Cost of public goods affects willingness to pay for them

Baron, J., & Maxwell, N. P. (1996). Cost of public goods affects willingness to pay for them. Journal of Behavioral Decision Making, 9, 173–183.

Jonathan Baron1
University of Pennsylvania

Nicholas P. Maxwell
University of Washington, Bothell


We provided information about costs and benefits of public goods, mostly forms of risk reduction. Judgments of willingness to pay (WTP) for the goods was affected by cost as well as benefit, even when subjects judged the benefit to be unaffected by cost. Cost information affected WTP when it took the form of estimated cost or when it was simply implied by past expenditures or by descriptions of how a good would be provided. Cost affected WTP both when each subject judged two cases varying only in cost and when each subject judged only a single cost version of each case. The results can be understood as overextension of a somewhat useful heuristic: things that cost more often yield more benefit. The findings suggest that contingent valuation methods may be improved by eliminating information from which costs could be inferred, so that respondents can focus more easily on benefits alone.


Contingent valuation (CV) attempts to measure the economic value of public goods, such as pollution reduction, by asking respondents to state the largest amount of money that they would be willing to pay, their willingness to pay (WTP), for a good. In theory, WTP provides a measure of a good's benefit that can be compared to the cost of providing it. If this benefit exceeds the cost, we infer that the good should be provided.
Users of CV typically assume that the value of the benefits is a function of the effects of the good in question, not its cost. In particular, inefficiency could result if WTP increased with increasing cost. For example, suppose that government program X has a little more benefit than program Y, so that WTP for X would be a little higher if the costs of the two programs were the same. But suppose that respondents would express higher WTP for Y than for X if they learned that Y was more costly. Then, if the government used WTP as an index of preference, and if cost were not a major issue, it would assume that people preferred program Y. A government that acted on this information would choose Y over X. The citizens would end up paying more and getting less benefit.
Some evidence suggests that reported WTP increases with cost of the good even when benefit is constant. For example, in one study, WTP for a hypothetical bottle of beer, to be consumed on the beach, depended on the source of the beer (Thaler, 1985). WTP was higher if the beer came from a fancy resort hotel than if it came from a mom-and-pop store. Because the beer was to be consumed on the beach, none of the atmosphere of the hotel would be consumed. Other evidence indicates that WTP is affected in part by judgments of what a fair price should be (Kahneman, Knetsch, & Thaler, 1986). In general, people like to pay what something is worth. They do not want to be taken advantage of by paying much more than the cost, and they are reluctant to take advantage of excessively low prices (Winer, 1986).
Some evidence points to a role of "fair prices" in WTP judgments for public goods as well as private ones. When respondents provide justifications for their WTP responses, they often refer to the cost of providing the good, or the cost per household (Schkade & Payne, 1994, p. 99). Green, Kahneman, and Kunreuther (1994), asked respondents about their willingness to pay donations for a program to teach English to immigrants. Respondents who were reminded that 20 million other households would be asked the same question differed from respondents without the reminder in two ways. First, they thought that it was less appropriate not to contribute at all. Second, they thought that the appropriate contribution was lower. Apparently, reminding them of the other contributors made them feel obliged to do their share, but it also may have made them wary of contributing more than was needed to cover the cost of the program. In this way, they did not simply report their values, but instead reported something about how much they thought such a program would cost.
Such results can be understood in terms of a disutility for unfairness, for exploiting the provider of a good by paying too little or for being exploited by paying too much. Inefficiency can still result, however, even if we consider this disutility when we evaluate the final outcome. For example, suppose that you have a slight preference for Heiniken's beer over Miller's, your friend asks you your WTP for Miller's from the resort hotel and for Heiniken's from the mom-and-pop store, and you say $5 and $4, respectively, because you think that the fair prices are $5.01 and $3.99, respectively, and you are somewhat affected by your taste. Moreover, your disutility for unfairness depends on the absolute value of the departure from the fair price. If the prices turn out to be $4.75 and $4.25, respectively, then any reasonable friend would get you the Miller's and you will pay $4.75. You will pay more for what you like less, and the degree of unfairness ($0.26) would be the same with either transaction, so you will not be compensated for your loss by engaging in a fairer transaction. The problem is that any reasonable decision procedure based on WTP responses must assume that lower prices are always better and that higher WTPs represent more utility from the good itself.
Many CV studies provide considerable information about how the public good will be provided; this information might lead subjects to focus on the cost of provision rather than the extent to which their own utility is affected. Inefficient decisions could thus arise.
In the present studies, we ask WTP questions concerning hypothetical public goods. We provide information about benefits and about costs of providing the good. We find that respondents are influenced by both kinds of information. Respondents are willing to pay more for goods that are more expensive, holding benefit constant.

