Credible interval estimates for official statistics with survey nonresponse
Manski, Charles F.
Government agencies commonly report official statistics based on survey data as point estimates, without accompanying measures of error. In the absence of agency guidance, users of the statistics can only conjecture the error magnitudes. Agencies could mitigate misinterpretation of official statistics if they were to measure potential errors and report them. Agencies could report sampling error using established statistical principles. It is more challenging to report nonsampling errors because there are many sources of such errors and there has been no consensus about how to measure them. To advance discourse on practical ways to report nonsampling error, this paper considers error due to survey nonresponse. I summarize research deriving interval estimates that make no assumptions about the values of missing data. In the absence of assumptions, one can obtain computable bounds on the population parameters that official statistics intend to measure. I also explore the middle ground between interval estimation making no assumptions and traditional point estimation using weights and imputations to implement assumptions that nonresponse is conditionally random.
I am grateful to Aanchal Jain for excellent research assistance and to Bruce Spencer for helpful discussions. I have benefitted from the opportunity to present this work in a seminar at the Institute for Social and Economic Research, University of Essex.
official statistics; surveys; nonresponse