Title: Statistical modeling with imperfect data
Authors: Li Tang - St. Jude Children's Research Hospital (United States) [presenting]
Abstract: Imperfect data is a long-standing statistical problem in practice. It often occurs due to method limitations or cost considerations. For example, in many real studies, either an exposure or a response variable or both may be misclassified, resulting in a common type of imperfect data. Another example of imperfect data involves study designs incorporating pooled samples. The third type of imperfect data one often encounters in reality is missing information in variables. As such, potential threats to the validity of analytic results (e.g., estimates of effects) are widely known. Although much of the discussion has been made in literature, it is often restricted to oversimplified settings, which may not be satisfied in practice. Thus, clear illustrations of valid and accessible methods that deal with complicated settings are still in high demand. We propose novel frameworks that allow flexible modeling of common types of imperfect data and emphasize the utility when oversimplified assumptions are not met.