Procuring agencies in conservation auctions typically have more information about the ecosystem service (ES) quality of conservation actions than landowners, and can affect auction outcomes by controlling participants’ access to this information. Our induced-value laboratory auction experiment explores the impact of sellers’ access to ES-quality information on auction efficiency when the conservation action choice is endogenous to offer formation. We find that providing ES-quality information allows sellers to identify and submit higher-quality conservation actions, an effect that counteracts previously identified efficiency losses from information rents.

The information structure of conservation procurement auctions has attracted significant attention due to concerns about adverse selection and other inefficiencies, with a particular focus on how information asymmetries between sellers and the conservation buyer drive auction outcomes. While the Conservation Reserve Program in the United States is the most well-known example, conservation auctions have been implemented by government agencies in a number of countries, including Australia (Hajkowicz et al. 2008; Stoneham et al. 2003), England (Short et al. 2000), and Germany (Ulber et al. 2010). A key characteristic of the conservation setting is that landowners’ privately-known costs of implementing conservation actions enable them to extract payments that are greater than their reservation rate (Arnold, Duke, and Messer 2013). Under certain conditions, auctions have theoretically been shown to reduce these information rents by motivating the honest revelation of participants’ opportunity costs (McAfee and McMillan 1987; Vickrey 1961), thereby allowing for the enrollment of more land into conservation for a given budget. However, field evidence from conservation procurement auctions suggests that landowner rents may comprise a significant portion of conservation expenditures (Ulber et al. 2011; Kirwan, Lubowski, and Roberts 2005).

From the perspective of a conservation agency seeking to enhance social welfare, socially efficient conservation procurement rests on the value of the provided public good exceeding the provision cost. Therefore, from a social welfare perspective, seller rents are a distributional consideration, not an efficiency consideration (Claassen, Cattaneo, and Johansson 2008). However, despite methodological advancements (e.g., Johnson et al. 2016; Guerry et al. 2015; Kareiva et al. 2011) there are numerous practical challenges to the quantification and valuation of a wide range of ecosystem services. As a result, ecosystem service values based on well-defined techniques (e.g., Conte 2013) are rarely available for use as a program acceptance criterion, especially when conservation provides multiple benefits. For example, the Conservation Reserve Program uses an environmental benefits index that serves as a proxy to rank the quality of offers—this measure does not reveal net social welfare. This practical approach, that is, minimizing procurement costs of an ecosystem service benefit index, can provide some measure of assurance that conservation dollars are being well-spent in lieu of net social welfare measures. This is an important consideration for the success of these programs as they face limited budgets and scrutiny over their cost-effectiveness (Hellerstein 2010). Given the scale of expenditures in some of these programs, increasing the cost-effectiveness of conservation procurement auctions could have a meaningful impact on ecosystem-service provision: as an example, the Conservation Reserve Program uses an auction process that results in almost $2 billion of conservation payments annually (Hellerstein, Higgins, and Roberts 2015).

A promising recent research path has focused on the role of environmental quality information and how it can be used to the buyer’s advantage in designing an auction to increase ecosystem service provision under a limited budget. A conservation-procuring agency may have an informational advantage in that it has the resources and ability to assess the quality of conservation activities across different parcels of land (Glebe 2013; Stoneham et al. 2003). In contrast, landowners may have relatively less knowledge about the landscape-scale functional processes that generate ecosystem services and the techniques used to estimate the value of these services. In conservation procurement auction experiments in which the assessed environmental quality is an exogenous attribute of a submitted offer, withholding this information has been shown to improve auction efficiency by preventing high offers from landowners with high-quality submissions (Banerjee, Kwasnica, and Shortle 2014Cason, Gangadharan, and Duke 2003).1 Within a similar theoretical framework, Glebe (2013) concludes that withholding information about quality can increase the cost-effectiveness of auctions in which there is a fixed number of participants.

