she will seek out future opportunities to use cocaine, and whether ultimately he or she requires treatment for cocaine dependence.
Despite the many investigations that have shed light on the suspected risk factors, there remains much to be learned about causal mechanisms determining use, dependence and addiction. Difficult methodological and data-related hurdles confront efforts to draw strong inferences about the individual-level and social-level circumstances, conditions, and processes that determine use.
For social scientists and policy analysts, the problem is especially complicated (see Manski, 2000 and Musto, 1995). Even if credible empirical evidence for social interactions should emerge, such evidence is likely to leave open basic questions about the processes at work. Does the stigma associated with drug use fall as the prevalence of use in the peer group rises? Do youth learn about the attractiveness of drug use by observing it in their environment (Feldman, 1968)? Manski (2000:130) illustrates this problem as follows:
To see the importance of understanding endogenous interactions at a deeper level, consider the crack cocaine epidemic of the 1980s, which appears to have subsided during the 1990s. A plausible explanation of the course of the epidemic begins with positive expectations interactions as youth of the ‘80s may have observed some of their peers initiate crack usage and apparently enjoy it. There also may have been positive preference interactions of the stigma-reducing type. Eventually, however, youth of the ’90s may have observed the devastating long-term outcomes experienced by addicts of the ’80s, and subsequently may have chosen not to initiate crack use themselves. If this story of observational learning is correct, then an information campaign warning of the devastating effects of crack addiction might have been effective in the early stages of the epidemic, but superfluous later on.
Without better data, researchers will continue to be unable to evaluate risk factors associated with intensification, abuse, addiction, desistance, and relapse. In particular, there are no consumption data and no longitudinal data available to the public. Lacking data on quantity, existing studies generally focus on prevalence of use within a specified time period. Without longitudinal data, the problem of causal interpretation of risk factors is likely to remain unresolved.
Concerns about the consequences of drug use for users and nonusers are central to U.S. drug policy, as manifest by drug prevention curricula, public service advertisements, and public statements by government officials. Yet much greater attention is given to statistical patterns of drug