Simulating the probability of False Confession-Wrongful Conviction conditional on features of police interrogation: A response to

Arguing that previous research on false confession wrongful conviction (FCWC) has consistently suffered from outcome-based selection bias (i.e., collider bias), Mourtgos and Adams (2026) pose the question: how much do currently legal interrogation tactics contribute to FCWC when one accounts for non-FCWC cases? Their conclusion is that about 1% of FCWCs are associated with lawful interrogation tactics. Initially, we replicate and recover the original estimates using an independently reconstructed dataset. We present an in-depth treatment of the posterior probability distributions that emphasizes the uncertainty of simulated risk estimates. We present alternative model specifications such that Bayes’ theorem can account for substantive differences between FCWC and non-FCWC interrogations. Our intentionally conservative simulations highlight that the risk of FCWC is contingent on the attributes of the interrogation, with typical routine interrogation (1 to 2 hours) closely conforming to this 1% estimate, while the risk of FCWC under protracted interrogation quickly exceeds thresholds that merit concern—reaching 10% at around 6 hours. This replication and extension are intended to serve as just one illustrative example of why we must ensure that the substantive differences between routine and atypical interrogation are not disregarded when assessing risk. Bayesian simulations are a powerful tool for studying phenomena with limited data. However, these simulations must accommodate a variety of biases before we can draw conclusions about the risk of interrogation-induced FCWC. Conclusions drawn from this work are too premature to enter into policy and practice—whether the interrogation room or the courtroom.