Integrity Review Report: "Research on the Impact of Parallel志愿 Reform on Student Source Distribution in Higher Education Admissions"

Academic fraud report · Geng Detector

Summary

Verdict: CLEAN (no misconduct indicators found). This integrity review examined a 2025 empirical education-economics paper by Sun Zhijun and Liu Zitong (Beijing Normal University), published in Tsinghua Journal of Education (DOI: 10.14138/j.1001-4519.2025.05.007812), which uses a staggered difference-in-differences design to evaluate China's parallel-volunteer college admission reform (2005–2015 data). Six checks were performed. Pixel-level image analysis was not feasible due to lack of high-resolution source files, but the paper relies on standard econometric event-study plots rather than biomedical imagery. Arithmetic verification of reported interaction-term subtractions was internally consistent (e.g., 2.971 − 2.231 = 0.740 exact; 3.014 − 1.690 ≈ 1.3; 4.539 − 2.417 ≈ 2.1). A 500-iteration placebo test yielded 32 significant coefficients (6.4%), close to the nominal 5% Type-I rate, suggesting no p-hacking. The authors cite de Chaisemartin and D'Haultfœuille (2020) on negative weighting in TWFE, indicating awareness of modern staggered-DID pitfalls. Timeline (received Sept 2024, expected Oct 2025) is plausible. No anomalies warrant follow-up.

Verdict

Clean. No evidence of image manipulation, fabricated numbers, p-hacking, methodologically dated practices, or anomalous publication timeline. The authors demonstrate internal arithmetic consistency, appropriate placebo calibration, and awareness of contemporary econometric concerns. The paper is recommended as a methodological teaching example rather than a candidate for PubPeer commentary.

Key findings

Image reuse / splicing: No pixel-level analysis possible (raw images not submitted), but the paper contains only standard econometric figures (event-study coefficient plots, placebo-test histogram) — not Western blots or microscopy — and text descriptions show no inconsistencies. No concern.
Data fabrication / RNG signatures: All reported interaction-term arithmetic in Section 5(b) verifies exactly or within stated rounding (1.3, 0.74, 2.1). No random-number artifacts.
Statistical anomalies: Placebo test with 500 random assignments produced 32 p < 0.05 coefficients (6.4%), consistent with the nominal 5% Type-I error rate; no evidence of p-hacking or inflated significance.
Methodological awareness: Explicit citation of de Chaisemartin and D'Haultfœuille (2020) on TWFE negative-weight bias in staggered DID, with event-study dynamic-effects estimation adopted as a remedy.
Timeline: Received Sept 10, 2024; expected publication Oct 2025 (Vol. 46, Issue 5). Roughly one-year cycle is normal to slightly fast for the CSSCI-indexed Tsinghua Journal of Education. No conflict.
Data provenance: Secondary data sources (Sina Education–Gaokao portal; China Statistical Yearbook) disclosed transparently, including deletion of missing observations and restriction to first-tier batches.

Evidence highlights

Arithmetic check (Section 5(b), East-region universities):
Comprehensive: 3.014 − 1.690 = 1.324 → reported ≈1.3 ✓
Science track: 2.971 − 2.231 = 0.740 → reported 0.74 ✓ (exact)
Liberal-arts track: 4.539 − 2.417 = 2.122 → reported ≈2.1 ✓
Placebo test (Section 4.3): 32/500 = 6.4% false-positive rate; well within sampling variation around the 5% nominal level.
Methodology citation: de Chaisemartin, C. and D'Haultfœuille, D. (2020) on two-way fixed-effects estimators with heterogeneous treatment effects — author awareness of staggered-DID bias documented.
Submission metadata: Received 2024-09-10; DOI 10.14138/j.1001-4519.2025.05.007812.

Notes

Pixel-level forensic imaging was not feasible because high-resolution source figures were not included with the manuscript text; this limitation is acknowledged rather than treated as a clean pass on visual evidence.
All verdicts are based solely on the textual, tabular, and metadata provided in the reviewed PDF excerpt; raw replication data were not requested, consistent with the report's no-concern conclusion.
Confidence is high for arithmetic, statistical-plausibility, methodological, and timeline checks; confidence in the image check is limited to "absence of textual red flags."
Disclaimer: This report is AI-assisted and intended for academic discussion only; final misconduct determinations require investigation by qualified institutional bodies.