There are a number of misunderstandings put forward by Kostic et al that we should clarify. First, it is important to correct the authors with regard to their understanding of the issue of confounding: a confound is a variable of relevance in epidemiological models of causation. To be clear, we did not suggest in our report that psychotic symptoms somehow cause psychiatric disorder. Symptoms and signs of course cannot cause pathology; rather, they act as clinical risk markers for disease. Using an analogy from respiratory medicine, the authors' suggestion that we should control for substance misuse (which is a potential cause of psychotic symptoms) makes no more sense than suggesting that respiratory researchers should control for cigarette smoking when looking at haemoptysis as a risk marker for lung pathology. That is, haemoptysis alerts the clinician to the likely presence of pathology (i.e. it is a risk marker); the cause of the pathology remains to be determined. Similarly, we showed that psychotic symptoms act as risk markers for a broader range of psychopathology than has generally been recognised (and, in particular, for multimorbid psychopathology). In the same way that there are multiple mechanistic causes for the occurrence of haemoptysis in lung pathology (e.g. cigarette smoking, infection, trauma), there are also likely multiple mechanistic causes for the occurrence of psychotic symptoms in psychopathology. In this regard, we would direct the authors to paragraph three of the Discussion, in which we put forward a number of suggestions for such causes.

Kostic and colleagues also wonder whether the response rate in study 1 or the fact that study 4 specifically overselected for psychopathology may have affected the validity of these findings. Unfortunately, we do not have space to provide a comprehensive explanation of the epidemiological impact of response rates on findings; however, it is important to clarify that, although response rates can introduce bias with regard to reported incidences or prevalences, they usually have little effect on statistical measures of association. With regard to study 4, which purposely overselected for psychopathology, this is, in fact, the very methodological basis of a case–control study. A statistical weight must be applied to determine population prevalences from such an approach but, as evidenced by the many thousands of case–control studies in the medical literature, this does not create problems for identifying associations that can be generalised to the population. Quite aside from this, we would remind the authors that the best way to address the possibility that sampling and other biases are responsible for a set of results is independent replication; our findings were replicated across multiple independent studies, led by multiple independent teams in multiple independent centres. With regard to symptom inclusion, in accordance with the guidelines of the interview instrument (the Schedule for Affective Disorders and Schizophrenia for School-Aged Children), ^{Reference Kaufman, Birmaher, Brent, Rao and Ryan1} hypnopompic, hypnagogic and drug-induced hallucinations were excluded, as were symptoms experienced only in the context of febrile illness.

Last, Kostic and colleagues state that there was no mention of the potential role of ‘school and family problems’ in our findings, although we specifically suggested this as an important issue in our discussion. In fact, we have already published results from study 4 (in this journal, in fact) on the relationship betweenpsychotic symptoms and a number of measures of school and family problems, including bullying, interparental domestic violence and physical and sexual abuse. ^{Reference Kelleher, Harley, Lynch, Arseneault, Fitzpatrick and Cannon2} We cited this in the paper. Furthermore, Kostic et al will be glad to know that a report on the relationship between childhood trauma and psychotic symptoms in another of the samples (study 2) is currently under review (details available from the authors on request). However, it is important to recognise that, again, the authors are raising an issue of causality in the relationship between psychotic symptoms and psychopathology; the point of the current paper, on the other hand, was to highlight new developments in our understanding of the importance of psychotic symptoms as clinical risk markers for psychopathology.

We appreciate that Kostic and colleagues are certainly not the only individuals who may have had conceptual misunderstandings about the above epidemiological points and we thank them for the opportunity to clarify some of these issues for the benefit of other readers with similar questions. We are also pleased to find that the Journal's readers are actively discussing the importance of assessing psychotic symptoms in the context of non-psychotic psychopathology. As well as recognising that psychotic symptoms are risk markers for a range of non-psychotic Axis I disorders in general, and for multimorbidity in particular, ^{Reference Kelleher, Keeley, Corcoran, Lynch, Fitzpatrick and Devlin3} we would also especially encourage discussion about findings on the importance of these symptoms as risk markers for suicidal behaviour in young people with psychopathology. ^{Reference Kelleher, Lynch, Harley, Molloy, Roddy and Fitzpatrick4} Considering the serious implications of these findings, an improved awareness of the significance of these symptoms among clinicians is urgently needed.