| Summary: | 碩士 === 國立陽明大學 === 公共衛生研究所 === 99 === Outcome-dependent sampling (ODS) design has been shown to have better performance in terms of coefficients estimation when sampling homoscedastic data via inverse probability weighting (IPW) method. In this study, we applied ODS method to sample heteroskedastic data, and investigate the performance of IPW coefficient estimation. We adopted weighted least square (WLS) method to study the influence of sampling design to parameter estimation after correcting error variance. ODS design tends to change the sample size of each stratum and samples in each stratum have different error variance measures. In this study, we combined weights from WLS and IPW methods to evaluate the performance of parameter estimation under the same simulation setting. According to the simulation results, we found that IPW method has larger standard error and WLS method tend to be biased in point estimation. However, the combined weighting method can produce results with smaller standard errors and biases. In this study, we concluded that when applying ODS design to heteroskedastic data, combined weighting method outperforms individual IPW and WLS methods in terms of coefficients estimation results.
|
|---|
