pairs of discordant twins, can be attributed to non-familial sources, such as individual-specific environments and causal processes (Figure 2). Only those pairs of twins where one twin uses cannabis but the other does not are selected. Within these discordant pairs, if the MZ twin who uses cannabis is also more likely to report use of hard drugs, relative to their identical co-twin who does not use cannabis (i.e. a significant odds-ratio when comparing the risk within the pair of twins), then even after accounting for shared genetic and familial influences, cannabis use is associated with use of hard drugs. The mechanisms that contribute to this residual association are person-specific (i.e. experienced by the cannabis-using twin but not their co-twin) and might include socio-environmental factors (affiliations with drug-using peers or exposure to adverse neighborhoods) or pharmacological (e.g. receptor cross-sensitization) and epigenetic modifications, which may act via putatively causal or non-causal pathways. Therefore, such a residual association in pairs of MZ twins is necessary but not sufficient evidence in favor of causation. On the other hand, absence of such a residual association provides strong support for the lack of causal inference. In addition, the pattern of association in discordant twins, when compared
