Raymond L. Chambers and Robert G. Clark
- Published in print:
- 2012
- Published Online:
- May 2012
- ISBN:
- 9780198566625
- eISBN:
- 9780191738449
- Item type:
- chapter
- Publisher:
- Oxford University Press
- DOI:
- 10.1093/acprof:oso/9780198566625.003.0012
- Subject:
- Mathematics, Probability / Statistics
Survey inference via sub-sampling describes the different resampling approaches that can be used in finite population inference. It starts with a discussion of the interpenetrated sampling idea that ...
More
Survey inference via sub-sampling describes the different resampling approaches that can be used in finite population inference. It starts with a discussion of the interpenetrated sampling idea that underlies the independent sub-groups methods of variance estimation, then moves on to the popular jackknife method, which uses the variation between dependent subgroups to estimate the prediction variance of a complex estimator. A linearised version of the jackknife that avoids its heavy computational burden in large samples is described. The bootstrap technique is finally introduced and two variants, the naive unconditional bootstrap and the conditional model-based bootstrap, are discussed. A modification to the latter bootstrap that makes it robust to misspecification of the second order properties of the working model for the population is proposed.Less
Survey inference via sub-sampling describes the different resampling approaches that can be used in finite population inference. It starts with a discussion of the interpenetrated sampling idea that underlies the independent sub-groups methods of variance estimation, then moves on to the popular jackknife method, which uses the variation between dependent subgroups to estimate the prediction variance of a complex estimator. A linearised version of the jackknife that avoids its heavy computational burden in large samples is described. The bootstrap technique is finally introduced and two variants, the naive unconditional bootstrap and the conditional model-based bootstrap, are discussed. A modification to the latter bootstrap that makes it robust to misspecification of the second order properties of the working model for the population is proposed.
