Benford Screening of 1.2 Million Meta-Analytic Values Finds No Corpus-Level Digit Anomaly

Abstract

Do the numerical outputs of large-scale meta-analytic workflows follow Benford law, or do they show corpus-level anomalies that might signal data-integrity concerns? We extracted leading digits from 1,175,056 values - point estimates, standard errors, and p-values - across the Pairwise70 corpus of 403 Cochrane review specifications. Digit frequencies were compared with Benford expectations using the Nigrini mean absolute deviation, chi-squared testing, and mantissa-arc uniformity, per field and pooled. The corpus first-digit mean absolute deviation was 0.013 (95% CI 0.013-0.014), within Nigrini marginal-conformity band, although the million-value sample made the omnibus chi-squared and mantissa tests reject uniformity. Effect-estimate and standard-error fields conformed (MAD approximately 0.009), p-values deviated as expected for a bounded quantity, and no review exceeded the critical threshold after Bonferroni correction. At corpus level these Cochrane outputs show no digit anomaly suggesting systematic fabrication or computational distortion in the meaningful effect-size and precision quantities. Benford screening is indirect and cannot detect fabrication that deliberately preserves expected digit frequencies.