Like a magician pulling a rabbit from his hat, Turkington draws a positive result for cognitive therapy for schizophrenia from the literature – only for McKenna to put it back in again (Reference Turkington and McKennaTurkington/McKenna, 2003). Does it exist or not? McKenna's arguments and table look convincing as, by excluding any study that does not have an active control, he reduces the number of studies he considers. But would he do the same for studies of antipsychotic medications? Or does he assume that patients, and raters evaluating patients, can detect no difference between taking, for example, placebo and haloperidol, or even haloperidol and olanzapine? In which case why are we giving them so much of the latter?

But even focusing only on the studies that he finds acceptable, he dismisses one (SoCRATES; Reference Lewis, Terrier and HaddockLewis et al, 2002) for having a positive effect over active control on auditory hallucinations (oh, for a drug that had such an effect over and above those currently available!) and another (Reference Sensky, Turkington and KingdonSensky et al, 2000) where a differential benefit of cognitive–behavioral therapy over befriending only became apparent 9 months after therapy ended. He completely omits other widely cited studies with active placebos and positive effects (e.g. Reference Drury, Birchwood and CochraneDrury et al, 1996). He then does an unusual meta-analytic exercise in dismissing two small pilot studies by weighing them against each other and finding them to cancel out. Other meta-analyses (e.g. Reference Pilling, Bebbington and KuipersPilling et al, 2002) using more conventional methodology have concluded differently and, fortunately, so has the National Institute for Clinical Excellence.

The rabbit exists and is multiplying rapidly (e.g. Reference Durham, Guthrie and MortonDurham et al, 2003).