Does money talk? — The effect of a monetary attribute on the marginal values in a choice experiment

Abstract

When designing choice experiments for nonmarket valuation the role of the price attribute is of major importance. In the energy sector the uncertainty of future direction of changes in prices makes it difficult to include an adequate price vector in the design. We separately investigate the implication of using price vectors with increases and decreases in tariffs from current levels, on marginal value estimate from choice experiment data developed using prospect theory. In addition, we also analyse the effect of excluding the price vector on these marginal values. By and large, our results support the neoclassical theory as we find that the means of the conditional estimates of the marginal values of attributes are unaffected by the direction of the price change and from exclusion of the price attribute. However, the distributions show a larger spread of values when the choice experiment implies a tariff decrease, which may have policy implications.

Introduction

To combat climate change, the energy sector has undergone major changes during the last decade. The most notable change is the increased investments in renewable sources of energy generation. Another feature of changes in the energy sector is the deregulation of the energy market, where one of the key issues is to establish incentives for the agents to undertake optimal investments from a societal perspective to ensure reliable supply of power. As a result, policymakers as well as other agents in the energy sector, need to use stated preference studies to evaluate many of the discussed changes, e.g., to value the environmental impacts of building a new wind farm or the decreased probability of future power outages (e.g., Bergmann et al., 2006, Borchers et al., 2007, Carlsson and Martinsson, 2008, Carlsson et al., forthcoming, Ek, 2005, Groothuis et al., 2008, Hanley and Nevin, 1999, Koundouri et al., 2009, Ladenburg et al., 2005, Longo et al., 2008, Meyerhoff et al., 2010, Scarpa and Willis, 2010). Since many of the suggested changes in the energy sector involve simultaneous changes of many attributes, choice experiments have been frequently applied. A choice experiment study provides policy makers with the marginal rates of substitution (MRS) between different attributes, most often between a non-monetary and a monetary attribute³ expressed as marginal willingness to pay (WTP). The estimated WTP may then be directly included in economic evaluations such as cost–benefit analyses.^4,5

The main objective of this paper is to study whether the marginal values, i.e., marginal WTP and MRS and their respective distributions, are affected by the direction of change in the price vector, described either as an increase or a decrease from the current price level in a choice experiment, with an application to the energy sector in Chile. A key consideration in our analysis is whether prospect theory, as originally devised by Kahneman and Tversky (1979) in a gambling situation and later extended to also include the case with certain outcomes in Tversky and Kahneman (1991), according to which gains and losses from a reference point are valued differently, can explain behavioural responses to different directions (increased and decreased) of change for the price attribute. Or indeed, whether the observed differences can be adequately explained by neoclassical theory, where changes from a reference point are, at the margin, valued equally.

The uncertainty regarding the direction of change in price is a salient reality in the energy sector. The effect on the electricity price paid by consumers due to the introduction of new sources of energy generation or to the reduction in probability of future power outages is often uncertain. This is partly related to the uncertainty surrounding future production costs and technology development as well as to the direction and extent of changes in taxes and subsidies in the energy sector. These factors make it difficult to design a choice experiment. However, if the marginal values are sensitive to the direction of change in the price vector then, in any valuation study, it is extremely important to correctly determine the absolute level of price changes, which of course includes the direction of change in price from the current level. At a broader perspective, the possibilities to generalize the results from one study to another, e.g., by using benefit transfer, are also limited if the marginal values are sensitive to the direction of changes in prices.⁶

In the empirical analyses of choice experiment data, linear-in-attribute utility function is often assumed and is considered a reasonable assumption for small enough changes. It is often implicitly assumed that the estimated marginal WTP and MRS from a choice experiment should be unaffected by the direction of a small price change (as long as the difference in absolute levels between the prices remains the same). However, frequent research, especially in psychology, has shown that preferences are reference-dependent; people's behaviour is affected by a reference price, which in our case corresponds to the current price level. Kahneman and Tversky (1979) formulated the so-called prospect theory, showing that people react differently to gains and losses related to the reference point in gambles (see also e.g., Thaler, 1985). Tversky and Kahneman (1991) extended their previous findings to also include the case with certain outcomes. The basic feature of the reference dependence is illustrated by an S-shaped value function around the reference point. The function is concave above the reference point, and convex below the reference point showing that the marginal utility for a loss is larger than for a gain by using separate value functions in the two different domains. This has been tested experimentally, where subjects either are endowed with a good or had the possibility to buy the same good. The experimental findings support prospect theory since whether or not a subject is endowed with the good significantly affects her valuation of the change (e.g., Bateman et al., 1997, Kahneman et al., 1990, Knetsch, 1989). However, in an overview paper, Mazumdar et al. (2005) summarize several marketing studies investigating the effect of asymmetric reference price on choices,⁷ i.e., when observed price of a good differs from the reference price of the consumer, where the overall findings when comparing previous studies weakly support prospect theory. However, as discussed in Mazumdar et al. (2005), the effect diminishes when segmenting people according to price sensitivity and brand loyalty. On the other hand, List (2003) and List (2004) find support for the idea that neoclassic theory explains behaviour among people with experience of the good.⁸

