Regency EF
A Caution in the Use of Bootstrap Confidence Intervals (304085)
*Chuchu Cheng, Boston College
Hao Wu, Vanderbilt University
Keywords: bootstrap confidence intervals, structural equation modelling
Bootstrap confidence intervals(CIs) are usually preferred for not relying on distributional assumptions. If the original sample is transformed before bootstrapping, the traditional formulations of bootstrap CIs are no longer appropriate. We used fit indexes in structural equation modelling as an example. Fit Indexes are popular measures of model fit . However, the CIs for most fit indexes are usually unavailable in software. Since their sampling distributions are analytically unknown, the CIs are difficult to construct, especially for nonnormal data. Nonparametric bootstrap is an ideal solution in these situations. While using YHY transformation in bootstrap CIs theoretically corrects the parent population, problems may arise if it is used blindly. Since the parent population after transformation is no longer the original sample, adjustments to the traditional formulations of bootstrap CIs are needed. We proposed adjusted formulae for bootstrap CIs to better apply YHY transformation. A simulation study was used to demonstrate this issue and out solution. The results showed the adjusted bootstrap CIs outperformed the original bootstrap CIs.