Domain-Level Covariance Analysis for Multilevel Survey Data With Structured Nonresponse
Domain-Level Covariance Analysis for Multilevel Survey Data With Structured Nonresponse
复制标题
DOI：
10.1198/016214508000000724
复制
发表时间：
2008-12-01
影响因子：
3.7
通讯作者：
Zaslavsky, Alan M.
中科院分区：
文献类型：
--
作者：
O'Malley, A. James;Zaslavsky, Alan M.
美国的医疗质量调查是对个人受访者进行的（即。医院病人、健康计划成员）来评估健康护理组织的绩效（即，医院、健康计划）。从而构成估计域。为了更好地理解和更简洁地报告质量的各个方面，我们分析了。质量在域级别上进行度量。而不是为观察到的反应和较低（患者）水平的无反应模式指定一个完整的参数模型。我们首先拟合考虑调查响应中的无响应模式的广义方差-协方差函数，然后使用这些广义方差-协方差函数为域平均响应指定似然函数。这使我们能够直接为不同项目的域均值之间的关系建模。因为反应尺度是有界的。我们假设这些平均值遵循截断的多元正态分布。我们使用EM算法计算最大似然（ML）估计，并使用马尔可夫链蒙特卡罗在贝叶斯模型下进行采样。终于来了我们对估计的或采样的域间协方差矩阵执行因子分析。使用后验抽签，我们评估后验分布的选定因素的数量和项目的分配组在传统的规则。我们比较ML估计这个因素结构与那些从几个贝叶斯模型与不同的先验分布的域间协方差。我们目前的数据分析来自医疗保健提供者和系统的消费者评估（CAHPS）的医疗保险优势健康计划的调查。
Health care quality surveys in the United States are administered to individual respondents (i.e.. hospital patients, health plan members) to evaluate the performance of health care organizations (i.e., hospitals, health plans). which thus Constitute estimation domains. For better understanding and more parsimonious reporting of dimensions of quality, we analyze relationships among., quality Measures at the domain level. Rather than specifying a full parametric model for the observed responses and the nonresponse patterns at the lower (patient) level. we first fit generalized variance-covariance functions that take into account nonresponse patterns in the survey responses, then specify a likelihood function for the domain mean responses using these generalized variance-covariance functions. This allows us to model directly the relationships among domain means for different items. Because the response scales are bounded. we assume that these means follow a truncated multivariate normal distribution. We calculate maximum likelihood (ML) estimates using the EM algorithm and sample under Bayesian models using Markov chain Monte Carlo. Finally. we perform factor analysis on the estimated or sampled between-domain covariance matrixes. Using posterior draws, we assess posterior distributions of the number of selected factors and the assignment of items to groups under conventional rules. We compare ML estimates of this factor structure with those from several Bayesian models with different prior distributions for the between-domain covariance. We present analyses of data from the Consumer Assessment of Healthcare Providers and Systems (CAHPS) survey of Medicare Advantage health plans.