The research, published in Frontiers in Environmental Economics, analyzed energy and resource sustainability across all twenty-seven Member States from 2010 to 2023. Rather than ranking countries on a single index, the study treated each country-year as a point in a multidimensional space defined by six indicators: primary energy consumption, gross domestic expenditure on research and development, raw material consumption, net domestic greenhouse gas emissions, resource productivity, and municipal waste generation. The dataset comprised a balanced panel of 378 observations, drawn from Eurostat, covering every Member State for each of the fourteen years.

The methodological core of the study was a K-Means clustering analysis, a technique that partitions data points into groups by minimizing the distance between each point and the center of its assigned cluster. Before running the algorithm, the researchers standardized all six variables by converting them to z-scores, which removes the distorting effect of indicators measured on wildly different scales. Candidate solutions ranging from two to ten clusters were then evaluated using seven separate validation criteria, including the within-cluster sum of squares, the silhouette coefficient, the Calinski-Harabasz index, the Davies-Bouldin index, the Gap Statistic, cluster-size structure, and resampling stability. This multi-criteria approach is designed to prevent the common pitfall of choosing a cluster count that looks good on one metric but fails on others.

The joint validation produced a clear verdict: four clusters, not three. Earlier work in this literature had often imposed a three-cluster structure, but the statistical evidence in this dataset pointed elsewhere. The resulting four-cluster map of the EU is striking in its economic logic. Cluster 1, containing 59.26 percent of all observations, represents a lower-intensity resource and innovation profile, combining relatively modest energy use, R&D spending, and material throughput. Cluster 2 pairs stronger innovation and resource productivity with higher waste generation and emissions. Cluster 3 is perhaps the most distinctive: it combines very high research and development expenditure with very high raw material consumption, yet comparatively low net greenhouse gas emissions. Cluster 4 captures a persistent high-intensity profile, with elevated energy use, emissions, and resource consumption all occurring together.

These clusters are not merely statistical artifacts. They correspond to recognizable economic archetypes within the Union. The large lower-intensity group likely encompasses many smaller and less industrially intensive economies. The high-R&D, high-material, low-emissions cluster suggests economies that have decoupled their innovation and industrial output from their carbon footprint to a significant degree, possibly through a combination of nuclear power, renewables, and service-heavy economic structures. The high-intensity cluster points to Member States whose energy systems and industrial bases remain deeply carbon- and resource-intensive. The persistence of these groupings over fourteen years indicates that they reflect structural features of national economies rather than transient policy fluctuations.

To test whether these regimes were robust, the researchers examined how sensitive the clustering was to preprocessing choices. An unstandardized specification, which leaves the raw metric scales intact, produced a partition that agreed with the baseline at an adjusted Rand index of only 0.135, a value indicating near-random agreement. This confirms that scale distortion in the raw data is severe enough to completely reshape the cluster structure if not corrected. By contrast, winsorization at the first and ninety-ninth percentiles, which caps extreme values, reproduced the baseline partition almost exactly, with an adjusted Rand index of 0.981. A Yeo-Johnson transformation and RobustScaler preprocessing were also tested as robustness checks. The message is unambiguous: in sustainability clustering, how you prepare the data matters as much as how you cluster it.

The temporal analysis revealed perhaps the study’s most consequential finding. Out of 351 possible transitions between adjacent years, only 3 involved a country changing clusters, yielding a persistence rate of 99.15 percent. In other words, Member States overwhelmingly remained within their structural regime from one year to the next. Yet despite this near-total immobility between clusters, the system as a whole was converging. The multivariate root-mean-square distance from each country-year observation to the annual EU centroid declined by 14.45 percent over the study period. The beta-convergence coefficient, which measures whether countries with initially worse outcomes improve faster than those with better ones, was −0.0358 and statistically significant at the p < 0.001 level.

This combination of high persistence and measurable convergence has a specific interpretation. Convergence in the EU’s energy and resource sustainability is not occurring primarily through countries jumping between structural regimes. It is occurring through gradual, within-regime improvement: countries are getting closer to the EU average on the measured indicators while remaining recognizably the same kind of economy they were in 2010. The implication for policy is that convergence and structural change are two different phenomena, and the tools that promote one do not automatically promote the other.

The study also uncovered significant heterogeneity in how individual indicators behaved. While the multivariate picture showed overall convergence, resource productivity displayed statistically significant divergence, meaning that the gap between the best and worst performers on this metric widened over time rather than narrowed. Research and development expenditure, meanwhile, showed no significant linear convergence in either direction. This asynchrony matters because it means that innovation capacity, material intensity, resource productivity, energy use, and environmental pressure are not improving in lockstep across the Union. A policy framework that assumes uniform progress across all dimensions will misdiagnose where intervention is actually needed.

The authors argue that these findings support differentiated policy design rather than one-size-fits-all targets. Because the four clusters represent genuinely different structural configurations, the levers that would move a high-intensity Member State toward lower emissions are not the same as those that would improve resource productivity in a low-intensity one. The methodological contribution is equally pointed: sigma-convergence and beta-convergence diagnostics, when applied separately from the clustering step, reveal dynamics that a purely cluster-based analysis would miss. For researchers studying sustainability transitions in any multi-country setting, the study’s insistence on explicit k-selection, resampling stability, scale correction, and panel-aware temporal diagnostics offers a template for producing defensible results in a field where methodological shortcuts are common.

Subject of Research: Clustering and convergence dynamics of energy and resource sustainability across EU Member States from 2010 to 2023

Article Title: Energy and resource sustainability patterns in the European Union through clustering and convergence dynamics

Article References: Barbuta-Misu, N., Georgescu, P. L., Zlati, M. L., Fortea, C., & Antohi, V. M. (2026). Energy and resource sustainability patterns in the European Union through clustering and convergence dynamics. Frontiers in Environmental Economics, 5, Article 1943790. https://doi.org/10.3389/frevc.2026.1943790

Image Credits: AI Generated

DOI: 10.3389/frevc.2026.1943790

Keywords: European Union, energy sustainability, resource productivity, K-Means clustering, convergence analysis, greenhouse gas emissions, R&D expenditure, municipal waste, Eurostat, structural regimes, policy design, sigma-convergence