Study 1. Hazardous waste.

The first study concerned hazardous waste. We provided information about the costs and benefits of cleanup. Cleanup was expensive, although not out of line with some real examples in the U.S. We asked subjects for WTP and for ratings of benefit.


Subjects were 53 students at the University of Pennsylvania and the Philadelphia College of Pharmacy and Science, solicited by advertising and paid $6/hour for completing the present questionnaire and others.
A questionnaire began, "Potentially hazardous chemicals have been deposited in various waste sites around the U.S. over the last few decades. Most of these chemicals were wastes from manufacturing. In many cases, the companies that produced these wastes did not know that they were hazardous.
"The Superfund program was created in the early 1980s to clean up these wastes. It has now become clear that cleanup is expensive and that choices must be made about which wastes to clean up, and to what degree. The following cases are hypothetical, but they are similar to real cases faced by decision makers in government and business.
"Each of the following cases describes a waste site under consideration for cleanup. Suppose that each waste site is near an urban area. Chemicals from the waste can get into the drinking water, and it can affect those who live near the site in other ways. We give here the number of people who will be affected in the next 50 years, e.g., the number of people who will get whatever disease the chemical causes. A `full cleanup' will completely eliminate this risk. A `partial cleanup' will eliminate half of the risk. The costs of these cleanups are in millions of dollars.
"In each case, in the last two columns, indicate the most you would be willing to pay for each kind of cleanup, if you lived in the affected area. Suppose that the payment will come from a temporary increase in local taxes (property, sales, local income tax). This is a one-time extra payment. Suppose that the cleanup will be done if more than 50% of the people in the area say that they are willing to pay the cost per person. (Naturally, the full cleanup will be done if people are willing to pay for both full and partial cleanup.)" (The purpose of this rule was to impress subjects that the question was not about a fair price. Baron & Greene [in press] found that it has little effect, and McFadden [1994] likewise found little difference between straightforward WTP questions and referendum-type questions.)
Finally, subjects were told, "In the last column, rate the degree of benefit that would result from full cleanup in each case. Assign 100 to the most beneficial cleanup and 0 to no benefit at all."
The table that followed listed three diseases - skin rashes, birth defects, and kidney cancer - each with four cases differing in the numbers, shown in Table 1. In the basic case, given first, 100 people were affected, the cost of the partial cleanup was $100 million, and full was $400 million. The next three cases varied the number of people affected and the cost of partial and full cleanup, respectively, by a factor of four: 400 people affected, $1,600 million cost of full cleanup, and $25 million cost of partial cleanup. The order of these variants was reversed for half of the subjects. Subjects provided WTP for both full and partial cleanup in each case.
Table 1: Cases (3 columns on left) and median responses (3 columns on right), Study 1, collapsed across diseases.
People Cost of Cost of Pay for Pay for Benefit
affected partial full partial full of
in 50 yrs. cleanup cleanup cleanup cleanup cleanup
100 100 400 $75 $233 63
100 25 400 $58 $292 75
100 100 1,600 $87 $300 50
400 100 400 $67 $300 78


