Scale dependence in energy poverty measurement: theoretical implications

Technical Summary: Scale Dependence in Energy Poverty Measurement

Problem Statement
Energy poverty (EP) is increasingly measured using multidimensional indices to identify vulnerable populations and guide policy. However, a critical unresolved issue is whether the phenomenon captured by these metrics remains consistent across different spatial scales. Current literature suffers from methodological heterogeneity, scalar variation, and restricted data access, which prevent coherent comparative evaluations. Existing studies often apply indices to non-harmonized datasets at different scales (household, municipality, country), making it difficult to distinguish whether variations in results stem from methodological design or the effects of spatial aggregation. Furthermore, the Modifiable Areal Unit Problem (MAUP) suggests that aggregation does not merely reduce resolution but can fundamentally alter the detection of household-level vulnerability and thermal deprivation patterns. There is a lack of systematic quantification regarding how composite energy poverty indices behave under harmonized empirical conditions across multiple scales.

Methodology
To address these gaps, the authors developed a multiscale validation framework and the Heidelberg Energy Poverty Dataset (HEP-Data).

• Dataset Construction: HEP-Data is a harmonized multiscale benchmark comprising 261,833 household-level observations. It utilizes a nested hierarchical territorial clustering process (MiniBatchKMeans) to aggregate data into six territorial scales: Household, Neighborhood, Parish, Municipality, Province, and Country. The dataset integrates real data, proxy variables, and Approximate Synthetic Values (ASVs) to ensure all 24 selected indices could be implemented under standardized conditions.
• Index Selection: The study analyzed 24 energy poverty indices derived from 23 publications, covering economic, multidimensional, and vulnerability-based approaches. Two indices (EVCI and GEVI) were excluded due to reliance on subjective criteria or cross-sectional variability requirements incompatible with the synthetic dataset.
• Analytical Protocol:
  1. Reproduction: Indices were reproduced using a hierarchical protocol of variable substitution and traceability, ensuring original calculation logics were preserved without structural modification.
  2. Scalar Analysis: The study grouped scales into Micro (Household), Meso (Neighborhood to Province), and Macro (Country) levels to assess dimensional information degradation and discriminatory capacity.
  3. Sensitivity and Stress Testing: A univariate sensitivity analysis was conducted using perturbations of ±5%, ±10%, and ±20% on input variables to calculate Normalized Sensitivity Coefficients (NSC). Additionally, stress tests were applied to evaluate index robustness under three scenarios: high vulnerability, no vulnerability, and average cases.

Key Results
The analysis reveals that spatial aggregation fundamentally reconfigures the measurement of energy poverty rather than simply aggregating household values.

• Multidimensional Information Degradation: As the scale moves from Micro to Macro, the dimensional structure of indices contracts significantly. At the Micro level, 141 variables across seven dimensions (including Infrastructure/Housing and Efficiency/Technology) are utilized. At the Macro level, this reduces to 48 variables, with Efficiency/Technology variables disappearing entirely and Infrastructure/Housing variables dropping from 39 to 5. This erosion limits the capacity to identify non-economic deprivations and complex vulnerability combinations visible only at local scales.
• Loss of Discriminatory Capacity: Aggregation alters the discriminatory power of indices. At the Country level, EP estimates for the same case study ranged from 0.00 to 0.88 depending on the index chosen. Binary indices showed extreme behavior: six of seven indices collapsed to 0.0% prevalence at scales larger than the Household level due to modal aggregation, while the TPR index converged to 100%. Only indices with specific aggregation algorithms (e.g., FGT, hEP) maintained stability across scales.
• Reorganization of Determinants: The hierarchy of critical variables shifts systematically with scale:
  • Micro Scale: Driven by the interaction of climatic exposure (Heating Degree Days, coefficient ~5.5) and household economic capacity (income coefficient ~-1.0).
  • Meso Scale: Climatic and spatial variables remain dominant, but distributive effects (e.g., gender gaps in poverty) emerge, which are invisible at the individual level.
  • Macro Scale: Thermal variables recede, and structural factors dominate. Household energy expenditure reverses its sign (from negative at micro to positive at macro, indicating that lower expenditure correlates with higher poverty due to consumption constraints). The national energy matrix (fossil vs. renewable share) becomes a primary sensitivity factor.
• Metric Architecture and Robustness: Stress tests revealed that conceptually similar indices respond differently to identical perturbations. For instance, expenditure-based indices like FGT showed high elasticity (spanning 0–0.974), while MEPI(2) remained invariant (0.333). This indicates that methodological choices (thresholds, normalization, internal structure) determine discriminatory capacity more than the number of variables included.

Significance and Claims
The paper claims to provide the first systematic quantification of scale dependence across composite energy poverty indices under harmonized empirical conditions. Its primary contributions are:

1. Structural Reconfiguration: It demonstrates that scale is a structural condition that redefines the informational architecture of energy poverty. Aggregation does not just smooth data; it suppresses the detection of hidden energy poverty (e.g., low consumption due to constrained income rather than low need) and shifts dominant determinants from household conditions to territorial and energy-system characteristics.
2. Methodological Validation: The study establishes that index selection is not neutral. It argues that metrics should be explicitly aligned with the scale of intervention: elastic indices for detecting household-level thermal deprivation and stable metrics for national benchmarking.
3. HEP-Data Benchmark: The authors introduce HEP-Data as a reusable, open resource for reproducing existing indices and testing new metrics under standardized, controlled conditions, addressing the lack of comparable benchmarks in the field.
4. Policy Implications: The findings suggest that macro-level indicators are insufficient for identifying household-level thermal deprivation and retrofit needs. Effective policy requires integrating metrics appropriate to the specific analytical scale, acknowledging that environmental and social objectives in energy policy depend on the spatial differentiation of measurement tools.

The authors maintain a modest stance, noting that HEP-Data is a methodological benchmark rather than a population-representative survey, and that the observed patterns reflect the specific context of Chile, necessitating future research in diverse geographic and socioeconomic settings to assess global generalizability.