Here, we build on prior research investigating manipulations of the information structure of a field-relevant auction design similar to that used by the Conservation Reserve Program. In particular, we assess the implications of quality information revelation in an auction setting where sellers can select both their price and their conservation activity. By treating both price and quality as choice variables in the formation of offers, this design more closely reflects conditions in the field, where landholders submit one offer and are given significant latitude in selecting what land and conservation practices they submit for consideration (Claassen, Cattaneo, and Johansson 2008). This design violates many assumptions of the standard benchmark auction model (McAfee and McMillan 1987) and extends significantly beyond prior analytical evaluations of conservation auctions, so we employ an experimental evaluation of the main information treatment effect.

We find that providing auction participants with detailed information regarding the quality of their conservation choices leads to improved auction performance, a contrary finding to related studies investigating exogenous quality auctions (Banerjee, Kwasnica, and Shortle 2014Cason, Gangadharan, and Duke 2003). We attribute this result to participants’ inability to identify and submit high-quality conservation actions when quality information is withheld. As a result, submissions are based solely on cost considerations, with effectively random quality. When quality information is withheld in our auction design, the subsequently lower information rents are more than offset by lower-quality conservation submissions, yielding an overall decrease in efficiency. More generally, these results demonstrate that (a) information restriction is not a panacea for cost-effective conservation procurement when quality is a choice variable, and (b) efficient auction design must account for both of these countervailing effects on efficiency.

Auction Background and Experimental Design

Conservation auctions are a type of procurement auction, a broad category that covers all auction variants with a single buyer and multiple potential sellers. After a brief overview of procurement auctions, we present the details of the experimental design.

Procurement Auctions

While the simplest implementation of a procurement auction includes price as the only relevant attribute of the good for sale, in practice there are very few instances of procurement where price is the only consideration (Rezende 2009). In the conservation context, the ecosystem services generated from a parcel of land are an important factor in determining the social welfare contribution of conservation expenditures. Contributions in the areas of auction and mechanism design have significantly improved our understanding of how to promote efficient procurement outcomes when both quality and price matter. A key finding of theoretical models exploring the optimal design of two-dimensional (quality and price) procurement auctions (e.g., Asker and Cantillon 2010; Asker and Cantillon 2008; Branco 1997; Che 1993) is that an optimal scoring rule does not necessarily reflect the buyer’s actual preferences—quality should be weighted less than its inherent value to the buyer. While these theoretical models are derived under conditions that typically do not reflect the conservation context, this finding makes it clear that auction outcomes can be influenced by taking advantage of private information in formulating the rules of the auction.2

In reaction to the limited insights into policy-relevant auction performance provided by theoretical guidance, recent research has employed simplified modeling approaches, reduced form empirical analyses of field data, and induced-value laboratory experiments to characterize auction performance under different design conditions.3 Laboratory auction experiments have proven valuable in the design of high-value auctions when theoretical guidance is incomplete or too reductive to capture the salient features of economic phenomena, as in the case of bandwidth allocation (Banks et al. 2003) and pollution credits (Friesen and Gangadharan 2013). Induced-value laboratory experiments are particularly useful in auctions with asymmetric information, as all parameters are known to the researcher, which is an inherent problem of the field setting. Experiments are commonly used to examine the impact of changes in auction design as a complement to game-theoretic approaches (Roth 2002), and have contributed significant insights to the field design of conservation auctions (Hellerstein 2010).

Relevant to our auction design, Schilizzi and Latacz-Lohmann (2007) and Cason and Gangadharan (2005) find that individualized-price (discriminatory) procurement auctions are typically preferred over fixed-payment programs for cost-effective conservation, though Schilizzi and Latacz-Lohmann (2007) demonstrate that repetition erodes the efficiency of the allocation—a similar finding to Rolfe, Windle, and McCosker (2009) in a field experiment setting. Beyond auction pricing rules, experimental research has provided insight into relaxing common assumptions about environmental quality, principally on the fronts of spatially-dependent environmental quality (Banerjee, Kwasnica, and Shortle  (2014); Reeson et al. (2011)), and, to a lesser extent, by treating environmental quality as an endogenous choice variable in the auction (Hellerstein and Higgins 2010).4 Lastly, several studies have examined the effect of changing information assumptions on auction efficiency when quality is exogenously determined (Banerjee, Kwasnica, and Shortle 2014Cason, Gangadharan, and Duke  2003; Haruvy and Katok 2013). These studies all assess auctions under significantly different auction design assumptions; however, a common theme is that when additional information about quality is provided to sellers, auction efficiency is reduced due to sellers leveraging their knowledge to extract rents from the buyer.

