Investigation Report: Falsified/Fabricated Descriptive Statistics and Inconsistent Reporting in 'Are the Humanities Useless? — Research on the Impact of Humanities Research Output of Universities on Economic Development'

Academic fraud report · Geng Detector

Summary

This report raises serious concerns about possible data fabrication and reporting inconsistencies in a 2026 article in Fudan Education Forum (DOI: 10.13397/j.cnki.fef.2026.01.011) by Fang Jianfeng and Xu Zhitong. The central finding is a strong suspicion that missing provincial data were imputed with fixed constants (notably 0.42 and 0.43) prior to entropy-weight TOPSIS-type calculations, producing implausibly identical composite scores across years for provinces such as Jiangxi (2012–2016 and 2020 all scored 0.43), Qinghai, Hainan, Ningxia, Gansu, and Inner Mongolia. A second concrete inconsistency is identified: the text reports a Kleibergen-Paap rk Wald F statistic of 23.536, while Table 5's footnote gives 23.256—a discrepancy attributable to incomplete editing. Methodologically, the paper is criticized for cookie-cutter 'kitchen-sink' econometrics (baseline regression, two-step mediation, threshold regression, SDM, heterogeneity) combined with a questionable instrumental variable (interaction of Ming-Qing jinshi counts and a time trend). Verdict, on the balance of probabilities, is that the descriptive-statistics irregularities are highly suggestive of fabrication or coding error that invalidate downstream regression results; the F-statistic mismatch is clear-cut. Limitations: the report relies on the published PDF; raw data and code were not independently accessed.

Verdict

🔴 Confirmed serious concerns. The recurring fixed-value pattern (0.42/0.43) across multiple provinces and years in Table 3 is highly anomalous and inconsistent with genuine entropy-weight composite scores; a textual inconsistency in the reported Kleibergen-Paap rk Wald F statistic (23.536 vs. 23.256) further undermines reliability. The downstream regression and IV-based causal claims built on these inputs cannot be regarded as credible without raw-data verification.

Key findings

Ghost values 0.42 and 0.43: Identical composite scores recur across distant provinces and years, defying normal stochastic variation expected from entropy-weighted indices.
Jiangxi 2012–2016: Composite score reported as exactly 0.43 for five consecutive years, and still 0.43 in 2020.
Qinghai: 0.42 in 2012 and 2013, jump to 0.51 in 2016, drop to 0.02, return to 0.42 in 2023.
Hainan: Spikes to 0.42 in 2016, with surrounding years near 0.06–0.09.
Ningxia: 0.42 in 2014 and 2015.
Gansu: 0.42 in 2014.
Inner Mongolia: 0.42 in 2013.
F-statistic mismatch (p. 95): Body text states Kleibergen-Paap rk Wald F statistic = 23.536; Table 5 footnote states Kleibergen-Paap rk Wald F statistic 23.256[16.380] — values diverge even in the first decimal place.
'Kitchen-sink' econometrics and weak IV: Baseline regression, two-step mediation, threshold regression, Spatial Durbin Model, heterogeneity, plus a Ming-Qing jinshi × time-trend instrument whose exclusion restriction is implausible.

Evidence highlights

DOI: 10.13397/j.cnki.fef.2026.01.011
Location: Table 3, p. 94 (provincial humanities research output, 2012–2023); p. 95 (endogeneity discussion and Table 5 footnote).
Pattern: Repeated decimals 0.42 / 0.43 appearing across non-adjacent provinces and years, including a five-year consecutive run for Jiangxi.
Numeric discrepancy: 23.536 (text) vs. 23.256 (Table 5 footnote), with critical value 16.380 also shown in the table.

Notes

The fixed-value pattern is most parsimoniously explained by fixed-value imputation of missing data (e.g., constant 0.42) or by a coding/logic error in the entropy-weight calculation pipeline; deliberate fabrication cannot be excluded but is not demonstrable from the PDF alone.
Regression coefficients, mediation, threshold, and spatial-Durbin results are downstream of the suspect composite scores and therefore inherit the same reliability problem.
Suggested follow-up: obtain the raw provincial input matrix and entropy-weight TOPSIS code; demand the authors reconcile 23.536 vs. 23.256; consider a PubPeer post and a formal complaint to the journal editorial office.
This report is AI-assisted and intended for academic discussion only; any final determination of misconduct rests with the relevant institutional investigation.