Ranking
STOCHASTIC-UTA: Stochastic UTilités Additives (preference-disaggregation under uncertainty)
Stavrou, D. I., Ventikos, N. P., Tsoukalas, V. D. · 2018
Strengths
- •Recovers weights implicitly from the DM ranking (no separate weight elicitation)
- •Explicit utility functions per criterion (interpretable; supports sensitivity analysis)
- •Native uncertainty handling via Monte Carlo + SMAA acceptability indices
- •Quantified robustness indices (b_i^r, a_i, central weights, confidence factor)
Limitations
- •Computationally heavy in Monte Carlo mode (N LP solves)
- •Requires DM to commit to a reference ranking (cognitively demanding)
- •LP may admit multiple optimal vertices: solver-dependent tie-breaking
- •Piecewise-linear assumption may be too restrictive for highly non-linear preferences
Method assistant
Grounded explanations: it explains the method, it does not compute.
Assumptions to verify
- •DM is able and willing to supply a reference ranking of all (or a representative subset of) alternatives
- •Preferences are additively separable across criteria
- •Piecewise-linear marginal utilities are an adequate approximation
- •For Monte Carlo mode: per-cell probability distributions are well-calibrated
When not to use
- •No reference ranking available: use SMAA or direct weighting
- •Criteria are strongly interacting: use ANP or Choquet integral
- •Need a closed-form weight vector explicitly: use AHP / BWM
Edge cases
- •LP infeasible (δ* too large): σ+, σ- become non-zero; solver still returns optimal but Z > 0: surface as warning.
- •Multiple LP optima: HiGHS deterministic tie-breaking → fixture reproducible. For research-grade use, average vertices (UTA stability analysis).
- •All alternatives tied in DM ranking: LP becomes trivial; method falls back to equal-weight average.
- •Cost criterion: negate column upstream (F1) before LP.
Common pitfalls
- •Reading u_j(g_max) as 'criterion weight' directly: that's true at the normalisation hyperplane but the actual u_j(g_ij) used in u(A_i) is the piecewise-linear interpolation, not u_j(g_max) alone.
- •Confusing Monte Carlo iterations N with breakpoints α_j: orthogonal LP knobs.
- •Skipping the σ+, σ- slack variables: without them the LP becomes infeasible whenever the reference ranking is even slightly inconsistent.
Commonly paired with
- •n_a_internal + STOCHASTIC-UTA (common)
How to cite
Stavrou, D. I.; Ventikos, N. P.; Tsoukalas, V. D. (2018). Robust Evaluation of Risks in Ship-to-Ship Transfer Operations: Application of the STOCHASTIC UTA Multicriteria Decision Support Method. In Lee, P. T. W. & Yang, Z. (Eds.), Multi-criteria Decision Making in Maritime Studies and Logistics (pp. 161-185). Springer.. https://doi.org/10.1007/978-3-319-62338-2_8