The National Research Council report Science and Judgment in Risk Assessment (NRC 1994a) addressed the extensive uncertainty and variability associated with estimating risk and concluded that risk characterizations should not be reduced to a single number or even to a range of numbers intended to portray uncertainty. Instead, the report recommended, risk managers should be given risk characterizations that are both qualitative and quantitative and both verbal and mathematical. The Commission concurs that qualitative descriptions of risk-related uncertainty are needed, but it does not agree that formal, quantitative uncertainty analyses are either necessary or useful for most risk assessments. When the Commission's risk-management framework is implemented, nonquantitative methods of communicating information about uncertainty to participants are likely to be more effective than quantitative methods. There are, of course, situations in which quantitative uncertainty analyses are likely to provide information that is useful in a decision-making process, and the Commission encourages the continued development and application of quantitative methods. There are also likely to be situations in which a quantitative uncertainty analysis can be used to improve qualitative information about uncertainty, even if the quantitative information is not what is communicated to the risk manager.

FINDING 3.3: The best way to present the results of a risk assessment so as to acknowledge variability and uncertainty is controversial. There is also confusion regarding the differences between variability and uncertainty. Variability comprises a population's natural heterogeneity or diversity, and it does not change through further measurement or study, although better sampling can improve knowledge about variability. Uncertainty reflects gaps in information about scientifically observable phenomena. Uncertainty sometimes can be reduced through further measurement or study. Several quantitative methods to describe risk-assessment uncertainties are being explored. Although there is general agreement as to the value of qualitative statements describing critical uncertainties in health risk assessments, formal quantitative approaches to uncertainty analysis are complex, difficult to perform, difficult to understand, and often unnecessary. Variability, in contrast, can be described much more readily and can be based on actual measurements.

RECOMMENDATION: Qualitative descriptions of the primary sources of uncertainty and the weight of the evidence associated with exposure, toxicity, and susceptibility should be included in risk characterizations intended for risk managers and the nontechnical public. Quantitative methods of describing the variability associated with exposure can yield useful information for risk management and should be included with qualitative descriptions in risk characterizations (see sections 3.2 and 5.1). However, a formal quantitative analysis of the uncertainties in risk estimates is not needed for most risk assessments.

RATIONALE

Support for routine, formal quantitative analysis of uncertainty is based on the desire to move away from poorly supported default assumptions and point estimates of risk that convey an unwarranted sense of accuracy and that fail to convey any sense of the confidence that the risk assessor has in the estimates or their inherent complexity. Providing a numerical range of possible risks that reflects uncertainty and variability is thought to allow more-informed and more-transparent decisions than are possible when only a single point estimate of risk is generated.

In the absence of some explanation of the weight of scientific evidence, communicating a range of population risks might be misconstrued by those unfamiliar with quantitative methods as implying that all the numbers in the range are equally likely or plausible and therefore equally valid for regulation. Many risk estimates are crude yardsticks for decision-making--as Thomas Gentile, of New York State's Division of Air Resources, noted in his testimony before the Commission, many state-level risk managers just want to know, "Is it safe or not?" In this context, the routine provision of a distribution or range of possible risks might only confuse and delay the regulatory process.

Generating ranges or probabilistic distributions of risk estimates instead of point estimates is thought to portray more accurately the range of possible risks experienced by an exposed population. When data are scarce, assumptions about the underlying shape of the risk distribution dominate; that is, when uncertainty is great, a range of probabilities based on assumptions would replace point estimates based on assumptions. As Thomas Starr, of ENVIRON, testified before the Commission, formal uncertainty analyses are not useful if there are disagreements about the underlying shapes of the distributions; folding assumptions about those shapes into a risk assessment incorporates the assessor's bias into the risk estimate. Approximating uncertainty thus introduces yet another source of uncertainty.

A report prepared by Cambridge Environmental Inc. for the Commission, Health Risk Assessments Prepared per the Risk Assessment Reforms under Consideration in the U.S. Congress (see appendix A.5 for abstract), showed that when chemicals that are not known human carcinogens are evaluated, most of the uncertainty in risk estimates results from uncertainty about a substance's toxicity. The probability distributions generated to account for that kind of uncertainty can take a variety of shapes, depending on the assumptions made and the data used--for example, whether a chemical that tested positive for carcinogenicity in a rodent bioassay is or is not a human carcinogen and whether some tumor rates were reduced, not increased. Methods for quantitatively describing the uncertainties associated with toxicity are still under development.

Providing distributions of risk is thought to counteract the perceived bias toward overestimating risk that is due to a compounding of conservative default assumptions. However, any range of a population's risks inevitably will include estimates in the upper end of the distribution that are at least as stringent as currently provided by point estimates. When confronted by an array of estimates, regulators and community groups are likely to choose from the more stringent portion of the range. Using formal uncertainty analysis is unlikely to lead to less-stringent regulation. If the risk-management process is perceived to be too stringent, the risk-management process, not the risk-assessment method, should be modified.