A Modified Extended Delete a Group Jackknife Variance Estimator Under Random Hot Deck Imputation in Business Surveys [chapter]

Paolo Righi, Stefano Falorsi, Andrea Fasulo
2014 Contributions to Statistics  
Item nonresponses commonly trouble the large scale surveys and if corrections are performed extra variability is introduced in the sampling errors. When the imputed values are treated as if they were observed the precision of the estimates are generally overstated. Modication of a variance estimator for contemplating item nonresponses is a ticklish issue. There is not a common judgement on which is the best estimation method. Usually the imputation procedure, the variance estimator properties
more » ... d the cost-eectiveness issues lead to choose a specic method. In the paper is proposed a method based on grouped jackknife (EDAGJK [2], [1]) easy to implement, not computer intensive and suitable whit random hot deck imputation. A simulative comparison on real business data with the bootstrap method with imputed data ([4]) and the Multiple Imputation ([3]) has been carried out. In the simulation the sampling strategy of Italian Small and Medium Enterprises survey has been taken into account and Taylor linearization technique when imputed data are treated as true values has been cosidered as well. The ndings show that the proposed method has good performances with respect to the other ones and it outperforms them in the terms of time spending. This paper summarizes some results of the BLUE-Enterprise and Trade Statistics project (BLUE-ETS project -Work Package WP6, http://www.blue-ets.istat.it). Main results Y1 Y2 Y3 Y4 18.36% 1.88% 8.73% 2.19% Table 1: Missing rates for the simulated nonresponse mechanism
doi:10.1007/978-3-319-05320-2_14 fatcat:j32yccpc5jgfvkxxynjhankxk4