December 3, 2015, 4:00pm
Location: Department of Statistics and Actuarial Science, University of Waterloo
Room: M3 – 3127
Speaker: J.N.K. Rao, Carleton University
Abstract: Item nonresponse occurs frequently in sample surveys collecting data on many items. It is customarily handled by some form of imputation to fill in the missing item values. However, imputation values are often treated as if there were true values in making inference form the imputed data sets. Imputed point estimators are often valid but the “naïve” variance estimates, treating the imputed values as true values, can lead to serious underestimation of the true variance even for large samples because the additional variability due to estimating the missing values is not taken into account. Impressive advances have been made on making efficient and asymptotically valid inferences from singly imputed data sets. The main purpose of this talk is to present an overview and appraisal of methods for variance estimation under single imputation. Fractional imputation and multiple imputation both use multiple imputed values for a missing item and reduce imputation variance relative to single imputation using only one randomly imputed value. Variance estimation under fractional and multiple imputation will also presented. Finally, the construction of reliable bootstrap confidence intervals under imputation will be examined.
This talk is sponsored by the CANSSI CRT Project “Statistical Inference for Complex Surveys with Missing Observations”.
Back to Seminar Series Schedule.