Simpson’s Paradox in the Engine Room: Unraveling Waste Generation on Training Ships

Accurate shipboard waste prediction is essential for MARPOL compliance, yet maritime research has predominantly relied on fleet-wide aggregated models that may obscure vessel-specific patterns.The occurrence of statistical paradoxes in hierarchical maritime data has not been systematically examined.This study provides the first systematic documentation of Simpson's Paradox in maritime operational environmental data, using shipboard waste generation as a case study.By analyzing engine running hours and waste generation from six Indonesian training ships, we demonstrate the risks of data aggregation in maritime predictive analytics.We compared fleetwide Generalized Linear Models with individual vessel regression approaches using 66 observations over 11 days.Simpson's Paradox emerged in Auxiliary Engine data: strong individual-level correlations (r = 0.993) were masked by weak fleet-wide correlation (r = 0.416), demonstrating how aggregation can fundamentally misrepresent underlying relationships.Individual ship models achieved substantially higher predictive performance (97.38% and 98.60%) than fleet-wide models (89.5% and 17.3%), with cross-validation (CV) confirming robustness.The findings reveal that fleet-wide aggregation can produce misleading predictions with significant operational consequences for waste storage planning and regulatory compliance.This study establishes the necessity of vessel-specific modeling in maritime environmental management and provides methodological guidance for analyzing hierarchical operational data.