Entropy-Based Uncertainty Management and Decision Support Under Strategic Agent Interactions in Institutional Survey...

Institutional governance systems increasingly rely on stakeholder surveys in strategic decision-making. Yet survey participants can act as strategic agents who shape organizational outcomes in their favor. By transforming the information content of response distributions, such behavior can systematically distort institutional decisions made under uncertainty. This study proposes a dynamic framework that detects strategic data manipulation in institutional surveys using Shannon entropy and Kullback-Leibler divergence, and converts this detection into decision support. The framework operates in three stages. First, it constructs a robust reference entropy profile from historical data. Second, it processes incoming survey responses as a sequential stream and compares them against this profile. Third, it detects manipulation at both the population and individual levels through a multi-layered anomaly scoring system. The reference profile was built from anonymized real survey data spanning 2021-2025, comprising 1233 participants and 19,728 clean observations. The framework was validated through 600 Monte Carlo scenarios derived from this profile, covering four manipulation types and five intensity levels, and was benchmarked against the Z-score and Isolation Forest methods. The findings are threefold. Straight-lining detection identifies individual suppression and inflation manipulations with perfect accuracy. KL divergence monitoring flags coordinated coalition entries before data collection is complete. Hierarchical clustering recovers a coordinated suppression group that individual scoring fails to isolate, with 76.5% cluster purity and ~59% recall. Policy impact analysis further shows that manipulation distorts dimensions in opposite directions: the gap between raw and verified means is positive in the dimension targeted by coordinated suppression but clearly negative in the dimension targeted by coordinated inflation. This bidirectional distortion shows why dimension-selective detection is necessary, as a single uniform correction cannot resolve it. The study contributes to the literature in two areas, integrating decision-making under uncertainty with strategic agent models, and survey integrity research with information-theoretic metrics.