materials for a public-use NHANES benchmark of small-sample regression instability and bounded shrinkage recovery

This archive contains reproducibility materials supporting an empirical benchmark study of small-sample regression instability and bounded shrinkage recovery using public-use National Health and Nutrition Examination Survey (NHANES) 2017–March 2020 pre-pandemic data. The materials include analysis scripts, configuration files, aggregate output tables, figure-generation materials, documentation, a codebook, and a file manifest. The benchmark evaluates recovery of full-sample pseudo-reference coefficients from intentionally small subsamples using public-use patient-level NHANES variables related to alanine aminotransferase, dietary supplement use, demographic factors, body mass index, and prescription-medication count. The archive does not redistribute raw NHANES source files, restricted datasets, proprietary datasets, non-public clinical datasets, or individual-level derived analytic datasets. Public-use NHANES source files should be obtained directly from the National Center for Health Statistics. The included outputs are aggregate or reproducibility-supporting materials suitable for public release. The bounded shrinkage implementation included in the archive is provided as a prespecified benchmark candidate for evaluating small-sample estimator behavior. The materials are not intended to support causal claims about dietary supplement use and liver-enzyme abnormality, nor do they establish general superiority over established small-sample methods.