WTP was higher when costs were higher, and lower when costs were lower. Table 1 shows the mean responses collapsed across the three diseases. (The pattern of results was the same for the three diseases, although the mean ratings were higher for more severe diseases. Only one subject gave a zero WTP for all three diseases in any condition.) Comparison of the first and third lines shows that WTP for full cleanup was higher when the cost was raised to $1,600 million. Benefit ratings for full cleanup were somewhat lower as cost increased. Comparison of the first and second lines shows that WTP was lower when costs were lowered to $25 million.
For analysis, we measured each effect as the log ratio of the critical case to the first case, and then we divided this by log(4), so that 0 indicated no effect and 1 indicated an effect proportional to the manipulation. For example, if the results in Table 1 were from one subject, the effect of increasing the cost of full cleanup on benefit for full cleanup for that subject would be log(50/63)/log(4). We used nonparametric statistical tests because of the many zero values, which lead to drastically nonnormal error distributions.
By this measure, the effect of cost on benefit was significantly negative (p=.002, two-tailed Wilcoxon test); that is, cleanups with greater costs were given greater WTPs and were rated to have less benefit. 42% of the subjects showed a negative effect of cost on benefit ratings, 44% showed no effect; and 13% showed a positive effect. (One subject was excluded for giving 0 to both cases.) The negative effect does not imply that subjects misunderstood the question. They simply took it to refer to the total package, including costs, rather than to the cleanup alone. The 13% who showed a positive effect, on the other hand, seemed to think that a more expensive cleanup would somehow be better, despite our effort to convey all the relevant information about benefits.
We focused on the plurality of subjects who said that the benefit was the same. In this group, higher cost increased WTP for full cleanup (mean of 0.28 on the 0 to 1 scale, p=.016, one-tailed Wilcoxon test). Most subjects, 56%, showed this effect to some degree, 22% showed no cost effect (equal WTP regardless of cost), and 22% showed a reverse effect. Reducing the cost also reduced WTP for partial cleanup (mean 0.30, p=.034; here, however, only 39% showed the effect, and 30% showed the opposite effect, albeit to a lesser degree). Results were essentially the same for the entire group (except that the effect of reduction was not quite significant).
The effect of the number of people affected in 50 years (100 vs.\ 400) was also significant for WTP for full cleanup (mean of 0.14, p=.042) and for benefit ratings (0.10, p=.004), but not for WTP for partial cleanup (mean -0.01). Results were essentially the same for the entire group. Notice that the effect of cost is at least as great as the effect of the number of people affected.

Study 2. Varying the number of payers

Green, Kahneman, and Kunreuther (1994), found that WTP diminished when respondents were reminded of how many other people are being surveyed. In light of the effect of cost on WTP, we propose that WTP diminishes because respondents are, to some degree, reporting what they feel is their fair share of the customary cost of the program. For example, if there are 100 people who will be paying, and the program will cost $1,000, then respondents tend to feel that they should pay no more than $10. If there are only 10 people paying, then it would not be unfair for each to pay $100. In the Green et al. (1994) study, respondents in the control condition probably underestimated how many people would be paying. Our hypothesis about this effect implies that WTPs will increase as the number of payers decreases, even when all of the scenarios remind respondents of how many people will be chipping in.
To test this, in Study 2, we varied the number of payers across scenarios. Study 2 concerned naturally occurring chemicals instead of human-caused pollution. (We did not expect this to matter.) We also asked subjects for explanations of why each factor was relevant or not.


We gave a questionnaire to 41 subjects solicited as in Study 1. The questionnaire began, "The situations described in this study are hypothetical, but they resemble real situations.
"Recent research has revealed the emergence of hazardous chemicals in communities around the U.S. These chemicals are naturally occurring. They result from temporary infestations of molds. The chemicals get into the drinking water, and become airborne when winds blow dusts into the air. Residents of these communities who drink and breath these chemicals have higher risks of contracting kidney cancer, a serious form of cancer. The chemicals from each infestation remain dangerous for about 10 years.
"Rather than waiting 10 years, communities can hire contractors to clean out the chemicals. This completely eliminates the risk from the chemicals." Then a paragraph described the information to be presented. The instructions continued:
"We also provide an estimate of the cost of cleanup. There will be further shopping for cheaper contractors, and the estimate is not binding: the contractors who provided these estimates may raise their prices after further study. So the actual cost may be higher or lower. To protect the community from cost over-runs, the contractors will be required to state final estimates before beginning work, and they will not be allowed to charge more than their final estimates.
"In each case, indicate the most that you personally would be willing to pay for each kind of cleanup. Suppose that the payment will come from a temporary increase in local taxes (property, sales, and local income tax). Your income is average for the area, about $50,000 per year. This is a one-time extra payment. Suppose that the cleanup will be done if more than 50% of the people in the area say that they are willing to pay the cost per person of the contractor's final estimate." (After some subjects responded with more than their income, we inserted another sentence reminding them of their income. We include all the data in the analysis, however. Analysis was based on within-subject ratios, so the high responders did not get more weight.)
A table, presented here as Table 2, listed eight cases (with the order reversed for about half of the subjects), consisting of all combinations of three factors: community population; cases of cancer; estimated cost of cleanup. The last column, labeled "Maximum pay for cleanup" was blank. Finally, subjects were asked, "Please explain how each of the three factors affected, or did not affect your willingness to pay," and the three factors were listed.
Table 2: Cases (3 columns on left) and geometric mean WTP (column on right), Study 2.
Estimated Your personal
Community Cases of cost of maximum pay
population cancer cleanup for cleanup
10,000 200 $10,000,000 $5,054
10,000 200 $1,000,000 $2,188
10,000 20 $10,000,000 $3,026
10,000 20 $1,000,000 $1,382
1,000 200 $10,000,000 $17,448
1,000 200 $1,000,000 $6,229
1,000 20 $10,000,000 $7,817
1,000 20 $1,000,000 $3,828


