Over the past decade, a consistent and frustrating pattern of empirical evidence has accumulated suggesting a significant diver- gence of willingness-to-pay (WTP) measures of value, where individuals buy an object, and willingness-to-accept (WTA) measures of value, where individuals sell the same object. Field-contingent valuation studies first uncovered the pattern, and laboratory markets have confirmed that the divergence is persistent (see e.g., Judd Hammack and Gardner M. Brown, Jr., 1974; Robert D. Rowe et al., 1980; Jack L. Knetsch and John

A. Sinden, 1984; David S. Brookshire and Don L. Coursey, 1987). The divergence is troubling in that the interpretation of stan- dard theory predicts that with small income effects WTP and WTA should be equiva-

*Shogren: Department of Economics, Iowa State University, Arnes, IA 50011, and School of Forestry and Environmental Studies, Yale University, New Haven, CT 06511; Shin, Hayes, and Kliebenstein: De- partment of Economics, Iowa State University, Ames, IA 50011. Tom Crocker, John Braden, and Helen Jensen provided helpful comments on experimental design. Young Park and Todd Holt supplied valuable research assistance. We gratefully acknowledge the financial support of the Food Safety Consortium. Two anonymous reviewers added comments that have im- proved the paper. All errors remain our own. This is Journal Paper No. 5-14801 of the Iowa Agriculture and Home Economics Experiment Station, Ames, IA (Pro- ject No. 2994).

lent, or at least within a tight bound (see Robert Willig, 1976; Alan Randall and John

R. Stoll, 1980). Moreover, since valuation measures are used for the study of many public-policy questions, these results raise questions about which procedure to use in practice. The evidence that WTA measures significantly exceed WTP measures suggests a need to reexamine the analytical founda- tions of value measures.

In response, Michael W. Hanemann (1991) has offered a straightforward expla- nation of why the value divergence occurs and by how much. By recognizing that sub- stitution effects have a more important role than previously realized, Hanemann dem- onstrated that the divergence can range from zero to infinity, depending on the de- gree of substitution between goods and given a positive income elasticity. Hanemann showed that one should only expect conver- gence of WTP and WTA value measures when the good in question has a very close substitute. When the good has an imperfect substitute, a value divergence will exist and will expand as the degree of substitution decreases.

This paper tests Hanemann's proposition in a nonhypothetical experimental auction market. Our results provide some support for his argument. We find that for two pri- vate market goods with a relatively close substitute (a candy bar and a coffee mug) the divergence of WTP and WTA value measures disappears with repeated exposure to the market. In contrast, for a private nonmarket good with no close substitute (reduction of human health risk) the diver- gence is robust and persistent, even given repeated market participation and full in- formation on the characteristics of the good.

This paper proceeds as follows. Section I describes our general experimental design in terms of Hanemann's model. Sections I1 and I11 outline the experimental procedures and results. Section IV describes a second experiment that explores the relationship between the substitution effect and the en- dowment effect. Our conclusions are offered in Section V.

I. Substitution Effects and the General
Experimental Design

Hanemann (1991) reconsidered the work of Randall and Stoll (1980) on value mea- sures given changed quantities of the good. He argued that the widely interpreted view of Randall and Stoll's result, which implies approximate equality between WTP and WTA is "misconceived." By recognizing that the difference in value measures depends on both income and substitution effects, Hanemann demonstrated that the fewer available substitutes the greater is the diver- gence between WTP and WTA. Intuitively, this conclusion makes a great deal of sense. One's willingness to accept compensation for a decrease in the level of a unique good, either private or public, need not equal one's willingness to pay for a good where one is constrained by income.

