Robust Dominance Levels: code and data for reporting composite-index rankings under construction uncertainty, with an application to the 2026 Environmental Performance Index

Data and code accompanying the manuscript “Robust Dominance Levels: An Algorithm for Reporting Composite-Index Rankings with Calibrated Order Statements” (Riansyah, Purwanto and Gernowo; submitted to the Journal of Information Systems Engineering and Business Intelligence).

Robust Dominance Levels (RDL) is a method for reporting a composite-index ranking only through the order statements that survive plausible construction choices. It (1) screens indicators by effective influence and pairwise discrimination, (2) simulates the construction space (weights at three levels, normalisation, aggregation, indicator set; optionally measurement error, RDL-M), (3) estimates for every pair of units the probability that one outranks the other, (4) keeps an order statement only when that probability is at least 1 − α, withdrawing any statement on a directed cycle so that the graph is acyclic, and (5) layers the resulting dominance graph into levels. A panel extension flags robust rank changes between two periods.

The method is applied to the 2026 Environmental Performance Index (EPI) of the Yale Center for Environmental Law & Policy: 177 countries, 47 indicators, 12 issue categories and 3 policy objectives, with current and baseline (about ten years earlier) indicator scores, and a focus on Indonesia and the ASEAN members. The package also contains the simulation evaluation that compares RDL with point ranks, quantile classes, rank-interval and tiering schemes on EPI and synthetic data, with 95% confidence intervals, a misspecification test and a sensitivity analysis of the specification of RDL.

Contents: data/ (indicator scores for the current and baseline periods, hierarchy and weights), code/ (the algorithm in rdl.py and the scripts for screening, uncertainty and Sobol analysis, application, evaluation and figures), results/ (every number reported in the manuscript, as JSON and CSV), figures/ (Figs. 1–4) and run_all.sh, which reruns everything with the fixed seeds used in the manuscript. A clean copy reproduces the shipped results exactly with Python 3.11 and the versions in requirements.txt.

The data are derived from the 2026 EPI (https://epi.yale.edu), which is distributed under CC BY-NC-SA 4.0; please also cite Wendling, Z. A., Emerson, J. W., Esty, D. C., de Sherbinin, A., Block, S., et al. (2026), 2026 Environmental Performance Index, New Haven, CT: Yale Center for Environmental Law & Policy. Code is released under the MIT licence; data, results and figures under CC BY-NC-SA 4.0.