Interpretability of Composite Indicators Based on Principal Components
Kris Boudt, Marco d’Errico, Hong Anh Luu, Rebecca Pietrelli
Abstract
Principal component approaches are often used in the construction of composite indicators to summarize the information of input variables. The gain of dimension reduction comes at the cost of difficulties in interpretation, inaccurate targeting, and possible conflicts with the theoretical framework when the signs in the loading are not aligned with the expected direction of impact. In this study, we propose an adjustment in the construction of principal component approaches to avoid these problems. The effectiveness of the proposed approach is illustrated in defining the Food and Agriculture Organization of the United Nations’ Resilience Capacity Index, which is used to measure household-level resilience to food insecurity. We conclude that the robustness gain of using the new method improves the reliability of the composite indicator.