We consider the long-run odds that narcotics users remain abstinent after m
ethadone treatment. A flexible split-hazard specification that allows for i
ndividual-level differences in both the long-run probability of eventual re
lapse and the short-run timing of relapse is developed. The model is applie
d to a comprehensive data set involving individual drug abuse and treatment
histories for over 800 addicts. Our findings indicate (1) that the short-r
un success of methadone programs does not automatically translate into long
-run abstinence, which suggests the need for aftercare, (2) the value of pr
eventing a teenager or young adult from initiating, and (3) the possibility
of identifying high-risk groups, both in terms of age of first daily use a
nd in terms of ethnicity.