Measurement error and farm size: Do nationally representative surveys provide reliable estimates?

We assess the reliability of measured farm sizes (ownership holdings) in the Living Standard Measurement Study - Integrated Surveys on Agriculture (LSMS-ISA) in Ethiopia and Malawi based on three survey rounds (2012, 2014, 2016) in Ethiopia and four rounds (2010, 2013, 2016, 2019) in Malawi. By using the balanced panel of households that participated in all the rounds, we utilized the within-household variation in reported and measured ownership holdings that, to a large extent, were measured with GPSs and/or with rope and compass. While this gives reliable measures of reported holdings, we detect substantial under-reporting of parcels over time within households. We find that the estimated farm sizes within survey rounds are substantially downward biased due to systematic and stochastic under-reporting of parcels. Such biases are substantial in the data from both countries, in all survey rounds, and in all regions of each country. Based on the analyses, we propose that the maximum within-household reported farm sizes over several survey rounds provide a more reliable proxy for the actual farm size distributions, as these maximum sizes are less likely to be biased due to parcel attrition. The ignorance of this non-classical measurement error is associated with a downward bias in the range of 20-30% in average and median farm sizes and an upward bias in the Gini-coefficients for farm size distributions. We propose ideas for follow-up research and improvements in collecting these data types and draw some policy implications.