Mon, April 10, 2017, noon:
Vimal Ranchhod (University of Cape Town)
10/14/2013, 12:00:00, 6050 ISR - Thompson
How prevalent is data fabrication by fieldworkers in South African surveys? Does this substantially affect the validity of subsequent empirical analyses? We document how we diagnosed such misbehaviour in the longitudinal National Income Dynamics Study. We found that the existence of fabrication was non-trivial, and affected about 7% of the sample. Since the fabrication was detected while fieldwork was still on-going, the relevant interviews were re-conducted and the fabricated data was replaced with authentic data. We thus have an observed counterfactual that allows us to measure how problematic such fabrication would have been, had it remained undetected. We implement a number of estimators using the data that includes the fabricated interviews, and compare these with the corresponding estimates that include the corrected data instead. For the outcomes that we consider, we find that while the two estimates do differ, the magnitudes of the differences are small and generally not substantial. This occurs for two reasons. Firstly, the fraction of fabricated data is relatively small, and secondly, the fabricated data is not too different from the corrected data. We conclude with a policy discussion for survey organizations that might have similar concerns.