Using ex ante approaches to obtain credible signals for value in contingent markets: evidence from the field.

(6) Carson, Groves, and Machina (2000) suggest that the use of two or more prices in value elicitation could (i) imply uncertainty of price, (ii) imply a willingness to bargain on behalf of the seller, or (iii) induce a perceived change in quantity/quality of the good. Price uncertainty would decrease the median or mean valuation (from the second question) for risk-averse agents, while the direction of change for the latter cases depends upon the response to the initial question. In the discussion of Carson, Groves, and Machina, however, the magnitude of the second price is always conditional on the initial response. Those who reply "No" are offered a lower price, while those who indicate "Yes" are asked a higher price. In all of these cases, the introduction of a second price signals that something else could be going on--the transaction involves more than is apparent at face value. Our price sequences, in contrast, are a design parameter that is purely exogenous. The sequence of prices is not conditional upon the response of the subjects, and the prices are offered aloud, for all to hear. Moreover, in cases of a potential real payment, it is made clear that the binding price will be determined randomly. These attributes of our experiment could attenuate any strategic responses of our subjects.

(7) Cheap talk treatment, pricing sequence A p-value = 0.0313; cheap talk treatment, pricing sequence B p-value = 0.0078; real treatment, pricing sequence A p-value = 0.0313; real treatment, pricing sequence B p-value = 0.0625.

(8) Consequential treatment, pricing sequence A p-value = 0.0313; consequential treatment, pricing sequence B p-value = 0.25.

(9) Hypothetical treatment, pricing sequence A p-value = 0.1563; hypothetical treatment, pricing sequence B, p-value = 0.1641.

(10) Our results are similar to those of Lusk (2003), in which responses derived from an elicitation mechanism that utilized cheap talk exhibited more responsiveness to price than those without cheap talk (i.e., hypothetical data). We find that both ex ante methods exhibited more responsiveness to variation in price than hypothetical data.

(11) [chi square] values are 3.9190 (p-value = 0.0477) and 6.3441 (p-value = 0.0118), respectively.

(12) Consider first other comparisons of A:$10, B:$5. [chi square] values are 1.7911 (p-value = 0.1808) for hypothetical treatment and 1.3371 (p-value = 0.2475) for consequential treatment. Consider next comparisons of A:$5, B:$10. [chi square] values are 0.4637 (p-value = 0.4959) for hypothetical, 2.1588 (p-value = 0.1455) for cheap talk, 0.9433 (p - value = 0.3314) for consequential, and 0.1383 (p-value = 0.7100) for real.

(13) An anonymous reviewer points out that an implication of the results of ALP is that demand for price changes derived within subjects will be more elastic than that derived between subjects. Visual inspection of the data and the statistical results, in general, does not lend support to this hypothesis, but this clearly depends on the parameters and experimental design.

(14) [chi square] values for the $5 price level are 0.04 (p = 0.82), 0.01 (p = 0.93), 0.13 (p = 0.71), and 0.94 (p = 0.33) for the hypothetical, cheap talk, consequential and real treatments, respectively (all df = 1). Corresponding [chi square] statistics for the $10 price level are 0.08 (p = 0.77), 0.14 (p = 0.69), 0.33 (p = 0.56), and 1.9 (p = 0.16) (all df = 1).

(15) We also conducted all analyses using only the first responses (i.e., the $5 responses from pricing sequence A and the $10 responses from pricing sequence B); our primary conclusions do not change.

(16) Percentage of "Yes" votes in the $10 real treatment = 18.8%; percentage of "Yes" votes in the $10 consequential treatment = 20.3%: percentage of "Yes" votes in the $10 cheap talk treatment = 29%; percentage of "Yes" votes in the $10 hypothetical treatment = 75%.

(17) [chi square] (df = 1) statistics for the hypothetical versus cheap talk, consequential, and real treatments are 20.9802, 34.6348, and 35.0672, respectively, for the $5 price level. For the $10 price level, the [chi square] (df = 1) statistics are 28.1351, 36.7114, and 40.6588 for the same sequence of tests. All p-values are below 0.0001.

