Forensic review: "Institutions and financial frictions: Estimating with structural restrictions on firm value and investment" (Claessens, Ueda, Yafeh, JDE 2014)

Academic fraud report · Geng Detector

Summary

This report assesses an econometrics paper on institutions and financial frictions, published in the Journal of Development Economics (2014). Verdict: cleared (white). No substantive academic misconduct indicators were identified. Biological image checks do not apply, so the review focused on data logic, statistical coherence, internal numbering, and methodology. Cross-checks of observation counts across Table 1b, Tables 2–4, and Table 8 were internally consistent: the 40 country-observation counts sum to 78,128 as claimed; dropping the US yields 62,525 (= 78,128 − 15,603); the relaxed-spec sample is 86,475. Reported t-values align with significance stars; weak-IV diagnostics (Kleibergen-Paap, Anderson-Rubin) appear in Table 8. The only noted irregularity is a duplicated equation number "(13)" used for the estimation equation in Section 3.2 and again for the measurement-error derivation in Section 4.6, a typographical/numbering issue of minor severity. No fabricated data, no P-hacking signature, and no citation/timeline inconsistencies detected. Confidence is moderate-to-high; limitations include inability to audit raw firm-level data.

Verdict

Cleared (清白). No actionable academic misconduct found. The paper passes cross-validation of sample sizes, statistical reporting, and timeline consistency. The only issue is a minor typographical duplication of equation number (13).

Key findings

Sample-size cross-check passes: 40 country observations in Table 1b sum to the headline 78,128. Dropping the US reduces the sample to exactly 62,525 (Δ = 15,603). An alternative external-finance specification yields 86,475 — all internally consistent.
Statistical reporting consistent: Reported t-statistics match significance markers (\*, \*\*, \*\*\*); p-values do not cluster at the 0.04–0.05 boundary, and coefficient signs remain stable across robustness tests.
Methodology self-consistent: Clustered standard errors and IV procedures are implemented as described; Table 8 reports standard weak-instrument diagnostics (Kleibergen-Paap, Anderson-Rubin).
Timeline and citations plausible: Submitted 2013-03-06, accepted 2014-05-21. References (e.g., Djankov et al., 2008; World Economic Forum, 2007) predate submission. Cross-sectional proxies (e.g., 2005, 2007) are transparently disclosed as data-availability compromises.
Minor formatting issue (non-fraud): Equation (13) is used twice — for the estimation equation in Section 3.2 and for the measurement-error derivation in Section 4.6 — with subsequent numbering incremented from the duplicate.

Evidence highlights

DOI: 10.1016/j.jdeveco.2014.05.004
Headline N = 78,128 firm-year observations; US contribution 15,603; reduced sample N = 62,525; alternative-spec N = 86,475.
Weak-IV tests (Kleibergen-Paap, Anderson-Rubin) reported in Table 8, consistent with the paper's IV/measurement-error framework.
Robustness variants (pre-tax income, US exclusion, alternative institutional proxies) preserve qualitative sign and significance patterns.

Notes

Image-based detection methods are not applicable to this empirical economics paper; review relied on numerical, statistical, and textual consistency checks.
Raw firm-level data were not re-analyzed; findings about data integrity are limited to internal cross-table arithmetic and reported statistics.
The duplicate equation number (13) is flagged as a low-severity editorial/typographical concern, not evidence of fabrication. If an erratum is issued, renumbering the second (13) (and downstream equations) would resolve it.