WTP increased with cost and benefits, and decreased with number of other payers, as hypothesized. Table 2 shows the geometric mean responses (rather than medians, since all responses were amounts of money - no WTPs were zero). Some responses were very large, probably intended as total cost rather than cost per person. To measure the size of the responses for each subject, we took the log of the maximum WTP for each subject. The effect of each factor was defined as the log of the ratio of the sum of the responses to one level of the factor to the sum of the responses to the other level, e.g., the sum of the odd items divided by the sum of the even items for the effect of cost. We used parametric statistics because very few of these values were 0 or 1.
The effect of all three factors was highly significant. The mean effects (on the 0-1 logarithmic scale described in Study 1) were 0.41 for population, 0.30 for number of cases, and 0.43 for cost (t > 5, p < .0005, for all). Note that the effect for population was in the direction of smaller WTP for greater population: subjects were considering the cost per person required. Subjects who did this tended also to be the ones who took cost into account: the correlation between the two effects was r=0.27 (p=0.04 one tailed).
The tendency to consider cost (but not number of payers) was correlated with the log of the maximum WTP (r=0.58, p < .0005). The cost effect was still present, however, even if we limit the analysis to those nine subjects whose maximum WTP was less than $1,000 (mean 0.14, p=.026). These were the subjects who were most obviously thinking in terms of their own income rather than in terms of total costs.
Most justifications for the relevance of cost simply summarized whether or not cost had an effect, and in which direction. Some suggested that subjects adjusted their WTP to "need," for example: "[Cost] was important because it gave me an idea of how much was needed on my part." "How much is required to be paid by how many of us." "What it costs is what it costs, and it should be cleaned up." One subject inferred that the quality would be better when the cost was higher. We know from Study 1, however, that this cannot explain the whole effect, so it is not surprising that this sort of comment was rare.

Study 3. Prior expenditures.

Discussions of the appropriate value of human life (e.g., Jones-Lee, 1989) often rest on the assumption that we can determine values from past expenditures. Two arguments support this assumption. First, if we are uncertain ourselves, our own past decisions and those of others can be a useful guide: maybe we knew our values better in the past, or maybe other people know their values better than we know our own. Second, when we are inconsistent because we spend more now for some good, such as safety, than we have spent in the past or vice versa, this inconsistency is a signal that gains can be achieved through reallocation. If we have spent $1,000,000 per life saved in the past, perhaps we could save more lives through the same kind of expenditure than we could by spending $10,000,000 per life through some other expenditure now.
But the use of past expenditures can also result from simple confusion of cost and value. The present study asks for values about pollution reduction in a list of cases in which the consequences are easily compared. All consequences involve reductions in cases of kidney cancer by reducing different chemicals. If WTP is consistent with past expenditures on reducing each chemical, then expenditures will be inconsistent across the cases in terms of dollars per case of cancer prevented.


