acterize the stages of risk analysis (Rowe 1977). The general idea is that one must have an initial identification of the problem that the risk analysis is expected to address (i.e., risk determination). With a well-defined problem, techniques derived from the natural sciences and engineering are deployed in a process of risk evaluation (measuring and combining the magnitude of the hazard and the likelihood of its occurrence). Finally a decision maker makes a decision (i.e., risk acceptance).
When risk analysis is used to support policy or contentious decisions, the simple three-stage characterization of risk analysis itself becomes controversial. Some have preferred to call the first stage hazard identification, implying that a hazard does not become a risk until the probability that it might occur has been measured. Some object to the normative connotations of the word evaluation and substitute the term risk measurement or risk assessment for the middle stage. Many have noted that risks may not be simply “accepted,” so the third stage is then relabeled as “risk management.” Because risk analysis of transgenic organisms is itself contentious, the committee chose to adopt the terminology hazard identification, risk assessment, and risk management when discussing risk in its decision support role.
Broadly conceived, the techniques of risk assessment are of five general kinds:
Epidemiological analysis. Events of interest are observed, and the statistical relations of these events in the sampled populations are analyzed. This epidemiological approach has been very effective in identifying disease risks among populations such as smokers, industrial workers exposed to certain substances, and persons with a specific genotype. The scientific rationale for this method is that empirical correlations provide a basis for predicting effects and may indicate cause. This method could be used to associate risks with particular transgenic plants that are intensively planted in large or specific areas.
Theoretical models. A theoretical model that mimics or simulates the causal interaction of elements in a complex system is used to identify likely sources of system failure. This approach is widely used to study the risk of failure in engineering contexts such as instrumentation and control design. It has also been applied in biology to develop strategies for ecosystem management. The scientific rationale for this method is that it provides the logical consequences of a set of scientific assumptions about risks. It has played a critical role in analysis of the risk of evolution of resistance to transgenic insecticidal plants (Alstad and Andow 1995, Roush 1997, Gould 1998), and could play a role in evaluating community-level non-target effects of transgenic plants (Andow 1994).
Experimental studies. Controlled experiments are conducted to iden-