Introduction: John C. Whitehead, Timothy C. Haab and Ju-Chin Huang
Stated preference approaches use hypothetical data to estimate the ex ante willingness to pay for various commodities. For example, the contingent valuation method can be used to estimate economic values of environmental resources, including nonuse values – those environmental values that are enjoyed by consumers who do not use the environmental resource on-site. Most economists, however, are firmly rooted in the revealed preference paradigm to estimate the use values of environmental resources. Revealed preference approaches are based on actual choices and actual choices based on real costs and benefits better reflect environmental values. During the decade of the 1990s the “contingent valuation debate” dominated the environmental valuation literature. Economists developed competing damage estimates associated with the Exxon Valdez oil spill, argued about whether nonuse values exist and whether the contingent valuation method is able to accurately measure them. Critics of the contingent valuation method argued that hypothetical behavior is too inaccurate to use for policy analysis. Proponents argued that contingent valuation generated value estimates that are no less accurate than those developed with revealed preference nonmarket valuation approaches. Economists tend to consider revealed and stated preference approaches as substitutes when choosing valuation methods. There are problems with this attitude. Since the revealed preference approaches rely on historical data, evaluation of new policies is often limited. Oftentimes, stated preference methods are the only way to gain policy relevant information. But, stated preference approaches are based on hypothetical survey data and respondents can be placed in unfamiliar situations. Revealed preference data is firmly planted in reality. Since the strengths of the revealed preference approaches are also the weaknesses of the stated preference approaches and vice versa, revealed and stated preference methods should be considered complements. The combination and joint estimation of revealed and stated preference data seeks to exploit the contrasting strengths of the revealed and stated approaches while minimizing their weaknesses (Cameron 1992; Adamowicz et al. 1994). Joint estimation has addressed a wide range of important issues in nonmarket environmental
valu ation, including the hypothetical bias of contingent valuation and behavior data, the valuation of quality change that extends beyond the range of historical experience, and the development of new econometric and survey methods that specifically address data combination (Whitehead et al. 2008). Data combination can be used to mitigate a large number of problems. First, some revealed preference data are limited to analyzing a range of behavior in response to a limited range of market or environmental change. Stated preference surveys can be designed to collect data on hypothetical behavior, which allows estimation of behavior beyond the range of historical experience. Second, general population surveys can be used to survey users and nonusers of an environmental resource and analyze the decision to participate in the market. But these data are limited when trying to understand changes in participation. Combining revealed preference data with stated preference data from surveys of the general population can be used to understand changes in participation and the market size with new products or environmental changes. Third, there is often high correlation between independent variables in revealed preference data. Multicollinearity among characteristics leads to statistically insignificant coefficient estimates which make it impossible to estimate the value of changes in the variables of interest. A related problem is endogeneity. For example, recreational fishing catch rates are correlated with fishing experience and both variables are related to fishing trip frequency. Analysis of a policy to value an increase in catch rates with revealed preference data can be confounded by fishing experience. An alternative strategy is to combine revealed preference data with stated preference data. Stated preference surveys can be designed to mitigate multicollinearity and endogeneity. Fourth, revealed preference data collection is relative inefficient. Oftentimes, a revealed preference cross-section survey will collect only one data point. Stated preference surveys can be designed to gather pseudo-panel data, supplementing the single revealed preference data point from a cross-section survey with one, or several, stated preference data points. More information from each respondent can lead to increased econometric efficiency. Fifth, stated preference data has limitations which can be addressed by combination with revealed preference data. Hypothetical bias results when hypothetical choices do not fully reflect budget and time constraints. Combining stated preference data with revealed preference data grounds hypothetical choices with real choice behavior. Sixth, combining revealed preference and stated preference data can be used to test the validity of both the revealed and stated preference methods. Convergent validity exists if two methods for measuring an unknown construct (i.e. willingness to pay) yields measures that are not statistically different. Predictive validity of jointly estimated revealed and stated preference data is the ability of the joint models to outperform independently estimated revealed and stated preference data models in predicting actual behavior. There are a number of types of revealed preference (RP) and stated preference (SP) data combination and joint estimation studies in the literature (see
Whitehead et al. 2008 for an extensive review). We consider a data combination study as any research effort that employs both RP and SP data in some way (see Carson et al. 1996 for a review of the early data comparison studies). In contrast, a joint estimation study occurs when the relationships between the independent variables and the dependent variables are estimated in a single model. Joint estimation studies are a subset of data combination studies. RP and SP data are combined and analyzed with three major classes of econometric models: frequency data, mixed data, and discrete data models.