The Editor is of course right to highlight the potential importance of the work by Morrison and colleagues (Reference Morrison, French and Walford2004) and how this must be weighed against possible methodological flaws. While the authors acknowledge most of these, there are some aspects of the study which deserve further clarification. For example, the exclusion of two cases after randomisation to the cognitive therapy group owing to the fact that they had apparently been psychotic at inception could be justified. It is stated that ‘all other participants were questioned about this possibility’ - however, can we be sure that psychosis at that earliest stage was rooted out equally assiduously in both groups, cognitive therapy and ‘control/monitoring’?

Another matter discussed is the randomisation procedure, which resulted in unequal group allocation. The authors state boldly that this was due to chance, and the methodology for randomisation (stratified for gender) as described seems to be sound. However, I am not sure I would be happy to accept a reprieve from a gloomy fate on the basis of ‘tails’ on the toss of a coin, in the knowledge that it had previously yielded ‘heads’ 37 times out of the last 60. Let's just imagine that somehow a gremlin interfered with the randomisation process so that the patients who seemed less likely to decompensate, the majority, were steered into the therapy group. This would produce the observed pattern. This gremlin need not even be credited with much clinical foresight since prediction of onset of psychosis in a very high-risk group of 20- to 21-year-olds is quite simple given one of the most robust findings in the epidemiology of schizophrenia, namely the later age of onset in females. So, as long as more females find their way into the intervention group, a better short-term outcome is virtually assured. Morrison et al ended up with 40% females in the cogntive therapy group v. 17% in the control group. It may all be due to chance and adjustable in the logistic regression analysis, but given the impossibility of delivering a psychological intervention blindly, the integrity of the randomisation procedure must be beyond question.