Journal of Official Statistics, Vol.3, No.4, 1987. pp. 375–387
Interval Estimation from Multiply-Imputed Data: A Case Study Using Census Agriculture Industry Codes
Donald B. Rubin and Nathaniel Schenker
Abstract:We describe the use of multiple imputation based on logistic regression models in a project to calibrate industry and occupation codes for public-use samples from the 1970 and 1980 United States Decennial Censuses. The coverage properties of interval estimates for two estimands are examined in a case study involving multiply-imputed 1980 agriculture industry codes. The use of just a small number of imputations per missing code is shown to yield much more accurate interval estimates than single imputation. Because the problem considered here involves high fractions of missing information, it is important when creating multiple imputations to account for uncertainty in estimating the parameters of the model for nonresponse. We relate these results to the theoretical results of Rubin and Schenker (1986) in simpler situations.
Keywords:Bayesian inference; logistic regression; missing data; nonresponse; public-use data; sample surveys.
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group