In response to Dr Miller we would like to state some general principles, to clarify our methodology and provide some additional results. First, we have no argument with the truism that causality cannot be inferred from correlation. Dr Miller seems to overlook the fact that, despite widespread awareness of the dangers of determining causality, the terms ‘risk’ and ‘protective’ are commonly used to describe associations identified in longitudinal studies. Indeed, identifying and interpreting such associations is the primary reason for conducting cohort studies. The reiteration of standard caveats should not be necessary in every article arising from these studies and would make for very tedious reading indeed.

The potential for inadequate control of confounding by unmeasured or omitted confounding factors is always a possibility in any multivariate analysis. Researchers are inevitably constrained by the measures they have at their disposal which, in turn, result from the constraints of research directions, design, responder burden and so on. Dr Miller criticises us for omitting socio-demographic measures while including correlated behavioural measures. In terms of the former, we assessed the influence of both parental education and metropolitan residence on cannabis dependence but as there was no evidence of univariate associations for either measure they were unlikely to be confounders (parental education, reference group ‘some tertiary’: completed secondary school OR 0.8 (95% CI 0.5–1.3); incomplete secondary OR 1.0 (95% CI 0.6–1.6); school in metropolitan Melbourne: OR 1.0 (95% CI 0.6–1.5)). As they were uninformative, these findings were omitted from the article in the interests of parsimony and conserving space. As the report focused on adolescent behavioural and mental health predictors of cannabis dependence, both parental substance use and peer substance use, although likely to be predictors, were not considered relevant to the question. Indeed, they were omitted from the analysis as their inclusion could have masked the associations of interest, exactly as Dr Miller describes.

We acknowledge that confounding occurred between some of the explanatory measures included in the multivariate analysis. We illustrated and discussed in some detail the confounding that occurred between early-onset cannabis use, cigarette smoking and antisocial behaviour. Furthermore, the interaction between problematic alcohol use and weekly cannabis use to which Dr Miller objects arose as post hoc examination of confounding.

Mr Palmer misunderstands the denominator of the reported symptom prevalences: we described overall symptom prevalence in the 1601 participants. Symptom prevalences in participants classified as being cannabis dependent were reported in an earlier publication and were: tolerance 17%, withdrawal 74%, unintentioned use 84%, persistent desire 91%, excessive time spent obtaining, using or recovering from use 74%, social consequences of use 18% and continued use despite acknowledged health problems 63% (Reference Coffey, Carlin and DegenhardtCoffey et al, 2002). Furthermore, participants classified as dependent cannabis users reported compulsive and out-of-control use more frequently than those classified with dependent alcohol use. That there is gathering evidence of social, physical and mental health harm, including dependence, arising from long-term cannabis use is now beyond debate. For a brief and informative review of the current literature on this topic see Ashton (Reference Ashton2002).

Mr Palmer debates what really constitutes cannabis dependence. That young people ‘are smoking because they like it’ does not preclude the possibility that they may be dependent. Alternatively, they may be using it to stop feeling awful, in the self-medication paradigm. He quotes an assertion that other non-challenging behaviours performed persistently may also fit dependence criteria. This may be so, but the harm that arises from these activities is a moot point. The issue that concerns us, and that we used the current gold standard instrument in population research to identify, is that cannabis dependence inevitably prolongs heavy use. No measure applied at interview can be considered to be completely sensitive and specific for all the reasons that Mr Palmer states but the unreferenced assertion that the ‘phenomena [are] grossly overreported’ is unsupportable in the light of extensive developmental and confirmatory work performed in treatment and non-treatment settings (e.g. Reference Nelson, Rehm and UstonNelson et al, 1999). We do not consider it a problem that individuals can be classified as dependent with different combinations of symptoms – conversely, we need to increase our understanding of symptom combinations and their significance (Reference Nelson, Rehm and UstonNelson et al, 1999).

The final point that Mr Palmer makes is to query the validity of our measure of cannabis use. He appears to have misread the definition – we did not ask about ‘smokes’ at all. We asked participants how often they ‘used cannabis’ without specifying the method of delivery. We assume the word ‘used’ is unambiguous and involves ingestion in some manner.

Finally, we follow no political agenda but seek only to inform the general public and policy makers using sound epidemiological evidence resulting from good study design, careful analysis and cautious interpretation. Our article represents a step towards filling the evidence void in the current polarised debate about important public health and policy issues surrounding cannabis use (Reference Strang, Witton and HallStrang et al, 2000).