To illustrate, assume that an individual derives utility, u = u(x, q), from consuming a numeraire good (x) and the good under consideration (q; e.g., candy bar, coffee mug, or health). In general, WTA will exceed WTP unless there is no income effect, in which case WTA equals WTP (see e.g.,

P. R. G. Layard and A. A. Walters, 1978 pp. 150-53). The standard argument is that income effects are small for the changes measured in most valuation studies so this cannot account for the observed WTPWTA divergence. Randall and Stoll (1980) take this a step further, arguing that when goods are sold in competitive markets with zero transactions costs (i.e., candy bars and coffee mugs), the goods possess the charac- teristics of money, so there will be perfect substitutability. Perfect substitution implies a linear indifference curve for x and q (i.e., a frictionless intermediate monetary exchange of commodity holdings).

Consider Figure 1. The top panel shows the Randall and Stoll(1980) scenario where intermediate monetary adjustments create perfect substitution between the two goods. Here the WTA measure is the quantity of the numeraire good required to compensate the individual for forgoing a change in q from q, to q,. This is the amount AD which puts the individual on the higher indiffer- ence curve, but which maintains the original q, consumption level. The WTP measure is the quantity of the numeraire good that one can take from an individual after the change to q,, while still leaving him or her as well off as before. This amount is BC. Given perfect substitution, BC equals AD. This is Hanemann's (1991 p. 637) proposition 1. Note that Randall and St011 (1980) also point out that, not only should WTP equal WTA, both should equal the average market sell- ing price of the good.

Now consider the case of goods such as health, where markets are incomplete or goods are lumpy or indivisible. Hanemann's insight was that certain goods like health cannot be perfectly exchanged for money. There is friction in the market, thereby vio- lating Randall and Stoll's case of intermedi- ate monetary adjustments in commodity holdings. This leads to indifference curves of the standard shape, strictly convex to the origin.' Hanemann demonstrated that in the quantity-change case WTA will exceed WTP; the divergence depends on the in- come elasticity divided by the Allen-Uzawa elasticity of substitution. Therefore an increase in income elasticity or a decrease in

a and all and Stoll (1980) do recognize this possibil- ity (note their brief discussion on p. 452).

I

WTP

WTAl

If we can reject the convergence proposi- tion, then we cannot support Hanemann's argument. In this case, other explanations such as the endowment effect or loss-aver- sion become more attractive (see e.g., Daniel Kahneman et al., 1990). If we cannot reject the proposition, however, then we can offer support to the conjecture that the degree of substitution is a key to understanding the disparity between WTP and WTA measures of value.

11. Experimental Procedures

The experiment was divided into two stages. Stage 1 was the market-good auction. Stage 2 was the nonmarket-good auc- tion. Subjects participated in both stages either for the WTP or WTA experiment. See the Appendix for instructions for the WTP experiment. The WTA experiment was identical to the WTP experiment in all as- pects except for the value measure and ini- tial ownership of the good.

In stage 1, each subject was provided an initial income of $3 and a small piece of candy. To facilitate learning and value for- mation, the auction was repeated over five trials. The number of trials was selected after pretesting to determining how quickly individual value measures stabilized. Note that to control wealth effects, we made the subjects fully aware that only one of the five trials was binding. The binding trial was selected at random by a Monte Carlo num- ber generator on a personal computer. In an attempt to elicit preference accurately we used a Vickrey second-price sealed-bid auction (see William Vickrey, 1961). The Vickrey auction has successfully elicited val- ues in various experimental settings (see Vikki Coppinger et al., 1980; Don L. Coursey, 1987; Shogren, 1990).

The market good was a regular-size brand-name candy bar. Each subject was asked the maximum he or she would be willing to pay to upgrade the small piece of candy to the brand-name candy bar. For each trial, each subject recorded a bid on a recording card that was collected by the monitor. The highest bidder's identification number and the reigning price (the second- highest bid) were posted as public informa- tion on a blackboard. Each subject was given a $3 endowment in the candy-bar stage.

