|
Abstract:
|
In an audit data setting, two-stage stratified sampling is typically done on relatively small populations and small samples. In the paper "Empirical Study on the Second-stage Sample Size" (Liu, Batcher, Petska and Luo, 2004) where a sampling design at both stages was used, the impact of the second stage sample size on the overall estimation was investigated empirically in terms of the bias and variance estimation. Different methods of the classical design-based approach and resampling approach were compared. In this paper, we look at this issue from the model-based approach. We construct models at the overall level and at the stratum level and compare their performances to design-based and resampling approaches. We also check the assumptions of error terms and investigate the gain from the stratification with PSU.
|
