In contrast to extensive attention on model selection for cross‐sectional data, research on model selection for longitudinal data remains largely unexplored. This is particularly the case when data are subject to missingness and measurement error. To address this important problem, we propose marginal methods that simultaneously carry out model selection and estimation for longitudinal data with missing responses and error‐prone covariates. Our methods have several appealing features: the applicability is broad because the methods are developed for a unified framework with marginal generalized linear models; model assumptions are minimal in that no full distribution is required for the response process and the distribution of the true covariates is left unspecified; and the implementation is straightforward. To justify the proposed methods, we provide both theoretical properties and numerical assessments. The Canadian Journal of Statistics 43: 498–518; 2015 © 2015 Statistical Society of Canada

中文翻译：

---

纵向数据边际分析的变量选择和推断程序，缺少观测值和协变量测量误差。

与广泛关注横截面数据的模型选择相比，纵向数据模型选择的研究仍未开发。当数据容易丢失和测量错误时，尤其如此。为了解决这个重要问题，我们提出了边际方法，该方法可以同时对缺少响应和容易出错的协变量的纵向数据进行模型选择和估计。我们的方法具有几个吸引人的特征：由于该方法是针对具有边际广义线性模型的统一框架开发的，因此适用性广泛。模型假设是最小的，因为响应过程不需要完全分布，并且真正的协变量的分布未被指定；而且实现起来很简单。为了证明所建议的方法的合理性，加拿大统计杂志43：498-518；加拿大统计局。2015©2015加拿大统计学会