(18) Using an F-test, we cannot reject the hypotheses Var([WTP.sub.consequential]) = Var([WTP.sub.real]) (F = 1.1285, p-value = 0.3186), Var([WTP.sub.hypothetical]) = Var([WTP.sub.real]) (F = 1.2283, p-value = 0.2084), or Var([WTP.sub.cheaptalk]) = Var([WTP.sub.real]) (F = 1.0759, p-value = 0.3853).

(19) A natural question concerning our consequentialism results is why they are different from Cummings and Taylor (1998), who report that treatments utilizing low levels of probability (p [less than or equal to] 0.5) produce results not in accord with a binding referendum (p = 1.0), but voting behavior associated with higher probability levels (p = 0.75) cannot be distinguished from that of a binding referendum. This remains an open empirical question, as Harrison and List (2004) point out when making a similar comparison to motivate the use of field experiments and what might cause differences between the lab and the field: "To provide a direct example of the type of problem that motivated us, when List [2001] obtains results in a field experiment that differ from the counterpart lab experiments of Cummings, Harrison, and Osborne [1995] and Cummings and Taylor [1999], what explains the difference? Is it the use of data from a particular market whose participants have selected into the market instead of student subjects, the use of subjects with experience in related tasks, the use of private sports-cards as the underlying commodity instead of an environmental public good, the use of streamlined instructions, the less intrusive experimental methods, mundane experimenter effects, or is it some combination of these and similar differences?"

The authors are, respectively, Assistant Professor, Department of Economics, East Carolina University and Professor, Department of Economies, University of Chicago and NBER. Thanks to the Editor and three anonymous reviewers who provided comments that Improved the paper. Glenn Harrison, Jason Shogren, and Laura Taylor also provided comments throughout the research process. Table 1. Experimental Design-Subjects by Treatment and Price Sequence Treatment Hypothetical Cheap Talk Consequential Real A: $5/$10 30 32 29 33 B: $10/$5 34 37 30 31 Table 2. Voting Behavior by Treatment Treatment Hypothetical Cheap Talk Pricing sequence A B A B First offer price $5 $10 $5 $10 Second offer price $10 $5 $10 $5 Subjects (n) 30 34 32 37

25 26 15 10 First (0.83) (0.76) (0.47) (0.27) Yes 22 29 10 17 Second (0.73) (0.85) (0.31) (0.46) Pooled n 64 69 Pooled Yes $5 54 32

(0.84) (0.46) Pooled yes $10 48 20

(0.75) (0.29) Treatment Consequential Real Pricing sequence A B A B First offer price $5 $10 $5 $10 Second offer price $10 $5 $10 $5 Subjects (n) 29 30 33 31

10 7 9 8 First (0.34) (0.23) (0.27) (0.26) Yes 5 9 4 12 Second (0.17) (0.30) (0.12) (0.39) Pooled n 59 64 Pooled Yes $5 19 21

(0.32) (0.33) Pooled yes $10 12 12

(0.20) (0.19) Note: Proportions are indicated in parentheses. Table 3. Experimental Statistics for Pair-Wise Comparisons across Treatments Treatment Hypothetical Cheap Talk Hypothetical -- 20.9802 (0.0000) Cheap talk 28.1351 (0.0000) -- Consequential 36.7114 (0.0000) 1.2682 (0.2601) Real 40.6588 (0.0000) 1.9038 (0.1677)

$10 Price level $10 Price level Treatment Consequential Real Hypothetical 34.6348 (0.0000) 35.0672 (0.0000) $5 Price level Cheap talk 2.6656 (0.1025) 2.5486 (0.1104) $5 Price level Consequential -- 0.2594 (0.6105) $5 Price level Real 0.04934 (0.8215) --

$10 Price level Note: The upper triangle contains test statistics for the $5 price level; the lower triangle contains test statistics for the $10 price level. All df = 1, and p-values are in parentheses.