mand for an illegal drug (Saffer and Chaloupka, 1995; Chaloupka et al., 1998; DeSimone, 1998).

  1. Lack of quantity data. Data on the quantities of drugs that consumers buy do not exist. To be most useful, a dataset should give quantities purchased in individual transactions, but even city-level aggregate consumption data might be useful.6 Because quantity data are not available, existing demand models use proxies. One common proxy is participation; this is a binary indicator (or a yes/no measure) of whether an individual has used a specified drug in a specified time period such as the past 30 days. Another proxy is the frequency of use (the number of times that an individual has used a drug in a specified time period). Some studies use both proxies—for example, Grossman et al. (1996) and Chaloupka et al. (1998) —estimate models of participation and of frequency of use conditional on participation. The accuracy of participation and frequency of use as proxies for quantity consumed is unknown (see Chapter 3 for further discussion of quantity data).

  2. Addiction. The utility that a consumer obtains from current consumption of an addictive drug depends on his past consumption (Becker and Murphy, 1988). Therefore, in a demand model for an addictive drug, current consumption depends on past consumption in addition to the price. If the consumer is foresighted, then current consumption also depends on future consumption (Becker et al., 1994). The dependence of current consumption on past and (possibly) future consumption increases the difficulty of obtaining suitable consumption data. Specifically, longitudinal consumption data measuring the quantity consumed over time are needed.7 The Monitoring the Future (MTF) survey provides longitudinal data on participation and use frequency by youths. Grossman et al. (1996) used these data to estimate a demand function for cocaine that takes account of the effects of addiction. No other consumption study reviewed by the committee was longitudinal. Instead, studies employed a


Aggregate consumption in the U.S. as a whole has been estimated by combining estimates of numbers of consumers, expenditure estimates obtained from arrested consumers interviewed by DUF (the predecessor of ADAM), and price estimates obtained from STRIDE (Office of National Drug Control Policy, 1997). The resulting consumption estimates are not available for individual cities and, in any case, are probably too crude for use in estimating demand functions.


Becker et al. (1994) estimated a model of the demand for cigarettes by using state-level aggregate consumption data (a time-series of cross-sections) instead of panel data. They assumed that state-level aggregates can be treated as consumption by a representative consumer. Becker et al. (1994) did not discuss the accuracy of this assumption. The assumption is problematic for cocaine, because cocaine consumers are highly heterogeneous. The representative consumer assumption cannot account for differences between the consumption patterns of casual and heavy users of cocaine.

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement