Aggregated Data and Compositional Variables: Methodological Note

The reform of personal data law in Europe makes it difficult to access personal data (even often non-nominative), especially when looking for data considered sensitive (and often income falls into this category). One solution often considered is the provision of spatially aggregated data. However, these data pose two technical problems. The first is that categorical data become compositions. The second is related to the ecological paradox that says it is dangerous to infer individual econometric relationships from aggregate data. We will see here how to work with compositional data (possibly just to validate a classical linear regression approach - easier to interpret). And we will discuss the second, but unfortunately it remains too general to be dealt with satisfactorily. Translated with www.DeepL.com/Translator