There is no universally accepted scientific method for computing composite indexes. Every step of the construction process involves choices which are dependent on the problem at hand, the modeller, the data constraints and the desired outcome. In this study, using the framework proposed by UNECA (2020) as a guideline, we construct a prototype of the African Human Security Index for 10 countries (Rwanda, Ethiopia, Sudan, Zambia, Ghana, Mali, Angola, Cameroun, Niger and Kenya) from 2008 to 2019. Since weights can have a significant effect on the overall composite index, we present a comparative analysis where we contrast the statistical-based weighting approaches (e.g., Unobserved Component and Factor Analysis) with model-free methods (e.g., Equal and Geometric weighting). While all weighting schemes have their pros and cons, we show that, due to the multi-dimensional nature of the African Human Security Index, it is essential to adopt a non-compensatory logic (such as Factor Analysis) in defining optimal weights because this approach takes into account the overlapping information across indicators and correct for potential correlation and compensability issues by investigating the underlying structure of the data. Furthermore, unlike the Unobserved Component Method which imposes an identification condition of, at least, three indicators per dimension, the factorial approach is flexible and could summarize multi-dimensional realities with as little as one indicator per dimension. This argument is important to consider in the African context where relevant data could be scarce and cross-country comparison limited. Last but not the least, it is crucial to point out that model-free approaches are by definition statistically unbiased but care must be taken in their usage as they may both disguise serious failings in some dimensions and invite simplistic policy conclusions due the compensability issues they suffer from.
Read full abstract