Does TargetDiscovery try to learn the causal graph from the data?
No. For the variable importance measure to have a simple causal interpretation, one will need to make sure that 1) the confounders are not on the pathway from the target variable to the outcome, and 2) the outcome occurs after the target variable and the confounders. In other words, the causal graph implied by this time ordering (confounders target variable outcome) is assumed to be known and will not be learned from the data. Our philosophy is that causal graphs cannot be realistically learned from data so that they will have to be specified based on external knowledge such as time-ordering information.
Related Questions
- When I try to retrieve data from the HSA I get an error message in the console about an invalid magic number for FITS file. What is going on?
- In BeadStudio’s Control Summary Report graph, how can I view the control data for the individual array section?
- Does TargetDiscovery try to learn the causal graph from the data?