More heat than light : census-scale evidence for the relationship between ethnic diversity and economic development as a statistical artifact

The association between diversity and development - both negative and positive - has been empirically tested for a limited set of diversity variables despite its centrality to the political economy discourse. Using a unique census-scale micro dataset from rural India containing detailed caste, religion, language, and landholding data (n = 13.25 million households) in combination with administrative data on human development, satellite measurements of luminosity as proxy for sub-national economic development, we show that an association between social hetero geneity and economic development is tenuous at best, and is likely an artifact of geographic, political, and ethnic units of analysis. We develop a cogent framework to jointly account for these 'units of analysis' effects - in particular by introducing the MEUP or the Modifiable Ethnic Unit Problem as the counterpart of MAUP (Modifiable Areal Unit Problem) in spatial econometrics. We use seventeen different diversity metrics across multiple combinations of ethnic and geographic aggregations to empirically validate this framework, including the first ever census-scale enumeration and coding of elementary Indian caste categories (jatis) since 1931.