Second, unreliability – a form of measurement error – is confounded with unique environmental effects in twin models, and some diagnoses appear more reliable than others in unselected samples. For example, in studies of adults employing the Virginia Twin Registry, the reliabilities of lifetime diagnoses of panic, major depressive, and generalized anxiety disorders tended to be lower (κ = 0.34-0.56) than those of cocaine, sedative, cannabis, and stimulant use disorders (κ = 0.47-0.68) (Foley et al., 1998, Hettema et al., 2001b, Kendler et al., 2001, Kendler et al., 2000). Since no psychiatric diagnosis is made with 100% reliability, heritability estimates tend to be underestimates, but the heritabilities of the disease-like conditions above would be expected to be slightly more underestimated than those of the substance use disorders. Nevertheless, it is unlikely that correcting heritability estimates for unreliability would substantially reorder the results in Figure 1. In fact, the reliability of cocaine dependence (κ = 0.47) was slightly lower than that of panic disorder (κ = 0.56) in the studies cited above, yet the heritability estimates from these studies were 0.79 and 0.37, respectively, for these diagnoses.
