We thank Large & Singh for their comments. But we would point out that we did not examine positive predictive value, as they say we did. We described the proportion of those classified as high risk who then acted violently. The two are only the same if an ascription of high risk, whether made using a structured risk assessment instrument (SRAI) or arrived at through clinical judgement, is treated as a ‘prediction’. Studies of the predictive validity of risk instruments out of necessity handle the data in this way ^{Reference Fazel, Singh, Doll and Grann1} and usually conclude that SRAIs demonstrate a moderate level of accuracy. As those who design SRAIs and others have repeatedly pointed out, however, fallible predictions are of limited value to clinicians. ^{Reference Otto and Douglas2} One thing that should help those clinicians is knowing what a classification of high risk means and, in particular, whether it means the same thing in different settings. We found that after controlling for time at risk, the rate of violence in groups classified as high risk using SRAIs shows substantial variation.