Explaining the Atomistic versus Ecological Fallacies in SES-Health Gradients

A long literature in economics and public health has examined the relationship between socioeconomic status (SES) and health, often relying on area-based measures when individual-level data are unavailable. Using comprehensive Dutch administrative data, we build on work estimating income-health gradients (e.g., Chetty et al. 2016) and study how the level of aggregation systematically shapes estimates by inducing so-called ecological or atomistic biases. Individual and area income proxy for different channels (e.g., personal resources vs. local surroundings), but are obviously correlated. Aggregating to the area level reduces meaningful variation and attenuates non-linearities. We find that area-level estimates exaggerate the income gradient relative to individual-level estimates and are less robust to including individual and area-level controls respectively. Moreover, interaction analyses reveal that area exposure is highly unequal: the health of low-income households are far more sensitive to neighbourhood conditions than high-income households. Together, these findings shed light on the sources of SES–health gradients and show how data aggregation affects both interpretation and empirical estimates.