A Non-Parametric Approach to Heterogeneity Analysis

This paper introduces a network-based method to capture heterogeneity in consumer microdata. We develop a permutation-based approach that repeatedly combines random samples of all agents' decisions, and partitions agents into jointly rational types. Aggregating these partitions yields a network that captures unobserved heterogeneity, where edges measure how often two agents share the same type across partitions. To evaluate how observable characteristics align with the heterogeneity, we implement permutation tests that shuffle covariate labels across network nodes, thereby generating a null distribution of alignment. We show that this test is exact, with asymptotic power of one. We further propose network-based measures that quantify whether nodes with the same observable attributes are disproportionately linked or clustered, along with standardized effect sizes that gauge each covariate's global influence. This yields a flexible, nonparametric measure of the heterogeneity structure. Finally, we apply our method to grocery expenditure data from the Stanford Basket Dataset.