Investigation Report: User Payment Models and Willingness to Pay on Online Paid Q&A Platforms

Academic fraud report · Geng Detector

Summary

Verdict: CLEAR — no evidence of academic fraud. This paper is a social-science survey study with no experimental figures, only a single textual regression table (Table 1). The forensic image module flagged high PRNU correlation (NCC=1.000) across 15 pairwise comparisons of 8 page images, but this reflects batch provenance from a single rendering/scanning pipeline applied uniformly to all pages, not duplicated or spliced experimental content. Internal consistency checks all passed: ΔR² values reconcile arithmetically (0.21−0.01=0.20; 0.18−0.01=0.17), Cronbach's α (0.91, 0.93) falls in an acceptable range, sample proportions (73/401=18.20%, 328/401=81.80%) are internally consistent, and the timeline (submitted 2017-11-19, revised 2018-03-28, published 2018-07) shows no anachronisms relative to cited events and web access dates (2017-09-10, 2017-03-02). Limitations: the image-based fraud-detection methodology is not applicable to purely textual survey papers, and several checks were not feasible because raw data were unavailable. Confidence is high that the flagged image signals are artefactual rather than indicative of misconduct. (DOI: 10.13266/j.issn.0252-3116.2018.13.004)

Verdict

CLEAR. No evidence of academic misconduct was identified. The paper is a social-science empirical survey study (focus group + online questionnaire) with no experimental figures, statistical plots, or microscopy images; the only figure is a textual hierarchical regression table (Table 1). All forensic signals attributable to PRNU image correlation are benign artefacts of a single rendering pipeline, not evidence of image duplication or manipulation.

Key findings

Image forensics inapplicable (PRNU NCC=1.000 across 15 pairwise comparisons of 8 page images): Because the paper contains no experimental figures, the "images" being compared are uniformly rendered/scanned full pages. NCC=1.000 reflects a shared rendering pipeline, not duplicated experimental content. Methodological precondition for the forensic module is not met.
Mismatched detection modality: Geng's six-form image-reuse/splicing checks do not apply to a textual social-science paper; quantitative-statistics checks also cannot be executed at the text level because raw measurement data are not reported.
Regression arithmetic is internally consistent: For asking-payment, Step1 R²=0.01, Step2 R²=0.21, ΔR²=0.20 (matches 0.21−0.01). For viewing-payment, Step1 R²=0.01, Step2 R²=0.18, ΔR²=0.17 (matches 0.18−0.01).
Reliability statistics are reasonable: Cronbach's α = 0.91 and 0.93, both within an acceptable range.
Sample proportions reconcile: Of 401 complete questionnaires, 73 (18.20%) were paying users and 328 (81.80%) were non-paying users; 73+328=401 is consistent with total N.
Demographic proportions are consistent: Female:male ≈ 3:2; age 18–25 = 55.8% (≈183); bachelor's degree = 60.4% (≈198); ≤1500 RMB = 38.4% (≈126). No internal contradictions detected.
Willingness-to-pay means are nearly identical: Asking-payment M=3.13, SD=0.82; viewing-payment M=3.12, SD=0.84; paired t=0.20 (n.s.), direction consistent.
Timeline is coherent: Submitted 2017-11-19, revised 2018-03-28, published in Vol. 62, No. 13 (July 2018). Events described (Fenda launched May 2016; Zhihu/Yike renamed in 2018; Zhihu/Wbpay in 2016) and web access dates for references [3] (2017-03-02), [4][14][16][17] (2017-09-10) all pre-date submission. No temporal inversion.

Evidence highlights

Single textual Table 1 (hierarchical regression); no Figures in the paper.
Statistical reporting uses social-science conventions (t, R², ΔR², F, significance stars: * p<0.01, p<0.05, * p<0.1).
328 valid records used for analysis; 401 total complete responses collected; 73 paying vs. 328 non-paying, fully reconciling to N=401.
DOI: 10.13266/j.issn.0252-3116.2018.13.004; all three authors (李武, 艾鹏亚, 许耀心) provide ORCIDs.

Notes

Limitations: image-based fraud detection is not designed for purely textual social-science papers; without raw data, deeper statistical checks (e.g., distributional anomalies) cannot be performed.
Recommendation: the forensic module could incorporate a paper-type pre-filter to skip image forensics for non-image-bearing manuscripts.
This report is AI-assisted and intended for academic discussion only; final determinations of misconduct rest with formal institutional investigation.