Stage 2 was the nonmarket-good auction. The procedures were similar to those in stage 1, with some noted exceptions. Each participant was given an initial income of $15. Two types of food items were then shown to the subjects with a description of each item. The first type was the test prod- uct, which represented food purchased from a local source with a typical chance of being contaminated with a food-borne pathogen from one-time consumption. Five food-borne pathogens were considered in five separate experimental sessions: Campylobacter, Salmonella, Staphylococcus aureus, Trichinella spiralis, and Clostridium perfrin- gens.' All five pathogens occur in the United States. The test product was provided to every participant as a free lunch. The sec- ond food type was stringently screened food. The stringently screened food had been tested for food-borne pathogens and had a low probability (one in 100 million) of caus- ing food-borne illness.

Description of Salmonellosis: Symptoms are those of a mild flu-like in- testinal disease of short duration with abdominal pains, nausea, vomiting, and diarrhea. The actual individual chance of infection of salmonellosis is 1in 125 annually. Of those individuals

who get sick, 1individual out of 1,000 will die annually. The average cost for medical expenses and productivity losses from a mild case of salmonel- losis is $220.

Given this information, "informed" bids were elicited in trials 11-20.

The computer randomly selected one of the 20 trials as binding. The highest bidder paid the displayed second-highest bidding price and ate the stringently screened food. The highest bidder's take-home income was $15 minus the price paid for the screened food product. The other bidders ate the test product and took home $15. Note that the subjects had to eat the food item to leave the experiment with the take-home income.

Table 1summarizes the experimental de- sign for both the WTP and WTA exper- iments. One hundred forty-two subjects participated in the experiment. Each experi- mental session included between 12 and 15 subjects. All were undergraduate and grad- uate students from Iowa State University (ISU), recruited campus-wide. Note that a subject participated in either the WTA or the WTP experiment, not both. Also, each subject was only confronted with one food- borne pathogen description, not all five, re- gardless of whether that subject was in the WTA or the WTP experiment. After each subject read the instructions and answered a set of questions to test his or her under- standing of the experiment and the monitor answered all relevant questions, the experi- ment began. All experiments were conducted in the ISU meat-testing laboratory with modern kitchen facilities. The ISU meat lab conducts food-tasting experiments on a regular basis. The lab is actively in- volved in all aspects of meat processing and handling, thereby providing a unique setting for our experiment.

111. Results and Discussion

Overall, we cannot reject the convergence proposition. Table 2 and Figure 2 illustrate that the WTP and WTA measures of value for the market good were not significantly different, with the exception of the first

aOne-tailed t test.

bMann-Whitney U test.

0.65 1

I

statistically significant. By trials 3, 4, and 5, the average WTP and WTA values converged to differences between 1 cent and 3 cents. We cannot reject the hypothesis of equality of the WTP and WTA measures. In addition, the results in Table 2 support Randall and Stoll's (1980) prediction that

....................................................................................................................

o,25 ....................................................................................................................

tion will also converge to the average mar-

I I I I

candy bar worth about 50 cents, the ob-

FIGURE2. WTP AND WTA COMPARISON:

served average value near 40 cents is strik-

CANDYBARS

ing. With perfect substitution, value measures converged, and they converged on the average market price.

Tables 3 and 4 and Figures 3-7 illustrate that the majority of the WTA measures for

H,: WTP = WTA

Salmonella WTP (1/137,000) WTA

Staphylococcus WTP aureus

(1/173,694) WTA

Trichinella WTP spiralis

(1/2,628,000) WTA

Clostridium WTP perfringens

(1/26,280,000) WTA

*Ha rejected at the 5-percent significance level. * * Ha rejected at the 1-percent significance level.

WITH ELIMINATION

Pathogen (probability of illness)

Campylobacter

(1/125,143)

Salmonella

(1/137,000)

Staphylococcus aureus

(1/173,694)

Trichinella spiralis (1/2,628,000)

Clostridium perfringens

Notes: Sample sizes are as follows: Camplyobacter (WTP = 13, WTA = 121, Salmonella (WTP = 13, WTA = 13), Staphylococcus aureus (WTP = 10, WTA = 131, Trichinella spiralis (WTP = 11, WTA = 131, Clostridium perfringens (WTP = 11, WTA = 13). Columns for t and U report results of one-tailed t tests and Mann-Whitney U tests, respectively. Sample standard deviations are in parentheses.