We gave the questionnaire to 19 subjects. The introduction to the questionnaire was like that of Study 2 except that the source of the chemicals was left unspecified (although the reader would probably infer that they were man-made). The chemicals at issue all caused kidney cancer. We presented eight cases, shown in Table 3. We presented the cases in sentences rather than in a table, e.g., "This chemical causes 10 cases of cancer per million if not cleaned up at all. 40% of it has already been removed at a cost of $20 per resident. How much would you be willing to pay for cleaning up another 40%?" The information about prior cost is in the second sentence. The cases varied in the benefit of cleanup as well as in prior cost. The benefit involved two variables: the number of cases of cancer caused by the chemical (5 or 10), and the proportion of the chemical cleaned up at each stage (20% or 40%). For about half of the subjects, the order of high and low costs was reversed.
Table 3: Cases (3 columns on left) and geometric mean WTP (column on right), Study 3.
Prior cost Geometric
Cases of Percent of cleanup mean
cancer cleanup per resident WTP
10 40 $10 $13.28
5 40 $20 $21.61
10 20 $10 $13.93
5 20 $20 $17.96
10 40 $20 $21.67
5 40 $10 $14.21
10 20 $20 $20.51
5 20 $10 $12.71


WTP was affected by prior expenditures as well as by one measure of benefits. Table 3 shows the geometric mean WTPs for each condition. (No WTPs were zero.) Results were analyzed both parametrically and nonparametrically, and the conclusions were the same. Effects were defined on the same 0-1 logarithmic scale as in Studies 1 and 2.
The effects of number of cases (mean 0.18) and of cost (0.41) were highly significant (t > 4, p < .0005), but the effect of quantity (percent removed) was not (mean 0.08). In fact, the effect of cost was significantly greater than each of the other two effects, although this was not at issue. The important result is simply that cost did affect WTP, quite substantially. Subjects used their past expenditures as a guide to their WTP, even when it introduced inconsistency across adjacent cases in their WTP per case of cancer prevented.

Study 4. Nonnumerical information, between subjects.

Studies 1-3 have made cost information explicit and have examined the effects of varying cost within subjects. These studies have shown that subjects attend to cost even when they know they are doing so. We might expect this knowledge to reduce the effect of cost, assuming that some subjects think that the only dimension relevant to their decision is the benefit. That is, if cost were implicit, it may be more influential. On the other hand, making costs explicit increases the salience of cost, possibly causing some subjects to attend to costs when they would not do so otherwise.
In Study 4, we made cost information implicit by describing only how goods are provided. Such description is typically included in CV surveys (e.g., Carson et al., 1992). We also presented only a single description of each good to each subject. No subject saw both cost levels of any good. To increase the sensitivity of the experiment, however, each subject evaluated six different goods, three in the high-cost version and three in the low-cost version. The general purpose of this study was to ask whether cost affects pricing under conditions somewhat more similar to those used in CV surveys. If anything, we provide less detail about the means of provision of the good than do many surveys.


The questionnaire presented six goods, listed in Table 4, in the same order to 138 subjects, solicited as in previous studies except that 34 were students at the University of Washington, Bothell (mostly employed adults over 25 years old). We wrote two versions of each good, which differed in the description of how the good was provided but which were identical in the description of the benefits. For example, the two versions of the first good both involved a reduction in the rate of violent crime on campus from 50 to 25 crimes per year, for 50,000 students. In the "high-cost" good, police increased from 50 to 200; in the low-cost good, the increase was from 50 to 100. Likewise: oil spills were reduced by "installing new, computerized, navigation equipment on oil tankers" (low-cost) or by "replacing the entire fleet of tankers with new ones, with double hulls" (high-cost); illegal immigration was reduced by a small increase and redeployment of immigration agents or a doubling of agents; extension of health insurance was achieved by paying the full cost of the insurance or by paying 25% (because of cost savings elsewhere); automobile exhaust pollution was reduced by a small filter or a new muffler system, each to be replaced each year; and bacteria were removed from water by installation of a new purification system or by an added step in an existing system.
About half of the subjects filled out a form of the questionnaire with odd-numbered items high-cost and even-numbered low-cost, and half filled out a form with the reverse assignment.
Table 4: Goods and geometric mean WTP, Study 4.
Good High-cost WTP Low-cost WTP
1. Prevent campus crime $247.15 $169.36
2. Prevent large oil spills $13.41 $9.86
3. Reduce illegal immigration $60.64 $27.66
4. Provide health insurance for uninsured $66.82 $50.60
5. Reduce respiratory disease from pollution $36.86 $23.81
6. Reduce illness from bacteria in water $34.85 $38.32


