Economists, and more recently, courts and public officials, use benefit-cost analyses to make optimal policy choices. The basic approach is to determine the policy choice or value of some decision variable that yields the maximum net benefit. In theory, when marginal benefit and marginal cost are continuous functions of some scalable policy variable Q, net benefits are maximized where MB = MC, i.e. where the marginal net benefit is zero.

In the real world, however, the choice is often between discrete policy options. The usual decision process involves calculating a B-C ratio for each option and immediately ruling out options with B-C < 1. At this point, the optimal choice may be the option with the highest B-C ratio (if that option is scalable), or the option with the largest B-C difference (if the options are not scalable).

But B-C analyses are rarely that straightforward. First, benefits and costs may accrue in different time periods, requiring use of discounting, and the choice of discount rate can have a dramatic influence on B-C calculations. For example, US Army Corps of Engineers projects typically involve large up-front construction costs and benefit streams that extend decades into the future. These projects may look economical at low discount rates, but would not be justified at higher discount rates. The choice of discount rate can often determine the B-C rank orders of alternative policy choices.

Second, B-C analyses often need to account for future risk or uncertainty. In economic parlance, "risk" is quantitative while "uncertainty" is not. Risk implies calculable probabilities of alternative outcomes; uncertainty implies these probabilities are not really calculable. Longer time horizons involve higher levels of cumulative risk or uncertainty.

It is not always obvious how a benefit-cost analysis should be framed. For example, a group of environmentalists sued to block construction of a hydropower dam in Hell's Canyon on the Snake River, arguing that the benefits of the hydropower did not outweight the cost of the wilderness recreation amenities that would be lost. The utility company calculated the benefit of the project as the total value of the electricity it would generate minus the construction cost. The environmental groups calculated the benefit of the project as simply the net cost savings of the hydropower versus electricity from the next-cheapest source. The court accepted the latter analysis.

The court also considered an asymmetric risk inherent in the project. The dam could be built at any time in the future, but once built, the wilderness recreation amenities would be gone for good. Dam construction is irreversible, but wilderness preservation is not. Thus the potential cost of being wrong in choosing the hydropower project was higher than the potential cost of being wrong in choosing wilderness preservation.

Benefit-cost analyses often have to account for the economic values of non-market environmental amenities, or the economic costs of human morbidity and mortality, that may be affected by the policy decision. Economists have developed various methods for estimating economic valuations of environmental amenities. One class of methods gauges the effects of environmental quality changes on markets for related goods. Another class of methods uses surveys to elicit respondent willingness-to-pay or willingness-to-accept valuations for hypothetical environmental quality changes.

Costs of human morbidity and mortality risks are estimated by multiplying the likely quantity of people affected by some per-person valuation of health or life. The available toxicologic and/or epidemiologic data that are used to quantify the number of people to be affected are typically sparse and noisy. Toxicologists use high-dosage short-term animals experiments to estimate low-dosage long-term effects on humans. Epidemiologists often identify multiple collear risk factors, so that the risks of individual factors are difficult to estimate.

Not only are actual risks difficult to quantify, but public perceptions such risks are often highly distorted. People also resist placing any dollar value on health or life (although courts do it all the time). However, economic self-valuations are implicit in their behavior (smoking, not wearing seat belts). In theory, however, people can evaluate health and lives meaningfully under the random conditions of John Rawls's "veil of ignorance." When you recognize that you face the same small random risk as everyone else, you may be able to personalize that risk in a statistically appropriate way.

There are several less rigorous alternative to B-C analysis that relax the requirement that benefits and costs be compared on a money metric: A cost-effectiveness analysis involves choosing some specific policy objective and then determining the least-cost method of achieving it. The benefits are not quantified. We simply assume the correct objective is chosen. An impact analysis skips eschews any single metric for comparison of benefits and costs. An environmental impact analysis will typically just catalog the physical impacts, mitigation options, short-run vs. long-run trade-offs and irreversibilities associated with alternative policy choices, leaving it to the policy-makers to apply their own subjective weights to these factors in deciding which choice is best.