aggregate model to explain the behaviors of needle exchange users and nonusers without giving appropriate attention to the potentially biasing effect of outliers in their data set.

Another design concern relates to the investigators' assumption that needle exchange nonusers are not gaining access to needle exchange needles. If a large number of nonusers is obtaining sterile needles that have originated from the exchange programs, it could be argued that they are indirectly participating in the needle exchange program or at least benefiting from it.

The method used to verify the stability of coefficients also poses some difficulties. The paper indicates that the final equations that use the data from all cohorts (n = 728) also incorporate the 1988 and 1992 cohorts (n = 405) on which the original coefficients were derived. This nonindependence of samples raises questions about the ability of this approach to reliably assess the stability of coefficient estimates across samples (especially because more than half of the final sample were members of the original sample).

It is inappropriate for the investigators to apply the circulation theory concept used in the New Haven evaluation study. The Chicago program is not a one-for-one program, and the protocol it follows allows for the number of distributed needles to exceed the number of returned needles by 5 on a per-visit basis. Kaplan's circulation theory requires that needles be exchanged on a one-for-one basis; as derived in Kaplan's work, it is this law of conservation of needles that provides the link between needle exchange rates and the level of infection in needles. In the absence of a one-for-one exchange, there is no physical guarantee that needle circulation times will decline as the numbers of needles distributed increases, because there is a net increase in the total number of needles in circulation.

In sum, based on the panel's interpretation, we cannot say that the data justify the conclusions the researchers have reached.

In a second paper, the same investigators (O'Brien et al., 1995b) explore the potential effect of needle exchange programs on HIV risk behaviors and incidence. Data from the 1994 interview and HIV serology test results of current injection drug users from all three cohorts (n = 728; 1988, 1992, and 1994 data), as well as data from the 1993 follow-up of the 1988 and 1992 cohorts (n = 405) described above, form the basis of the reported analyses.

Four measures of injection risk behaviors were derived. A dichotomous variable indicating overall risky injecting behavior based on injection drug users' reported use of others' used needles, consistent use of bleach,