*Denotes rejection of H, at the 5-percent significance level.

**~enotes rejection of H, at the 1-percent significance level.

VOL. 84 NO. I SHOGREN ETAL.: WILLINGNESS TO PAY/ ACCEPT

(I)

a

To test for a possible endowment effect we repeated our experiment, but the auctioned good was now an Iowa State coffee mug available from the campus bookstore for $5.20. All design aspects followed as closely as possible to our earlier experiments: a Vickrey auction, 15 subjects per treatment, mugs placed directly in the sub- ject's hands, a $15 initial endowment, and repeated market participation (ten trials for the ISU mug stage). In addition, we repli- cated the initial candy-bar stage ($3 initial endowment and five trials) to maintain con- sistency with the food-borne pathogen treat- ments. The baseline treatment was a WTP auction that allowed the highest bidder to upgrade his or her plain plastic mug (worth $1.60) for an ISU mug. We compared two WTA treatments to the WTP baseline. The first treatment was a WTA auction where the low bidder would receive compensation (second-lowest bid) for exchanging his or her ISU mug for a plain mug. The second treatment was identical to the first, except now subjects were told they could purchase (at the market price) an unlimited number of ISU mugs right outside the door immedi- ately after the experiment. This eliminated any transaction costs that could create fric- tion in trade. Our hypothesis is summarized below.

TABLE 5-COMPARISON OF MEAN WTP AND WTA BIDS (IN DOLLARS) IN MUG TREATMENTS

Trial Treatment n 1 2 3 4 5 6 7 8 9 10 WTP, baseline 15 2.37 2.40 2.89 2.90 2.70 2.86 2.74 2.62 2.68 3.09 withoutmugsale (1.484) (1.275) (1.111) (1.072) (1.047) (1.139) (1.022) (1.291) (1.187) (1.357)

WTA, without 15 6.55 4.17 3.19 3.06 2.88 4.14 3.53 3.25 3.12 2.30 ISU mug sale (4.894) (3.261) (2.990) (2.401) (1.903) (5.125) (3.411) (3.632) (2.522) (1.970)

WTA, with 15 6.50 4.74 3.67 2.98 3.27 3.75 2.65 3.13 3.29 2.49 ISU mug sale (4.452) (4.147) (3.202) (3.104) (3.262) (4.331) (2.746) (3.010) (3.135) (2.175)

Note: Sample standard deviations are in parentheses.

design and that of Kahneman et al., and our use of the Vickrey auction may play a role (see e.g., John H. Kagel et al. [I9871 on overbidding). Regardless of why our results differ from Kahneman et al., we observe the convergence of WTP and WTA values for the candy-bar and coffee-mug treatments but a persistent divergence in the values for the food-borne pathogen treatments. Hane- mann's (1991) substitution effect appears to organize observed valuation behavior in a predictable fashion. Future research should continue to compare how alternative elicita- tion mechanisms affect revealed values in experimental markets, holding constant monetary incentives, opportunity for market expkience, subject pools, and the auctioned or traded good.

APPENDIX Experimental Znstructions [Exact Transcript]

You are about to participate in an experiment about decision making. Please follow the instructions carefully. The United States Department of Agriculture has provided funds for this research.

In this experiment, you will be asked to decide how much you would be willing to pay for safer food [to decide the minimum amount you would be willing to accept for taking the test product food, instead of keeping your safer food]. The experiment has two stages.

Your starting income will be $3 in stage 1. Your income will be $15 for stage 2. Your take-home income will consist of your initial income ($3+$15) minus [plus] the values of goods purchased.

You will submit your bidding price on a recording card. Note only one of the trials in stage 1will be binding and only one of the twenty trials in stage 2 will be binding. A number will be randomly selected to identify these binding trials.

