Whenever marginal prices are nonconstant, as in most hedonic and household production function markets, ordinary least squares estimates of the price elasticities of structural equations will be biased. A two-stage least squares estimation procedure is developed and applied to estimate hedonic (and potentially household production function) demand functions when price gradients are nonlinear.

The hedonic travel cost method is a technique which reveals how much users are willing to pay for the individual characteristics of outdoor recreation sites. The prices of recreation attributes are estimated by regressing travel costs on the bundles of characteristics associated with each of several potential destination sites. The demand for site characteristics on site quality is then revealed by comparing the site selection of users facing different attribute prices. The technique is applied to value steelhead fish density in Washington State streams.

Decision makers performing cost-benefit analysis must often deal with the problem of how to aggregate the benefits across states of nature accruing from an uncertain public investment project.' Option price and the expected value of consumer's surplus are two potential measures of these aggregate benefits.2 The expected value of surplus has been proposed because it is readily measured and because risk pooling (Paul Samuelson, William Vickrey; 1964) and risk spreading (Kenneth Arrow and Robert Lind, 1970) tend to encourage risk neutral behavior. Option price has been favored on the vague notion that people would be willing to pay something extra above expected surplus to preserve the opportunity to purchase a good (Burton Weisbrod, 1964). As Daniel Graham cogently argues in this Review (1981), however, option price is but one of an infinite number of contingent payment schemes. The literature has provided no justification for focusing upon it as an ideal measure of benefits under individual risk. Graham further argues that policymakers ought to adopt the compensating contingent payment plan which maximizes expected revenue. This maximum payment plan, by definition, is never less and will often exceed any other contingent payment scheme. Consequently, Graham argues that both the expected value of surplus and option price are underestimates of the true value of project benefits. Our purpose in this comment is twofold: first, we show the role of project and nonproject insurance in a model of individual risk; and, second, we argue that option price, not the maximum payment plan, is the optimal rule when no fair insurance is available. In Section II, we show that if fair insurance is available against all risks, all contingent payment plans yield identical revenue. A similar result holds if insurance is available for nonproject risks and the effect of the project to an individual is small (the Arrow-Lind model). In Section III, we explore the case where either the project has a large uninsurable effect on the individual, or there is no insurance against even nonproject risks. We argue that the very phenomena (moral hazard, adverse selection, and complexity) that eliminate the market for private insurance also prevent the government from making otherwise desirable contingent payments. If contingent payments are too costly, the government's only remaining choice is to collect payments that are constant across states, which makes option price the relevant measure of benefits.