Causal Models for Patterns of Nonresponse
从因果模型角度研究分类变量缺失数据的无应答机制，发现多数因果模型意味着不可忽略的应答机制，并给出替代估计方法。
Abstract The problem of missing data for categorical variables is examined from the perspective of modeling the mechanisms of nonresponse. Log-linear causal models, as formulated by Goodman, are studied for the relationship of the survey variables to response; under some conditions several such models are estimable from the observed data. For nested patterns of nonresponse, a specific causal model represents exactly the assumption of ignorable response. Most causal models, however, imply nonignorable response mechanisms and yield alternative estimates for the distribution of the survey variables.