Estimating the productivity impacts of technology adoption in the presence of misclassification

Abstract

This article examines the impact that misreporting adoption status has on the identification and estimation of causal effects on productivity. In particular, by comparing measurement error-ridden self-reported adoption data with measurement-error-free DNA-fingerprinted adoption data, we investigate the extent to which such errors bias the causal effects of adoption on productivity. Taking DNA-fingerprinted adoption data as a benchmark, we find 25% “false negatives” and 10% “false positives” in farmers’ responses. Our results show that misreporting of adoption status is not exogenous to household characteristics, and produces a bias of about 22 percentage points in the productivity impact of adoption. Ignoring inherent behavioral adjustments of farmers based on perceived adoption status has a bias of 13 percentage points. The results of this article underscore the crucial role that correct measurement of adoption plays in designing policy interventions that address constraints to technology adoption in agriculture.