Jainer & Shivanandaswamy's comments about the problems of bias in case–control studies are well made. However, our study ^{Reference Mason, Rimmer, Richman, Garg and Johnson1} was designed to avoid such problems by recruiting all patients with a likely diagnosis of schizophrenia in contact with general practitioners in a defined catchment area. There was no possibility of influencing the selection of individuals since they were all patients with a diagnosis of schizophrenia on a community mental health team's case-load.

The community mental health team concerned looked after an area of high socio-economic deprivation and the study included patients who had drifted down the social scale from more affluent rural areas where one would expect a lower prevalence of middle-ear disease. If there is any bias in this study it is likely to favour the null hypothesis rather than that suggested by Jainer & Shivanandaswamy.

In addition, perhaps the most striking finding in this study was the excess of left-sided middle-ear disease. In this case, the odds ratio of 4.15 meets the recommendation of Sackett et al ^{Reference Sackett, Straus, Richardson, Rosenberg and Haynes2} that an odds ratio of greater than 4 should be used to establish an association in case–control studies.