On longitudinal data, the measurements collected from the same subject are correlated. Liang and Zeger (1986) (Ref.1) developed the generalized estimating equations (GEE) by incorporating correlation matrix, for this purpose. This approach assures consistency even when correlation structure is not correctly specified. The correlation structures commonly used include exchangeable MA (1),and AR (1). Quantile regression is another alternate method for analyzing repeated measurements because of its ability to describe the whole conditional distribution. One method is to assume an independence working model. This though has many practical advantages, but may lose efficiency in parameter estimation in the presence of high correlation. Many approaches were developed by various studies to overcome this problem. This article uses a Gaussian pseudolikelihood approach to simultaneously estimate the correlation parameters and select a working correlation structure. This is achieved by estimating the parameters and calculating the corresponding Gaussian pseudolikelihood for the working correlation models and the one with dominant pseudolikelihood criterion can be accepted as the correlation structure will be very close to the true structure (23 Refs.)
展开▼
机译：在纵向数据上，将从同一对象收集的测量值相关联。为此，Liang and Zeger（1986）（Ref.1）通过引入相关矩阵开发了广义估计方程（GEE）。即使未正确指定相关结构，此方法也可确保一致性。常用的相关结构包括可交换的MA（1）和AR（1）。分位数回归是分析重复测量的另一种替代方法，因为它能够描述整个条件分布。一种方法是假设独立工作模型。尽管这具有许多实际优势，但是在存在高相关性的情况下可能会失去参数估计的效率。各种研究开发了许多方法来克服这个问题。本文使用高斯伪似然方法来同时估计相关参数并选择一个有效的相关结构。这是通过估算参数并为工作相关模型计算相应的高斯伪似然性来实现的，并且可以接受具有显性伪似然准则的模型，因为相关性结构将非常接近真实结构（23参考）。
展开▼
