Estimation of the latent mediated effect with ordinal data using the limited-information and Bayesian full-information approaches
Estimation of the latent mediated effect with ordinal data using the limited-information and Bayesian full-information approaches
复制标题
DOI：
10.3758/s13428-014-0526-3
复制
发表时间：
2015-12-01
影响因子：
5.4
通讯作者：
Choi, Jaehwa
中科院分区：
文献类型：
--
作者：
Chen, Jinsong;Zhang, Dake;Choi, Jaehwa
在社会或行为研究中，经常会遇到带有有序数据的潜变量。虽然潜在变量与有序数据的中介效应（潜在中介效应，或LME）可能看起来是LME与连续数据和潜在变量与有序数据的直接组合，但联合收割机将两者结合起来的方法学挑战并不简单。本研究涵盖了复杂的模型结构LME和制定有序数据使用贝叶斯全信息方法的点和区间估计LME。我们还结合联合收割机加权最小二乘（WLS）估计与偏差校正自举（BCB;埃夫隆杂志的美国统计协会，82，171-185，1987）的方法或传统的三角洲方法作为有限的信息的方法。我们通过模拟研究评估了这些不同方法在各种条件下的可行性，并提供了一个实证例子来说明这些方法。我们发现，贝叶斯方法与合理的信息先验是首选点和区间估计的兴趣和样本量为200或以上。
It is common to encounter latent variables with ordinal data in social or behavioral research. Although a mediated effect of latent variables (latent mediated effect, or LME) with ordinal data may appear to be a straightforward combination of LME with continuous data and latent variables with ordinal data, the methodological challenges to combine the two are not trivial. This research covers model structures as complex as LME and formulates both point and interval estimates of LME for ordinal data using the Bayesian full-information approach. We also combine weighted least squares (WLS) estimation with the bias-corrected bootstrapping (BCB; Efron Journal of the American Statistical Association, 82, 171-185, 1987) method or the traditional delta method as the limited-information approach. We evaluated the viability of these different approaches across various conditions through simulation studies, and provide an empirical example to illustrate the approaches. We found that the Bayesian approach with reasonably informative priors is preferred when both point and interval estimates are of interest and the sample size is 200 or above.