Testing competing explanations for the inverse productivity puzzle

We use plot-level data from ICRISAT to assess competing explanations for an old empirical regularity - the inverse relationship between land productivity and farm size. The presence of farmers who simultaneously crop multiple plots with di¤erent sizes is used to test (and reject) explanations based on household heterogeneity. The panel nature of the data is explored to test (and refuse) explanations based on plot fixed characteristics. We are then left with explanations based on time-varying plot features or measurement errors in the plot size. Theoretically, the input choices should reflect both plot-specific features and the true plot size. Empirically, the inverse relationship vanishes when we control for input use.