In this commentary, the author focuses on the use of design elements for increasing the severity of causal mediation tests. The estimation of causal mediation effects from observational data rests on rather stringent assumptions. In introducing and exemplifying ratio-of-mediator-probability weighting (RMPW), Hong and Nomi (henceforth HN) make these assumptions very explicit and rather transparent. However, the crucial question in practice is whether these assumptions are actually met or whether they are violated to an extent that would invalidate the conclusions drawn. Of course, HN are aware of the possibility of unmet assumptions and suggest sensitivity analyses. Although sensitivity analyses are very helpful for probing the estimated effects' sensitivity to unobserved confounders, they can neither entirely rule out potential threats to validity (like differential selection, attrition, instrumentation, or history effects) nor show that a threat to validity might actually be present. Thus, in addition to sensitivity analyses, one should always think about design elements that help in ruling out plausible threats to internal validity. The author's discussion is not restricted to marginal mean weighting through stratification and RMPW--it also applies to other techniques for estimating total, direct, or indirect effects.