The perceived impacts of tourism on host communities, and associated resident attitudes toward tourism, continues to be an important issue. This much is indicated by governmental policy pronouncements like the World Tourism Organization Bali and Manila declarations (WTO, 1996 and WTO, 1997) and coverage in academic publications ( Akis; Boissevain; Faulkner; Haralambopoulos; Huang; Kang; Lindberg; Schneider; Schroeder; Wall and Wall). Pearce, Moscardo and Ross (1996) provide a thorough review and evaluation of previous research in this field.

Within this growing literature on tourism’s impacts, the primary focus has been on measuring attitudes and evaluating their relationship to perceived impacts. An alternate approach is to measure impacts in a metric that is common across the conventional impact categories: economic, sociocultural, and environmental/ecological. Measurement of impacts in a common metric enables identification of resident preferences and tradeoffs regarding these impacts and facilitates evaluation of the desirability of specific tourism projects and/or overall development paths. One possible common metric is economic value, and Lindberg and Johnson (1997a) used contingent valuation (CV) to estimate the economic values of selected impacts associated with tourism. A related technique is choice modeling (CM), which has grown out of conjoint analysis and has been used primarily for market research and evaluation, in tourism and other fields ( Dellaert; Dellaert; Haider; Jeng; Louviere and Morley).

Though from a different heritage, CM parallels, but is more general than, contingent valuation. The former allows for tradeoffs between multiple attributes, while the latter involves tradeoffs between only two attributes, one of which typically is money and the other of which typically is a detailed policy option. In simplified terms, dichotomous choice CV asks respondents to decide whether they are willing to trade a specified amount of money (a variable level of one attribute) for a specified good or service (a fixed level of a second attribute, such as a specific reduction in traffic within their community). Choice modeling asks respondents to decide which of several alternatives they would prefer, with the alternatives being packages of attributes at varying levels and one “base” or “other” alternative being none of the packages (levels fixed to zero for all attributes). If respondents choose a certain alternative, then they are assumed to prefer the levels of attributes in that package over the levels of attributes in the other ones and over their current situation of having none of the packages. The attributes used in the resident survey reported here were number of new jobs, amount of tax reduction, percent increase in rubbish, and number of additional cars on the road. For each of the eight choices presented, respondents were asked to choose between two alternatives, with one being a package of attributes and levels and the other being not to accept the package. Responses are assumed to reflect resident preferences and willingness to accept the negative attributes (rubbish and cars) in exchange for the positive attributes (jobs and reduced taxes).

CV and CM are relatively new techniques, and their validity and reliability remain subject to debate. Many of the applications and evaluations of these methods relate to their performance in environmental/recreational valuation and consumer choice, respectively, such that their performance in the present context, or similar ones, remains an area for research and evaluation. In a review of CV, CM, and other stated preference approaches, Morrison, Blamey, Bennett and Louviere (1996) note that there are theoretical and/or empirical indications of several biases associated with CV, including embedding effects and the related part-whole bias, hypothetical bias, payment vehicle bias, strategic bias, starting point bias, information bias, and non-response bias.

Some of these potential biases, such as hypothetical, information, and nonresponse bias, are inherent in any stated preference approach, but can be avoided or minimized through careful survey design and administration. Other potential biases, such as embedding, part-whole, payment vehicle, and strategic bias, may be more likely to occur in CV studies than in CM studies. For example, Lindberg, Johnson and Berrens (1997) were not able to reject the presence of part-whole bias in their CV study of resident tradeoffs, while Morrison et al (1996) note that CM’s more explicit attention to differences in attribute levels may make it less prone to this form of bias. Boxall, Adamowicz, Swait, Williams and Louviere (1996) provide further comparison of these two approaches. More generally, the benefits of the CV approach are that it directly estimates economic value for input into cost-benefit analysis, statistical estimation is relatively easy, and distributional effects can be estimated easily using predicted values. The benefits of the CM approach are that the scenarios accommodate multiple attributes (though respondents may focus on only a subset of these attributes) and often are more realistic to respondents than are CV scenarios. Relative to the latter scenarios, CM ones also involve less focus on financial gains or losses; depending on one’s view of actual and ideal consumer behavior, this may be a benefit or a drawback of CM.

Both CV and CM fundamentally differ from traditional social impact approaches (Lindberg and Johnson 1997a). The focus of CV and CM is on the tradeoffs that residents are willing to make between the positive and negative impacts of tourism. The measurement of these tradeoffs enables analysts to determine whether the impacts associated with a specific tourism development option (such as a project or policy) will add to or detract from resident economic welfare (used in this article in the sense of satisfying Pareto compensation criteria) on the basis of implementation leading to a net gain in aggregate economic value for residents. The information generated by applying these techniques can be used to inform, and to reduce uncertainty in, the decision-making process relating to tourism development. An idealistic and simplistic model would postulate that politicians and other public decision makers act in the aggregate interest of citizens. However, even if one assumes this to be true, it would require that decision makers accurately perceive the disparate citizen interests and then weigh them across individuals. Moreover, the postulate that decision makers act solely in the aggregate interest of their constituents (and not at all in their own personal interest) has been rejected as untenable (Elster and Williamson).

The direct measurement of citizen preferences using CV and CM approaches sidesteps the problems of personal interests and probable misperceptions on the part of decision makers, and these approaches follow in the tradition of public referenda. However, these particular approaches do require assumptions regarding decision rules and weighting systems. The common decision rule is to choose the project or path (or no change option) that maximizes aggregate welfare, as measured by willingness-to-pay or willingness-to-accept. The welfare of each citizen is commonly weighed equally, though other weighting systems can be used, particularly if there is concern that a willingness-to-pay approach will bias results in favor of high-income citizens, who are most able to pay. An alternate decision rule for CM is to choose a project or path with a probability of resident support at least equal to 50% or, in the case of multiple projects/paths, to choose the one with the highest probability of resident support.

The outcomes of respondent choices in the CM approach typically are described using random utility models. These models assume that consumer choices can be represented as a process in which the attributes of alternatives relevant to a given choice are evaluated in terms of the utility they provide the consumer. The part-worth utilities associated with each of the attributes are assumed to be cognitively integrated into an overall utility for each alternative, after which the alternative with the highest overall utility is selected. The utility function consists of two basic parts: a deterministic component that describes the structural utility that the consumer derives from the alternative, and a random component that describes the error over the structural utility. This error can be due to various sources such as measurement error, omitted explanatory variables, and unobserved variations in taste (Ben-Akiva, M. and Lerman, S., 1985. . Discrete Choice Analysis: Theory and Application to Travel Demand MIT Press, Cambridge MA.Ben-Akiva and Lerman 1985).