Our study examines the impact of withholding quality information from sellers on auction efficiency using information treatments and parameters similar to those of Cason, Gangadharan, and Duke (2003). The quality information treatment in Cason, Gangadharan, and Duke (2003) is conducted in the context of a multi-round auction where participants submit prices on three conservation actions, all of which are evaluated in the auction. We adopt a similar design, with two main modifications that are more consistent with field conservation auctions such as the Conservation Reserve Program (Hellerstein, Higgins, and Roberts 2015): we employ a single-round auction format and only allow participants to submit one of three potential conservation actions.5 These modest changes significantly alter the information available to participants in a way that also interacts with the quality information treatment. In a single-round auction, participants cannot conditionally update their beliefs about the comparative strength of their score because they do not have access to the information signals provided by provisionally-accepted offers in a multi-round auction. This uncertainty is compounded when quality becomes a choice variable in the auction, as participants cannot determine whether they are submitting low-quality conservation actions. The impact of access to quality information in our more field-representative setting is unresolved.

Experimental Design

Our induced-value lab experiment used a within-subject design to explore how withholding information about the quality of conservation actions from sellers impacts auction performance and participant behavior. Participants played the role of landowners reacting to the incentive structure of a conservation procurement auction in exchange for cash payouts based on their performance. Bidders competed against each other by submitting offers composed of: (a) a conservation action and (b) a corresponding price at which they would be willing to undertake that action. A selected action’s quality and the offered price were countervailing acceptance criteria: higher-quality and lower-cost offers were most preferred by the buyer. To maximize earnings, and therefore their payout for the session, sellers needed to judiciously balance their asking price against the probability of their offer being accepted.

Twelve experimental sessions, each with twelve participants, were conducted at Fordham University in the spring of 2014. The 144 unique participants were undergraduates recruited from economics classes and previous laboratory experiments at Fordham. At a conversion rate of 120 experimental dollars to one U.S. dollar, the mean payment was $31 per participant, including a fixed $10 show-up payment for sessions that lasted approximately 90 minutes.6 This computerized auction was implemented using z-tree (Fischbacher 2007), automating both the offer-submission process and the auction outcomes.

Sessions were designed to assess the main treatment effect of how participants respond to varying levels of information about the quality of conservation actions in the context of a procurement auction in which quality is endogenous. We varied how this information was presented to participants across treatments, while keeping all other factors constant. In each auction period, participants were presented with three potential conservation actions to choose from, each represented as a colored item for sale, abstracting the decision context from the conservation procurement setting. Actions each had a private cost that was known only to the seller, and an ecosystem service provision level, represented as a quality index, that was known to the buyer and may or may not have been known with certainty by the seller, depending on the information treatment.7

In the Quality Value treatment, quality information for each conservation action was displayed directly to the sellers. In the Quality Rank treatment, the quality values for each conservation action were not known with certainty; instead, sellers saw a numbered ranking of their three options, based on their relative quality values. This treatment was included to approximate the information that landowners may rely on in the absence of provided quality information. While the nominal quality values may not be known with certainty, landowners may have a sense of the relative quality of each choice based on prior auction experience or practical experience in the field. In the No Quality treatment, sellers were not provided with any information about the quality of their conservation actions.8

Sessions began with the experimenter reading aloud the instructions (see appendix), which emphasized that there should be no communication throughout the experiment. Following the instructions, participants completed a well-established exercise used to elicit participants’ risk preferences (Holt and Laury 2002), based on their preferences over a sequence of paired lotteries that were presented to minimize inconsistent responses (Garboua et al. 2012). This exercise was included because behavior in laboratory experiments has been shown to depend on participants’ attitudes toward risk (e.g., Goeree, Holt, and Palfrey  2002Kagel and Roth 2002).

