2026-10-07T15:04:27Z https://api.figshare.com/v2/oai

oai:figshare.com:article/34040517 2026-10-01T06:32:11Z category_29869 category_29173 category_29821 category_28963 item_type_9 month_year_10_2026

Mean-preserving online shrinkage for prediction-assisted finite-population auditing: replication package julian hoxha (24494312) Applied statistics Machine learning not elsewhere classified Numerical and computational mathematics not elsewhere classified Data management and data science not elsewhere classified mean-preserving online shrinkage prediction-assisted inference prediction-powered inference finite-population inference confidence sequences sequential inference sequential auditing reference-label acquisition betting-based inference e-values probability calibration online learning binary class proportion estimation Replication package for the manuscript “Mean-preserving online shrinkage for prediction-assisted finite-population auditing” by Julian Hoxha and Marius Panxhi.This repository contains the complete computational materials used to reproduce the statistical experiments, tables, and figures reported in the manuscript and its technical supplement. The study considers sequential estimation of a positive-label proportion in a fixed finite population when predictions are available for every item but reference labels are acquired sequentially by simple random sampling without replacement.The principal proposed procedure is mean-preserving online shrinkage (MPOS) coupled with an explicit binary-design evidence factor and integer-total confidence-sequence inversion. MPOS adaptively contracts item-level working probabilities toward their remaining prediction mean using only previously acquired reference labels. The procedure is designed to preserve time-uniform coverage while controlling excessive prediction confidence.The package includes implementations and results for MPOS, the binary-design optimizer, integer-total confidence sequences, integer uncapped comparators, reference-only methods, finite-population prediction-powered inference (PPI) comparisons, native PPI/PPI++ fixed-budget comparisons, ridge-sigmoid and smoothed-isotonic calibration comparisons, record-disjoint benchmark evaluations, and the supporting OPC and RVO investigations described in the supplementary material.The evaluation includes recurring public PPI benchmarks and additional fixed datasets, with matched acquisition orders used for method comparisons. The package retains favorable, unfavorable, and inconclusive comparisons. It does not present repeated checking orders as independent application datasets, and fixed-budget PPI/PPI++ intervals are kept separate from anytime-valid confidence-sequence comparisons.The archive contains the executable Python code, required fixed inputs, prediction and reference arrays needed for reproduction, acquisition-order specifications, configuration files, individual experimental outputs, numerical validation checks, table-generation scripts, and figure-generation scripts. The supplied commands reproduce the manuscript’s reported numerical results and regenerate its figures and tables.The study uses existing public numerical datasets and released prediction arrays. No new human participants were recruited and no new annotation campaign was conducted. Reference-acquisition counts are therefore retrospective experimental measures of statistical workload rather than observed deployment expenses.Authors: Julian Hoxha and Marius Panxhi Corresponding author: Julian Hoxha, <a href="mailto:julian.hoxha@aum.edu.kw" target="_blank">julian.hoxha@aum.edu.kw</a>Please cite both this repository and the associated journal article when reusing the materials. 2026-10-01T06:32:11Z Software Software 10.6084/m9.figshare.34040517.v1 https://figshare.com/articles/software/Mean-preserving_online_shrinkage_for_prediction-assisted_finite-population_auditing_replication_package/34040517 CC BY 4.0