Comparison of bootstrap methods for missing survey data: A simulation study

Article type: Research Article

Authors: Ahmad, Tauqueer^{; *} | Singh, Randhir | Rai, Anil

Affiliations: Indian Agricultural Statistics Research Institute, Library Avenue, New Delhi – 110012, India

Correspondence: [*] Corresponding author. E-mail: [email protected]

Abstract: Bootstrap technique is used in the estimation of variance of non-linear statistics in case of complex surveys. This technique is gaining popularity for survey data with missing observations. In this paper, bootstrap techniques with missing observations have been compared through a simulation study under different imputation techniques. The technique namely "Proportional Bootstrap Without Replacement (PBWO)" for missing observations has also been compared with the Rescaling Bootstrap Without Replacement (RSBWO) method for complete data set. Further, the efficiency of Proportional Bootstrap With Replacement (PBWR) technique for missing observations has been compared with the standard bootstrap technique for complete data set. An optimum number of bootstrap samples required for the reliable estimation of variance in the case of missing observations has also been obtained.

Keywords: Variance estimation, imputation techniques, without replacement, Monte-Carlo estimator

DOI: 10.3233/MAS-2006-1108

Journal: Model Assisted Statistics and Applications, vol. 1, no. 1, pp. 43-49, 2006