Quantifying the Impact of Sampling Error in Composite Index Construction

Abstract Composite indicators are widely used to assess multifaceted phenomena; however, they rarely account for the uncertainty inherent in input data. This study quantifies the uncertainty of composite index scores by propagating sampling error from input data, providing a foundation for more effective decision-making by reducing maladaptation when data quality is ignored. We analyze how the margin of error (MOE) propagates across different methodological configurations, including normalization, aggregation, and weighting, and conduct global sensitivity analysis to compare its influence against conventional design parameters. Results show that index scores are shaped as much by methodological choices as by empirical realities. We argue that data gaps and unknown unknowns must be recognized as part of the validation space, particularly when uncertainty varies systematically across geographies. Moreover, we find that communities most affected by social vulnerability often face the greatest data uncertainty, reinforcing patterns of epistemic injustice. These findings support the integration of MOE as a core design component in composite indices and advocate for scenario-based, uncertainty-aware frameworks. This reframes index construction from a technical task into one of epistemological humility and methodological responsibility.