WTP was higher when implied cost was higher. Table 4 shows the geometric mean WTP responses for the high-cost and low-cost versions of each item. Recall that the odd-numbered high-cost and even-numbered low-cost responses come from half the subjects, and the rest of the responses from the other half.
WTP responses were higher for all but one of the high-cost items than they were for the low-cost items. To test this statistically, for each item, we converted the WTPs to ranks across subjects, including both high- and low-cost versions of the item. (Each subject saw one version of each item.) For each subject, we subtracted the mean rank of that subject's three low-cost items from the mean rank of that subject's three high-cost items. The differences were significantly greater than zero (t=2.41, p=.009 one tailed), indicating that subjects were offering relatively more for the high-cost items. (The difference was also significant, p=.003, when we used the difference of the geometric means, but this method excluded the 25% of WTP responses that were zero. The number of these responses did not depend on cost.)
Items 1 and 4 provided quantitative information from which the effect size of cost might be inferred: doubling vs. quadrupling the police force; and paying the full cost vs. 25% of health insurance. The effect sizes, using the method of previous studies were 0.55 and 0.20 for these two items, respectively. (That is, these were the log ratios of the geometric means of the two conditions, respectively, divided by log(2) and log(4), respectively.) These effects are on the same order as the within-subject effects.
In sum, sensitivity to cost is present even when cost information is not itself provided in monetary form and when the manipulation of cost is not apparent to the subjects.


Cost information affected WTP when it took the form of estimated cost (Studies 1-3) or when it was simply implied by past expenditures (Study 3) or by descriptions of how a good would be provided (Study 4). Cost affected WTP both when each subject judged cases varying only in cost (Studies 1-3) and when each subject judged only a single cost version of each case (Study 4).
The results can be understood as overextension of a useful heuristic of relying on costs. This heuristic is useful for several reasons. First, cost does typically provide information about value. When we are not sure how much to pay for something, the willingness of others to pay is a good guide to value. When buying stocks, most investors have little else to go on. Only a few people need information about true value in order for markets to work efficiently. Second, the stated cost of something is a good guide to the willingness of others to pay. Contractors don't bother to make bids if they think that their bid has no chance of being accepted. So cost is also a guide to the opinions of others. Third, it may be unfair to pay less than the cost plus a fair profit.
Heuristics, however, take on a life of their own (Baron, 1994). People come to think of heuristics as useful guides even when they forget the justifications of the heuristics in terms of their original purposes. In particular, people attend to cost even when they have all relevant information about benefit, so that they no longer need cost information as a guide to benefit. In Study 1, even those subjects who acknowledged the sufficiency of the benefit information we provided were also influenced by costs. Moreover, given the description of how a decision would be made, it was unnecessary for subjects to take into account the desires of others, since these would be taken into account by the decision procedure. Finally, there was no reason to believe that the contractors would accept less than a fair price for the work required in each case. Although subjects may have ignored many of these specifications, they may also do so in other surveys. The heuristic of attending to cost may even prevent subjects from thinking about such issues.
In view of our findings, it is somewhat disturbing that many CV surveys (e.g., Carson et al., 1992) spend considerable time explaining the physical means of providing the good (e.g., escort ships) as well as the effects of providing it (e.g., reduced deaths of animals). Our findings suggest that attempts to measure economic value of public goods might try to eliminate information from which costs could be inferred. If this were done, gaps in information about benefits (such as those discussed by Baron, 1995) might become more apparent both to respondents and researchers.
The use of the cost heuristic might affect real decisions as well as hypothetical ones, so that the problem of valuation cannot necessarily be solved by observations of behavior in real markets. For example, before the development of contingent valuation methods, researchers relied upon travel-cost methods to establish values of public resources. Travel-cost methods looked at real market transactions, or foregone opportunities for real market transactions. Thus, to evaluate a particular national park, a researcher would survey park patrons, asking them how much they were spending on hotels, meals, gasoline, and other travel expenses. These data were assumed to reveal the value of the park, because it was assumed that these purchase decision were based on perceptions of the value of the park.
If market transactions reflect only tradeoffs between prices and benefits, then this approach of valuing visits to parks would be sensible. However, market researchers have amply demonstrated that market transactions are influenced by the sort of cost considerations explored in the research reported here. Studies of supermarket sales have shown that shoppers are less likely to buy an item at a particular price if they believe that the usual price was lower (Winer, 1986; Mayhew & Winer, 1992; Putler, 1992; Rajendran & Tellis 1994). Rather than considering only the benefit they can derive from an item, shoppers consider how the item's current price compares to its usual price. This effect of consumers' conceptions of usual prices (termed "reference prices") has been found to influence travel decisions as well: Crompton and Love (1994) found that people tend to avoid a festival if the cost of attending appears to be higher than the cost of attending other festivals. This suggests that reference prices contaminate the data gathered in travel-cost methods.
In the long run, with repeated decisions, use of the cost-heuristic might be overcome. People may discover that paying the "going rate" or "fair price" for something does not bring the satisfaction of other expenditures, so they may stop paying it. Similarly, they may find that sometimes they are willing to pay more than the fair price for something they really want. On the other hand, we know of no evidence that people learn to overcome their biases against prices that seem unusually high.
Measurement of the value of public goods, however, almost necessarily prevents such adaptation. We have found that subjects are prone to rely on costs in stating willingness to pay. Perhaps the only way to avoid this is to focus as much as possible on the benefits to be provided rather than the means of providing them.


