Strengthening causal inference via genetically informative studies

For many health-related phenotypes, randomized control trials are impractical or unethical, so alternative methods of causal inference are needed. Historically, longitudinal observational studies appear to have been the most... [ view full abstract ]

For many health-related phenotypes, randomized control trials are impractical or unethical, so alternative methods of causal inference are needed. Historically, longitudinal observational studies appear to have been the most popular, but more recently other research designs and analytical methods have been pursued.  Two more recent approaches are of interest to the behavior geneticist: the use of data collected from relatives such as twins, and Mendelian randomization.  The latter is an application of a mediation model for three variables, typically a genetic variant, an intermediate trait believed to be directly caused by it, and an outcome variable whose causal relationship to the intermediate trait is of interest.  Each method relies on assumptions, some of which may be difficult to test directly.  In this talk, I discuss the three approaches individually, and the potential for combinations of them to provide tests of key assumptions, and to yield actionable information on causal processes.  The further potential for multivariate methods, such as network modeling via partial correlation or structural equation modeling is also considered.