Three comments on the RIR method

The critique was initially made through an example trial for which the empirical event rate is zero in the treatment arm, and Walter et al. argued that a shortcoming of the fragility index is that this event rate is not incorporated. (Since outcome modifications by default occur in the treatment arm, these modifications thus have probability zero of occurring!) Frank et al. reanalyze the example with the RIR method and frame the RIR method as addressing the critique. [...]a Bayesian view of the RIR method suggests that some nonstandard choices were made. The posterior expectation of the Table is calculated by taking the posterior expectation of each entry and noting that E[YT∣XT,XC]=kXC+0.5nC+1≈kXCnC. Since this posterior expectation is sometimes fractional and whole numbers are required for the table, Frank et al. sometimes implicitly round it down (as in their Section 3.1) and sometimes implicitly round it up (as in their Section 3.2).