Fraud Detection Report: 'The role of export competitiveness in driving Renminbi cross-border settlement: Micro-level evidence from China' (PLoS ONE, 2025)

Academic fraud report · Geng Detector

Summary

This report assesses a 2025 PLoS ONE article by Danlu Bu and Lin Jiang examining determinants of Renminbi cross-border settlement using firm-level data (DOI: 10.1371/journal.pone.0318099). The verdict is strong suspicion of systematic data fabrication. Five text-and-table anomalies are documented. The most damning are in Table 4, where every heterogeneity subgroup across firm size, industry concentration, and leverage categories reports an identical N of 5013, summing exactly to the full sample of 10026—an implausible outcome suggesting copy-paste rather than genuine subsample regressions. Table 3 Column (2) likewise reports the same N=10026 after 'lagging all control variables by one period,' which should have caused sample attrition given the 2014–2022 panel. Additional concerns include implausible extreme values (FC range -5.691 to 3.968) despite claimed 1% winsorization, an incoherent economic interpretation (a 226.5% decrease per unit change in a bounded index), and suspiciously precise year-over-year firm-level figures (3.93%, 7.31%, 8.33%) presented as motivating evidence. Confidence is high for findings 1 and 2; lower for findings 3–5, which require dataset verification. No image-based checks were possible.

Verdict

🔴 Strong suspicion of systematic data fabrication. Multiple internal inconsistencies between described methods and reported table values indicate that key regression outputs were likely fabricated or copied rather than computed from the underlying micro-data.

Key findings

Table 4 (heterogeneity analysis): Every subsample—high/low firm size, high/low industry concentration, high/low leverage structure—reports an identical N of 5013, summing to exactly 10026 (the full sample). In real panel data with missing values and varying median splits, perfectly symmetric subsamples are statistically near-impossible, strongly indicating copy-paste fabrication.
Table 3 Column (2) (lagged-controls robustness test): Despite the authors' claim that all control variables were lagged by one period to address endogeneity, N remains 10026. With a 2014–2022 panel, lagging should drop approximately one year's worth of observations, reducing N to roughly 8,000–9,000.
Winsorization vs. summary statistics contradiction: The authors claim 1% upper/lower winsorization, yet Table 1 shows FC minimum of -5.691 and maximum of 3.968. Approximately 100 observations each sitting at these extreme values is economically implausible for a foreign-currency settlement share.
Implausible economic interpretation: The text states that a one-unit increase in COMP reduces FC by 226.5%. However, COMP (entropy-weighted index) ranges only from 0.167 to 0.303, so a 'unit increase' cannot occur; nor can a proportional dependent variable decrease by more than 100%.
Suspiciously precise narrative figures: The introduction cites Zoomlion's '3.93%, 7.31%, 8.33%' year-over-year declines in foreign-currency-denominated items over 2020–2022, an unusually precise sequence not typically disclosed in annual reports, suggesting reverse-engineered or fabricated motivating evidence.

Evidence highlights

Table 4 (pp. 13–14): all subgroup N values = 5013; 5013 × 2 = 10026 = full sample N.
Table 3 (p. 11), Column (2): N = 10026 despite claimed one-period lag of controls; panel covers 2014–2022 (9 years).
Table 1 (p. 7): FC Min = -5.691, Max = 3.968 while 1% winsorization is declared.
Page 9: stated coefficient interpretation "decreases by 226.5%" for a unit change in COMP (range 0.167–0.303).
Page 3: Zoomlion cited year-over-year declines of "3.93%, 7.31%, 8.33%" for 2020–2022.
DOI: 10.1371/journal.pone.0318099 (PLoS ONE, published April 11, 2025).

Notes

Image-based duplication and splicing checks (the first and third detection methods) could not be performed because only the text PDF was available; pixel-level and noise-level scans were not feasible.
Confidence is high for Findings 1 and 2 (direct numerical contradictions with stated methods). Findings 3–5 are suggestive but require verification against the authors' raw data and Stata/R do-files.
This report is AI-assisted and intended as a discussion prompt; final determination of misconduct requires investigation by the journal (PLoS ONE) and/or the authors' institution (Southwestern University of Finance and Economics).