今日云讲堂 | 苗旺：Introducing the specificity score

嘉宾简介

报告摘要

There is considerable debate and doubt about the use of P value in scientific research in recent years, particularly after its use is banished in several prestigious journals. Much scientific research is concerned with uncovering  causal associations, however, P value is mostly a measure of the significance of a statistical association, which could be biased from the causal association of interest and lead to false/trivial scientific discoveries particularly in the presence of unmeasured  confounding. In this talk, I will introduce a score measuring the specificity of causal associations and a specificity score-based test about the existence of causal effects in the presence of unmeasured confounding. Under certain conditions, this approach  has controlled type I error and power approaching unity for testing the null hypothesis of no causal effect. A visualization approach using a heatmap of specificity is proposed to communicate all specificity score/test information in a universal and effective manner. This approach only entails a rough idea on the broadness of the causal associations in sight, e.g., the maximum or upper-bound number of causes/outcomes of an outcome/treatment, but does not  require to know exactly the exclusion of certain causal effects or the availability of auxiliary variables. This approach is related to Hill's specificity criterion for causal inference, but I will discuss the difference from Hill's. This approach admits for joint causal discovery with multiple treatments and multiple outcomes, which is particularly suitable for gene expressions studies, Mendelian randomization and EHR studies. Identification and estimation will be briefly covered. Simulations are used for illustration and an application to a mouse obesity dataset detects potential active effects of genes on clinical traits that are relevant to metabolic syndrome.