Quantitative Evaluation of China's Pediatric Drug Policies Based on Policy Modeling Consistency Index Model (DOI: 10.12114/j.issn.1007-9572.2024.0456) — Forensic Report

Academic fraud report · Geng Detector

Summary

Verdict: Highly suspicious (orange). This preprint in Chinese General Practice applies a Policy Modeling Consistency (PMC) index to 23 Chinese pediatric drug policies. Because the underlying 'data' are subjective binary codings of policy texts rather than experimental measurements, the forensic focus shifts to internal consistency and factual accuracy rather than image fabrication. Three confirmed findings raise concern: (1) the impossible date '2021.09.31' for policy P15 in Table 1; (2) duplicated references [18] and [32] sharing an identical URL despite citing two distinct policies (third vs. fourth batch pediatric drug lists); (3) all 23 policies scoring identically 0.25 on X2 (policy time-frame) and 0.50 on X9 (policy perspective), the latter directly contradicting the paper's own claim of ~87% micro / ~13% macro perspective. Additional statistical anomalies (Benford deviation across four leading digits, last-digit parity imbalance) are noted but weakened by the small sample, bounded [0,1] range, and low-precision ratio structure of PMC scores. Overall posterior probability of misconduct is 52.7% (95% CI 19.6%–79.7%); findings warrant author response and editorial inquiry but do not constitute proof of fraud.

Verdict

Highly suspicious. Three findings are confirmed as standing on direct textual evidence; two statistical signals are flagged but partly mitigated by benign explanations (small-sample, bounded, low-precision ratio data); image-based forensic flags are explained by uniform software-generated charts. No finding independently proves fabrication, but the pattern—impossible date, duplicated reference URLs, and mathematically self-contradictory coding results across all 23 policies—exceeds the threshold for editorial follow-up.

Key findings

Finding 6 (CONFIRMED): Policy P15 listed with publication date "2021.09.31" in Table 1; September has only 30 days, making the date impossible (factual typo).
Finding 7 (CONFIRMED): References [18] and [32] share an identical URL (http://www.nhc.gov.cn/yaozs/s7656/201908/9a10b2382fe94d84817d9044d90dda15.shtml), but [18] cites the 2019 third-batch list while [32] cites the 2023 fourth-batch list—a clear copy-paste reference error.
Finding 8 (CONFIRMED): All 23 policies have X2 = 0.25 in Table 6, implying every policy has exactly one of four time-frame binary variables marked 1—an implausibly uniform result across heterogeneous policy documents.
Finding 9 (CONFIRMED): All 23 policies have X9 = 0.50 in Table 6, directly contradicting Section 3.1's claim of ~87% micro / ~13% macro perspective (mean should be 0.87 or 0.13, not 0.50).
Finding 1 (PARTIAL): Benford first–fourth digit distributions all deviate strongly (MAD up to 0.1715; χ² up to 768.8; all p = 0.0000), but applicability to small-sample, bounded, discrete PMC scores is limited.
Finding 2 (PARTIAL): Last-digit χ² = 32.4 (p = 0.0002) and odd/even ratio 31/94 = 0.33 (p = 0.0013), again partly attributable to small-denominator ratio structure.
Finding 3 (INSUFFICIENT): Column-difference uniformity flagged as potential additive fabrication, but X10 = 1.00 and X2 = 0.25 have plausible benign explanations; not a stand-alone smoking gun.
Findings 4–5 (BENIGN-EXPLAINED): PRNU NCC = 1.000 and Fridrich copy-move detections are explained by charts generated through a single software export pipeline and shared axis/grid elements, consistent with the paper's methodology description.

Evidence highlights

Table 1, P15: "2021.09.31" — impossible calendar date.
Reference list: [18] URL = [32] URL (identical), despite citing 2019 third-batch vs. 2023 fourth-batch lists.
Table 6, X2 column: 23/23 values = 0.25; column mean = 0.25.
Table 6, X9 column: 23/23 values = 0.50; contradicts §3.1 self-report of 87% micro / 13% macro.
Statistical forensics on Table 6 numeric columns: 4-digit Benford MADs 0.1020 / 0.0525 / 0.0770 / 0.1715; last-digit χ² = 32.4 (p = 0.0002); odd/even = 31/94.
DOI: 10.12114/j.issn.1007-9572.2024.0456 (ChinaXiv preprint 202412.00288v1.pdf, posted 2024-12-19).

Notes

Posterior probability: 52.7% (95% CI 19.6%–79.7%); BF = 21.18 (strong). Evidence treated as conditionally independent—may overestimate joint strength.
Recommended actions: request from authors the raw 41-binary-variable coding table for all 23 policies; raise queries on PubPeer; notify the journal editorial office and the authors' institutional research integrity committee.
Limits: PMC scoring depends on subjective coder judgment; uniform coding could reflect a single researcher's consistent (if implausible) interpretation rather than fabrication. The image-based signals (PRNU, copy-move) are adequately explained by a uniform chart-export pipeline.
Disclaimer: This is an AI-assisted forensic summary for academic discussion; final determination of misconduct requires institutional investigation.