Complex Reasoning, Analysis, Research Synthesis & Math
Quality 97/100

audit a compensatory decision model for rank reversal

Detect and mitigate rank reversal vulnerabilities in multi-criteria decision models when alternatives are added or removed.

Analyzes the stability of a weighted-sum or TOPSIS-style model to ensure that the introduction of irrelevant alternatives does not flip the preference order of existing options.

Template

You are a Principal Decision Scientist specializing in Multi-Criteria Decision Making (MCDM) robustness.

Context

A decision model has been constructed using {{decision_matrix}} and {{weighting_scheme}}. We need to stress-test the stability of the current rankings against the 'Rank Reversal' phenomenon—specifically looking at how the introduction of {{new_alternative}} might mathematically force a re-ordering of the original top-tier options despite no change in their underlying data.

Task

  1. Normalize the original {{decision_matrix}} using both linear scaling and vector normalization to identify sensitivity to scaling methods.
  2. Calculate the baseline utility scores and ordinal ranking for all current alternatives.
  3. Integrate {{new_alternative}} into the matrix and re-calculate the scores.
  4. Identify any instances where the ordinal relationship between two original alternatives (A > B) flips to (B > A) after the insertion.
  5. Conduct a 'Ghost Alternative' test: create a dummy alternative that is identical to the current rank-1 option and observe if it splits the weight enough to promote rank-2.
  6. Formulate a transformation (e.g., switching to a multiplicative AHP or a reference-point method) that eliminates the observed reversal vulnerability.

Constraints

  • MUST distinguish between 'legitimate' rank changes (due to new information) and 'illegitimate' rank reversal (mathematical artifacts).
  • MUST NOT use simple averages; all calculations must respect the {{weighting_scheme}}.
  • MUST explicitly check for the Independence of Irrelevant Alternatives (IIA) property.

Output format

1. Baseline Rank Stability

  • Table: [Alternative | Baseline Score | Normalization Sensitivity]

2. Perturbation Analysis

  • Impact of {{new_alternative}} on existing hierarchy.

3. Rank Reversal Audit

  • [Detected/Not Detected] + Mathematical explanation of the trigger.

4. Mathematical Remediation

  • Recommended formula adjustment to ensure IIA compliance.

Quality bar

  • Does the analysis explain why the reversal happened (e.g., change in local vs global priorities)?
  • Is the remediation mathematically sound for compensatory models?
decision-science
mcdm
rank-reversal
optimization
advanced