Commentary on "Mediation analysis without sequential ignorability: Using baseline covariates interacted with random assignment as instrumental variables" by Dylan Small.

I applaud Dr. Small for advancing causal mediation analysis and thank the editors for the opportunity to comment on this valuable article. Small's project was to relax and test the assumptions on which a previously proposed model relies; in the second half of this discussion I will assess those assumptions and others on which the model hinges. But first I will review the various schools of mediation analysis and situate the estimand considered by Small within the somewhat esoteric domain of mediation estimands.