Article Info
Authors: Tengfei Shao, Xu Wang, Kotomichi Matsuno, Masayuki Goto
Journal: Decision Analytics Journal
Publication date: 6 October 2026
Article: 100754
DOI: 10.1016/j.dajour.2026.100754
Access: Open access
Research Focus
The article audits claims made when a Data Envelopment Analysis intensity matrix is read as a reference network. It distinguishes between structure imposed by the envelopment programme, extremeness relative to a declared reference technology, and evidence that multiple technologies generated the observed units.
Analytics Protocol
The study audits a pipeline using one confidential retail panel and six public panels. Each claim is evaluated against an explicitly defined inferential object and reference model, with the analysis reporting observed rank, uncertainty, the model’s own error rate, and the point where the analysis stops.
DEA Reference-Network Findings
One exact result is that a basic feasible solution of the variable-returns-to-scale programme has at most m+s positive intensities per unit; the article reports no violation across 63 solver-tolerance cells. A degree-preserving rewiring null fires in every replicate of both structurally valid single-regime arms, indicating invalid size for that test.
Calibration and Interpretation
The declared-technology reference ensemble does not identify latent regimes. It misses regimes that are not visible as network blocks, fires on a single-technology panel with concentrated benchmark mass, and shows a false-fire rate moving from 0.025 to 0.125 between two plausible technologies. With 200 draws, it reproduces its verdict about half the time near the threshold.
Application and Limits
For the application panel and five public panels, the statistic falls below its reference and the verdicts are stable. On the application panel, object classification is unassessed; consequently, no community-informed decision rule is constructed as a confirmatory analysis. The study’s deliverable is a calibration record, not a validation certificate.
Key Takeaways
- DEA reference-network claims require explicit inferential objects.
- Reference models should declare their technology and error behavior.
- Observed ranks need uncertainty and reproducibility information.
- Unassessed classifications should stop confirmatory rule construction.
Cite this paper
APA: Shao, T., Wang, X., Matsuno, K., & Goto, M. (2026). An analytics protocol for calibrating benchmarking-community claims in data envelopment analysis. Decision Analytics Journal, 100754. https://doi.org/10.1016/j.dajour.2026.100754
MLA: Shao, Tengfei, et al. “An analytics protocol for calibrating benchmarking-community claims in data envelopment analysis.” Decision Analytics Journal, 2026, article 100754. https://doi.org/10.1016/j.dajour.2026.100754.
IEEE: T. Shao, X. Wang, K. Matsuno, and M. Goto, “An analytics protocol for calibrating benchmarking-community claims in data envelopment analysis,” Decision Analytics Journal, Art. no. 100754, 2026, doi: 10.1016/j.dajour.2026.100754.
Source
Read the original ScienceDirect article for complete publication and methodological details.
View the ScienceDirect article