Data variability or expert judgment? A comparative analysis on the robustness of weighting methods in MCDM

Demirer, Ömür
OPSEARCH · Springer Nature · 2026-10-20 · DOI: 10.1007/s12597-026-01301-z

Аннотация

Objective weighting methods such as entropy, CRITIC, and standard deviation derive weights from data alone but are criticized for rewarding dispersion rather than importance, and when this criticism holds has not been established. This study determines when data-driven weighting is reliable and when it misleads. A factorial Monte Carlo simulation compares three objective methods with three subjective methods (AHP, BWM, SWARA) against a known ground truth, with the alignment between criterion importance and data dispersion as the primary factor. Results reveal a beneficial–harmful sensitivity gradient: entropy is the most discriminating yet most fragile method, and in the harmful regime its degradation exceeds that of any subjective method roughly tenfold when experts are unbiased. Robustness checks under alternative reference weights, biased experts, SAW and VIKOR aggregation, and varied matrix sizes confirm that the gradient is a property of the data, whereas the subjective advantage disappears under expert bias. The regime cannot be detected from the data alone, but a minimal-expert diagnostic, checking whether the data-driven weighting ranks the most important criterion near the top, provides an early warning and underpins a conditional decision guideline. A real microzonation case study with an eight-expert panel illustrates the framework. The broader implication is that the reliability of data-driven weighting is a property of the data-generating process rather than of the method, that it can be screened at the cost of a single expert judgment, and that the stakes grow with problem size.

Читать у издателя →

Похожие статьи