More recently, prospect theory has been explicitly investigated in choice experiments. Some studies have investigated reference point when analysing gains and losses in transportation studies. For example, Hess et al. (2008) find evidence of asymmetrical responses to increases and decreases in the value of attributes (travel time and travel cost) compared to the reference alternative in a study of choices on tolled routes. Masiero and Hensher (2010) investigate prospect theory in a choice experiment applied to freight transport to see whether changes in the transport costs affect the marginal willingness to pay and willingness to accept for other attributes and find support for prospect theory. They show consistent disparity between the two measures, with loss in benefits being higher than gains.⁹ Apart from some studies exploring preferences for travel or access time (Hess, 2008), the applications using choice experiments are scarce (e.g. Lanz et al., 2010 address and confirm this asymmetry in water services). Therefore on balance, the results are mixed since some studies, especially laboratory experiments and choice experiments, have found evidence of the existence of prospect theory to explain choices, while others have found the opposite e.g. List, 2003, List, 2004.

In the case of choice experiments, the increase in price from the current level represents the loss in prospect theory terminology, as people have to pay more for electricity than what they are currently paying, while a decrease from current price level represents the gain. This has direct implication on the marginal WTP given that the marginal utility for the non-market attribute is unaffected by price change in a choice experiment study. The marginal WTP for an attribute in a choice experiment is expressed, if we assume a linear-in-attribute utility function, as the ratio between the marginal utility of the attribute and the marginal utility of income. If the marginal utility of income is affected by price change as predicted by prospect theory, then the increase in price would result in a larger marginal utility of income compared to a decrease. Therefore, the marginal WTP is lower for the same attribute if the price is expressed as an increase from the initial level compared to a decrease.

A second key objective of this paper is to explore the impact of excluding the price attribute altogether from the choice experiment. In some cases the price is not a relevant attribute since it is already decided that a change will take place (regardless of the price), or for a given cost of the project, the focus is on the optimal combination of attributes, in which case the policy of interest relates to obtain the MRSs between non-monetary attributes in order to compare the impact of different projects. For example, Aas et al. (2000) and Wattage et al. (2005) use choice experiment methodology to study preferences for different fisheries management programmes in Norway and the UK respectively by studying the relative importance of diverse (non-monetary) attributes. Another example is Casey et al. (2008), which use attributes, such as better opportunities for education and health care, to evaluate the willingness to accept risk for different spill risk scenarios in indigenous communities in Brazil. Thus, it is enough to use the information on MRS to investigate the combination of attribute levels that maximizes the social welfare (or minimize the disutility) conditional on the fact that a project will be implemented. In such case, we investigate the effect on marginal values, i.e., MRSs, of excluding the price attribute from the set of descriptors of the alternatives used in the choice experiment. We do so by comparing the results to the standard case where it is included. It has been shown that preferences measured by choices in the presence and absence of prices differ (e.g., List, 2002, Slovic and Lichtenstein, 1983, Tversky and Thaler, 1990). In a recent study, Carlsson et al. (2007) compare the results from choice experiments with and without the price vector in exploring preferences for food products. They find that preferences were affected by the inclusion of a price vector and more specifically that the MRSs between attributes are lower when a price vector is included. We extend their analysis by investigating the effect on MRS of excluding a price vector from the choice experiment compared to the cases where there is an increase or a decrease in price from the current level. To this end, we use a sample of respondents who faced a choice task without the price attribute, but to ensure complete comparability between treatments we use the same experimental design used in the cases where the price level is included.

Our research questions are explored with choice experiments on the introduction of wind power plants in various types of location in Chile. We use a split sample design where subjects were allocated to a survey with either increased or decreased electricity prices from current level, or no price vector at all. In our application on the construction of future wind farms in Chile, the future generation cost and design of policy instruments used by policy-makers are genuinely uncertain, and thus make our research design credible.

The rest of the paper is organized as follows. In Section 2, we present the policy context and describe the design of our choice experiment study and the econometric model. Section 3 reports the obtained results and the discussion of findings. Finally, Section 4 concludes the paper.