Abstract This study addresses the challenge of addressing spatial heterogeneity in composite indicators when spatial interaction between geographies is very weak or nonexistent due to distance, making geographically weighted regression models unfeasible. The study explores the applicability of a locally weighted performance matrix generated by the Benefit of the Doubt (BoD) and Ordered Weighted Averaging (OWA) methods in recognizing that the relevance of factors associated with multidimensional phenomena varies significantly across geographies. The ability of these methods to operationalize spatial heterogeneity without a distance-based spatial weights matrix is demonstrated in the context-sensitive Global Innovation Index for OECD Countries. The results show that both methods perform well across explanatory power, information value, discriminant power, ranking uncertainty, and the proportion of outliers. However, the OWA correlates more strongly with GDP per capita, high-technology exports, patent applications, residents, research and development (R&D) expenditure, and researchers in R&D than the BoD. These results indicate greater compatibility between the OWA and the compensability assumptions of the national innovation systems theory. The weights of the OWA operator can be adjusted to reflect the theory that above-average-performing subsystems do not compensate for a poor-performing innovation subsystem. Furthermore, the delimitation of innovation capacity can be achieved by assigning higher weights to the poorer-performing sub-indicators. Finally, validation using the novel Locally Performance-Weighted Random Forest machine learning method confirms the superiority of the OWA over the BoD and its ability to represent the context-sensitive Global Innovation Index.
Libório et al. (Fri,) studied this question.