Improving willingness to pay estimates for quality improvements through joint estimation with quality perceptions.

• Author: Whitehead, John C.

1. Introduction

   The contingent valuation method (CVM) is a stated preference approach to the measurement of the value of changes in the allocation of nonmarket environmental and natural resources (Mitchell and Carson 1989). The CVM has clear advantages when compared to revealed preference methods in which actual behavior is used to develop estimates of value (e.g., hedonic price method, travel cost method). Stated preference methods are most useful when an ex ante policy analysis must consider proposals that are beyond the range of historical experience. The CVM is more flexible than the revealed preference methods, allowing the estimation of the impacts of a wide range of policies. The CVM can be used to estimate nonuse values (i.e., passive use values) and ex ante willingness to pay under uncertainty (Whitehead and Blomquist 2006).

   Several issues indicate that the CVM is not a flawless approach to measuring environmental values for policy analysis. (1) The methodological challenges include the potential for hypothetical bias, temporal bias, sensitivity of willingness to pay estimates to multipart policy (i.e., embedding, sequencing), and the bias of a reliance on willingness to pay, relative to willingness to accept questions, when the appropriate property rights are held by the respondent (Whitehead and Blomquist 2006). Hoehn and Randall (1987) define a "satisfactory benefit cost indicator" as one that does not overstate the present value of net benefits of policy. More methodological research is needed before I can conclude that the CVM estimates of willingness to pay are satisfactory benefit-cost indicators. For example, if willingness to pay suffers from hypothetical bias, benefits will be overestimated. Nevertheless, the CVM (and other stated preference approaches) is the only option for estimation of the benefits of a broad range of policy questions.

   This paper addresses a potential problem where willingness to pay statements are based on subjective perceptions about the environmental quality change instead of the objective change that is prescribed by the policy. In this case, willingness to pay may be biased if the subjective change in quality diverges from the objective change. I argue that standard attempts to control for this divergence may fail. An alternative instrumental variables approach is introduced that may improve the accuracy of willingness to pay estimates.

   In the next section I describe the relationship between willingness to pay and quality perceptions, I then describe the potential empirical problem. Next, the empirical willingness to pay model is formally described. The survey used to collect the data and the data used to implement the model are also described. The application is to water quality improvements in the Neuse River, North Carolina. Empirical results using two different quality measures are presented. Conclusions and suggestions for future research follow.

2. Willingness to Pay and Quality Perceptions

   The theoretical construction of willingness to pay for quality improvement shows that willingness to pay is a function of prepolicy and postpolicy quality levels, among other variables (Whitehead 1995). CVM surveys should carefully describe both quality levels and ask for respondent willingness to pay for the change in quality (Mitchell and Carson 1989). A crucial assumption is that respondents are valuing the objective quality change that the survey asks them to value. This assumption may not hold in many applications, especially those in which one or both quality levels are not explicitly described and when heterogeneous respondents have varying levels of prior information about the quality change.

   For example, in a well-funded study that employed in-person interviews, Carson and Mitchell (1993) thoroughly describe baseline national water quality as "not boatable" and improved water quality as "boatable, fishable, and swimmable" using visual aids and extensive text. In contrast, many CVM research budgets are not adequate to pursue extensive descriptions of existing quality and changes in quality. With smaller research budgets that may lead to mail or telephone interviews, important text detailing the environmental quality change may be discarded. For example, in the CVM application presented here, respondents are asked to value a water quality improvement from the current water quality level to a water quality level that is fishable, swimmable, and drinkable. The current water quality is not explicitly described to respondents during the telephone interview. I rely on existing respondent knowledge about current water quality.

   Heterogeneous respondents may have varying subjective perceptions about the current environmental quality level and the hypothetical changes described during the CVM interview. This may be true even when current quality and the quality change are thoroughly described, as in Carson and Mitchell (1993), but it is especially true when the quality change is not explicitly described and assuming that perceptions about quality are homogeneous. In the current application, some might consider current water quality to be too poor for fishing and swimming. Other respondents might consider current water quality to be fishable but not swimmable. With either explicitly described quality change or implicitly understood quality change, CVM questions elicit willingness to pay values that may vary based on differences in respondent quality perceptions. The variation in willingness to pay due to the variation in quality perception will not be accounted for by the researcher who ignores the differences in quality perceptions across respondents, adding to the error of the willingness to pay estimates.

   Ignoring the divergence between perceived quality and objective quality (i.e., quality as described in the survey) in empirical models of willingness to pay leads to the well-known omitted variable problem. For examples of studies that may suffer from omitted variable problems, Hurley, Otto, and Holtkamp (1999) estimate the willingness to pay for delaying nitrate contamination in drinking water and Stumborg, Baerenklau, and Bishop (2001) estimate the willingness to pay for a reduction in phosphorus pollution in lakes. In both cases the perceived quality change is likely to vary across respondents. Neither of these studies includes measures of attitudes or perceptions about the pollution problem in their models of willingness to pay. These omitted variables may cause bias in the estimates of coefficients on variables that are correlated with perceived environmental quality. More generally, omitted variable bias may help explain some poor results from CVM research, such as poor fits and even unexpected signs.

   One solution to the omitted quality variable problem is to include a proxy variable for quality in the model. In the case of willingness to pay for quality improvements the approach is to elicit perceived quality, or variables that may be related to quality (e.g., attitudes, satisfaction ratings), from survey respondents and include these measures as determinants of willingness to pay. Many CVM studies have followed this approach. For example, Kwak, Lee, and Russell (1997) and Yoo and Yang (2001) measure status quo drinking water quality with scale variables measuring "the respondent's attitude toward current tap water quality" and "degree of satisfaction the respondent has with current tap water quality." Both studies find that as the proxy for current drinking water quality increases, willingness to pay decreases.

   Most studies that include quality perceptions in the willingness to pay model ignore the fact that varying subjective quality perceptions are due to the heterogeneity of respondents and the information and attitudes that they bring to the CVM survey. In contrast, Danielson et al. (1995) estimate the determinants of perceived air and water quality and find that they depend on demographics, environmental knowledge, and environmental attitudes. This approach illustrates a problem with including quality perceptions in willingness to pay models. Quality perceptions may be affected by the same unobserved characteristics that influence willingness to pay. If unobserved tastes are correlated with both perceived quality and willingness to pay, the coefficient on the quality perception variable will be biased in a willingness to pay regression model. The bias is due to the correlation in the error terms in the...