Baron, J. (1994). Nonconsequentialist decisions (with commentary and reply). Behavioral and Brain Sciences, 17, 1-42.
Baron, J. (1995). Rationality and invariance: Response to Schuman. In D. J. Bjornstad & J. R. Kahn (Eds.) The contingent valuation of environmental resources: methodological issues and research needs, pp. 145-163. London: Edward Elgar.
Baron, J., & Greene, J. (in press). Determinants of insensitivity to quantity in valuation of public goods: contribution, warm glow, budget constraints, availability, and prominence. Journal of Experimental Psychology: Applied.
Carson, R. T., Mitchell, R. C., Hanemann, W. M., Kopp, R. J., Presser, S., & Ruud, P. A. (1992). A contingent valuation study of lost passive use values resulting from the Exxon Valdez oil spill. Natural Resource Damage Assessment, Inc.
Crompton, J. L., & Love, L. L. (1994). Using inferential evidence to determine likely reaction to a price increase at a festival. Journal of Travel Research, 32, 32-36.
Green, D. P., Kahneman, D., & Kunreuther, H. (1994). How the scope and method of public funding affects willingness to pay for public goods. Public Opinion Quarterly, 58, 49-67.
Jones-Lee, M. W. (1989). The economics of safety and physical risk. Oxford: Basil Blackwell.
Mayhew, G. E., & Winer, R. (1992). An empirical analysis of internal and external reference prices using scanner data. Journal of Consumer Research, 19, 62-70.
McFadden, D. (1994). Contingent valuation and social choice. American Journal of Agricultural Economics, 76, 689-708.
Putler, D. S. (1992). Incorporating reference price effects into a theory of consumer choice. Marketing Science, 11, 287-309.
Rajendran, K. N. & Tellis, G. J. (1994). Contextual and temporal components of reference price. Journal of Marketing, 58, 22-34.
Schkade, D. A., & Payne, J. W. (1994). How people respond to contingent valuation questions: A verbal protocol analysis of willingness to pay for an environmental regulation. Journal of Environmental Economics and Management, 26, 88-109.
Winer, R. S. (1986). A reference price model of brand choice for frequently purchased products. Journal of Consumer Research, 13, 250-256.


1This research was supported by N.S.F. grant SBR92-23015. Send correspondence to Jonathan Baron, Department of Psychology, University of Pennsylvania, 3815 Walnut St., Philadelphia, PA 19104-6196, or (e-mail)

File translated from TEX by TTH, version 3.59.
On 12 May 2005, 07:04.