Working Paper: NBER ID: w29749
Authors: Fernando M. Aragon; Diego Restuccia; Juan Pablo Rud
Abstract: We examine empirically whether the level of data aggregation affects the assessment of misallocation in agriculture. Using data from Ugandan farmers, we document a substantial discrepancy between misallocation measures calculated at the plot and at the farm levels. Estimates of misallocation at the plot level are much higher than those estimated with the same data but aggregated at the farm level. Even after accounting for measurement error and unobserved heterogeneity, estimates of misallocation at the plot level are extremely high, with nationwide agricultural productivity gains of 562%. Furthermore, we find suggestive evidence that granular data may be more susceptible to measurement error in survey data and that data aggregation can attenuate the relative magnitude of measurement error in misallocation measures. Our findings suggest caution in generalizing insights on measurement error and misallocation from plot-level analysis to those at the farm level.
Keywords: misallocation; agriculture; plot-level analysis; farm-level analysis; Uganda
JEL Codes: O11; O13; O4; O55; Q1
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| data aggregation level (C43) | perceived extent of misallocation (D61) |
| plot-level data (Y10) | inflated estimates of reallocation gains (D61) |
| measurement error (C20) | inflated productivity dispersion (O49) |
| land input (Q24) | productivity (plot level) (E23) |
| land input (Q24) | productivity (farm level) (E23) |
| plot-level analysis (D79) | misleading conclusions about misallocation (D61) |
| farm-level data (Q12) | accurate assessment of misallocation (D61) |