Torrey and Yolken raise two concerns about our study (Solmi et al. Reference Solmi, Hayes, Lewis and Kirkbride2017). The first relates to the use of psychotic experiences, as opposed to diagnosed schizophrenia, as the outcome for our analyses. They suggested that this approach could be over-inclusive and result in a dilution of the association between the exposure and the outcome. We agree with Torrey and Yolken that psychotic experiences and schizophrenia are different outcomes and we acknowledged this issue in our study as a potential limitation. Adolescents who report psychotic experiences are, nevertheless, at increased risk of developing psychosis and other psychopathology later in life (Fisher et al. Reference Fisher, Caspi, Poulton, Meier, Houts, Harrington, Arseneault and Moffitt2013). In this cohort, they have also been shown to share a number of features and risk factors that are typically observed in individuals with schizophrenia [e.g. lower IQ (Horwood et al. Reference Horwood, Salvi, Thomas, Duffy, Gunnell, Hollis, Lewis, Menezes, Thompson, Wolke, Zammit and Harrison2008); impaired social cognition (Sullivan et al. Reference Sullivan, Thompson, Kounali, Lewis and Zammit2017); higher residential mobility (Singh et al. Reference Singh, Winsper, Wolke and Bryson2014) and neighbourhood deprivation (Solmi et al. Reference Solmi, Colman, Weeks, Lewis and Kirkbridein press)]. We therefore hypothesised that, if a link between cat ownership and schizophrenia existed, we would have found this association with broadly-defined phenotypical presentations of psychosis, as we do with these other risk factors.

The second concern related to our choice of confounders. Torrey and Yolken argue that the statistically significant association which we reported between cat ownership in childhood and psychotic experiences at age 13 years in univariable analyses was artificially removed by inappropriate adjustment for variables (i.e. indicators of socioeconomic status) which lie on the causal pathway between our exposure (cat ownership) and our outcome (psychotic experiences). Adjustment for confounding is necessary to take account of alternative potential explanations for an observed association. We do not believe socioeconomic status lies on the causal pathway between cat ownership and psychotic experiences, since that would imply that owning a cat leads to lower socioeconomic status, which, in turn, causes psychotic experiences. Different socioeconomic backgrounds might, however, be associated with different patterns of cat ownership, and are also known to be associated with psychosis [with higher rates of the latter in more deprived settings (Morgan et al. Reference Morgan, Kirkbride, Hutchinson, Craig, Morgan, Dazzan, Boydell, Doody, Jones, Murray, Leff and Fearon2008)], thus providing theoretical support for our use of socioeconomic status as a confounder (our full causal model is presented in online Supplementary Figs S1 and S2). Torrey and Yolken's proposal that low socioeconomic status might lead to increased exposure to Toxoplasma gondii in faeces compared with those of higher socioeconomic status implies effect modification. We thank Torrey and Yolken for their interest in our paper, and hope that our reply clarifies these issues.