You cannot reveal your bids to any other participant. Any communication between bidders during a trial will result in an automatic penalty of $3.

Step 1: You own the candy [candy bar] free in front of you. Your initial income is $3.

Step 2: Let's say you are willing to pay $X for the piece of candy and $Y for a candy bar. The difference ($Y -$X) is what you are willing to pay to upgrade your piece of candy for a candy bar. Please indicate your willingness to pay to trade the piece of candy for a candy bar. Do not state what you would pay for an entire candy bar. Only state the difference ($Y -$X) you are willing to pay.

Step 3: Please write your bid (difference) for the one candy bar on the recording card. The monitor will announce the highest bidder and display the price of the candy bar (second-highest bidding price) on the blackboard.

Note: For example, if the highest bid was $a and the second-highest was $P, the highest

bidder would receive the candy bar and must pay $P. Step 4: There will be five trials. Step 5: Only one trial will be binding. After the five trials, a number will be randomly

selected to determine which trial is binding. The highest bidder of that trial will exchange the piece of candy for the candy bar and must pay the displayed price (i.e., the second-highest bid).

Note: In the event that there is a tie for the highest bid, those participants will be asked to bid again.

Please answer the following questions, which are designed to help you understand stage 1. Do not hesitate to ask the researchers if you have questions.

1. Suppose that person A is the highest bidder in the first trial, person B is the highest bidder in third trial, and person C is the highest bidder in fifth trial. If, after five trials are finished, we randomly select the third trial, then who will purchase the candy bar?

    

2. If your $a bid is the highest in the third trial, and the second-highest bid is $P, what price will you pay for the candy bar? $

    

3. If your bid is not the highest in the third trial, which is randomly selected, how much should you pay for the piece of candy? $

    

Step 1: There are two types of food. The features are each described below.

Test Product Stringently Screened

This food has a typical chance This food has been subjected to of being contaminated with the stringent screening for food-borne pathogen Salmonella ; Salmonella. There is a 1in

i.e., it is purchased from a 100,000,000 chance of getting

local source. salmonellosis from consuming

Step 2: You own a test product sandwich free in front of you. Everyone has the same sandwich. You also have initial income, $15. Step 3: Let's say you are willing to pay $X for the test product sandwich and $Y for the stringently screened sandwich. The diference ($Y -$X) is what you are willing to pay to reduce the risk of illness from the food-borne pathogens. Please indicate your willingness to pay to reduce the risk of illness. Do not state what you would pay for the entire stringently screened sandwich. Only state the diference ($Y -$X) you are willing to pay. The highest bidder will upgrade his or her test product sandwich for the stringently screened sandwich. He or she will pay the second-highest bidder's price. Step 4: There will be twenty trials. Step 5: After all twenty trials are complete, we will randomly select one binding trial to determine who buys the stringently screened food. Note: The sandwich has to be eaten to leave with the take-home income.

Please answer the following questions, which are designed to help you understand stage 2. Do not hesitate to ask the researchers if you have questions.

1. There are twenty bidding trials. If person A is the highest bidder in the first trial, person B is the highest bidder in the eighteenth trial, and the eighteenth trial is selected, then who will receive the stringently screened food?

    

2. If your $a bid is the highest in the eighteenth trial, and the second highest bid is $P, what price will you pay for the stringently screened food? $

    

1. What do you think is the chance of becoming ill from Salmonella, given that you eat an average amount of typical food products in the United States over one year?

    

2. What do you think are the important sources of the food-borne pathogen, Salmonella, in

    

Answer: chance out of 1million people the United States?

Symptoms are those of a mild "flu-like" intestinal disease of short duration with abdominal pains, nausea, vomiting, and diarrhea. The actual individual choice of infection of Salmonel- losis is 1in 125 annually. Of those individuals who get sick, 1individual out of 1,000 will die annually. The average cost for medical expenses and productivity losses from a mild case of