Missing(ness) in Action: Selectivity Bias in GPS-Based Land Area Measurements

Land area is essential to analyses undertaken by agricultural economists. While handheld GPS-based area measurement has become a practical alternative to farmers' error-prone area assessments in household surveys across the developing world, non-ignorable shares of GPS-based area measures are often missing in public datasets. Using nationally-representative household survey data from Tanzania and Uganda, we demonstrate that the missing GPS-based plot areas constitute a non-random subset of the unit-record data, but that the missing data can be reliably simulated by multiple imputation, and that the effects of the imputation on key agricultural statistics are non-trivial. As GPS-based land areas could inform an extensive array of statistical and economic applications, we focus on one of many policy-relevant applications of the method while underscoring the importance of completed data analysis and highlighting the practicalities of operationalizing the approach to answer similar research questions. The application of the method in the estimation of the inverse scale-land productivity relationship confirms and strengthens the presence of such relationship in Tanzania and Uganda. The study demonstrates the usefulness of judiciously reconstructed GPS-based areas in alleviating concerns over potential measurement error in farmer-reported areas, and with regard to systematic bias in plot selection for GPS-based area measurement.