In the context of chemical risk assessment, uncertainty has typically been defined narrowly. For example, Science and Decisions (NRC, 2009) defines uncertainty as a “[l]ack or incompleteness of information” (p. 97). It defines variability as the “true differences in attributes due to heterogeneity or diversity” (p. 97), and, as with previous reports (NRC, 1983, 1994), it does not consider variability and heterogeneity to be specific types of uncertainty.
In other settings, such as research on climate change, variability is considered a type or nature of uncertainty (CCSP, 2009). In addition, even reports related to chemical risk assessment highlight the importance of evaluating both uncertainty and variability in risk assessments and considering both in the decision-making process (NRC, 1983, 1994, 2009). When EPA makes a regulatory decision it must consider what information it might be missing, the variability and heterogeneity in the information that it has, and the uncertainty in that variability and heterogeneity. The committee, therefore, discusses variability and heterogeneity in this report. When using the term uncertainty generically, it includes variability and heterogeneity in its definition.
Variability occurs within a probability distribution that is known or can be ascertained. It can often be quantified with standard statistical techniques, although it may be necessary to collect additional data. If variability comes, in part, from heterogeneity, populations can be divided into subcategories on the basis of demographic, economic, or geographic characteristics with associated percentages and possibly a probability distribution over the percentages if they are uncertain. That stratification into distinct categories can tighten the probability distribution within each of the categories. Variability in the underlying parameters often depends on personal characteristics, geographic location, or other factors, and there might not be an adequate sample size to detect true underlying differences in populations or to ensure that data are sufficiently representative of the population being studied.
There are many variables outside of the decision maker’s control that can affect the appropriateness of a particular decision or its consequences (for example, socioeconomic factors or comorbidities). Modeling those factors is not always feasible. For example, there may be too many socioeconomic factors that require large samples to analyze, insufficient time to conduct an appropriate statistical survey and analysis, or a prohibition of using some sociodemographic variables in the analysis. A longer-term research agenda can often evaluate such variables, and retrospective