Next, participants engaged in ten practice auction periods (under the Quality Value information setting), to become familiarized with the software interface and the basic auction design. The cost and quality distributions used in the practice periods differed from those used for data collection, which was noted in the instructions. After the practice periods, the four auction treatments were conducted. An auction period began with all participants observing private information about their three potential conservation actions. Based on this information, the auction proceeded with participants selecting one of the conservation actions and an accompanying price. The submitted offers were then ranked based on the score of the selected conservation action, with score defined as quality/price.9 Finally, using a discriminatory pricing rule, offers were accepted in decreasing order of their score until the price of the marginal offer exceeded the conservation budget.10 Individual net earnings for an auction period were the difference between the price and the conservation cost for accepted offers, and zero otherwise.

Participants were not made aware of the distributions from which cost and quality parameters were drawn. The budget in each auction period was 4,500 experimental dollars, a figure that was constant across treatments, but unknown to participants. The only information provided to participants after each period was whether or not their offer was accepted and an updated personal cumulative earnings total. Messer, Duke, and Lynch (2014) indicate that seller rents are sensitive to restricting information on the budget level and prior auction outcomes in an experimental discriminatory land procurement auction, although the treatments considered in that study do not provide a clear directional conclusion about the optimal information-revelation strategy for our study; therefore, we appropriated the information setting utilized by Cason, Gangadharan, and Duke (2003).

With the goal of providing general policy guidance for a common environmental procurement auction format, the parameters in our experiment were not linked to a specific field context. We chose to proceed with independent cost and quality distributions, which was made clear to participants in the auction instructions. Evidence from the field suggests that the environmental benefits of retiring land from production and the costs of doing so may be negatively correlated, positively correlated, or uncorrelated (Babcock et al. 1996; Heimlich 1989), while Cattaneo et al. (2005) show that each of these three relationships is possible when considering the environmental benefits and the costs of actions undertaken on working lands to provide these benefits. Our random endowment generation process for cost and quality draws and the process of creating cost heterogeneity are based on approaches used in Cason, Gangadharan, and Duke (2003) and Hellerstein and Higgins (2010) to ensure that the results here are readily comparable to the broader literature.

All costs and prices were denominated in experimental dollars. Each cost draw cij for player i and conservation choice j was drawn from a uniform distribution on support (500, 1,000) and each quality draw, qij was drawn from a uniform distribution on support (50, 100).11 For each period, one of the conservation actions was randomly selected to receive a cost discount that varied across players. For this conservation action, one-third of the players received a discount off of their initial conservation cost of 250, one-third received a discount of 125, and the remaining one-third received no discount.12 Participants were not explicitly informed about the cost discounts applied to their conservation actions—each observed only the final net cost of their conservation action when they received information about their endowments at the beginning of the period.

All treatments were featured once in each session, and no person was allowed to participate in more than one session. While this within-subject design increases the sample size and minimizes the error variance associated with participant heterogeneity, it comes at the cost of potential carryover effects between treatments (Charness, Gneezy, and Kuhn 2012). Although participants were not given information about the distribution of parameters or how they might vary, participants may have been able to infer this based on the order of treatments because the parameterization did not change across treatments. However, because the treatment effect does not pertain to knowledge about the distribution of potential quality realizations, but on the random realization itself, this potential sequencing effect is not expected to bias results.13 Consequently, we did not counterbalance the treatment sequencing across sessions, instead employing a randomization of treatment sequencing that approximated a counterbalanced design due to concerns about recruiting enough participants for our desired design.

Results

The treatment effect in this experiment is intended to assess the impact of withholding quality information on auction efficiency when conservation quality is a choice variable. The impact on auction performance is reported first, and then the impact of access to quality information on bidder behavior is reported.

Auction Performance

In procurement auctions where the offered good has only one valued attribute, it is straightforward to calculate how successful the auction was for the procurer based on whether the least-costly projects were selected. In our procurement auction, cost and environmental quality are both valued attributes. Therefore, a different metric is needed to assess auction efficiency. An appropriate measure of auction efficiency in this setting is how much quality is purchased per dollar spent, relative to the optimal quality-per-dollar ratio. This metric, the percentage of optimal cost-effectiveness ratio (POCER), has an advantage over total procured quality for measuring efficiency because the budget is rarely fully exhausted in auctions for discrete conservation activities, even when it is constant across successive auctions. Subsequently, different expenditures for the optimal set versus the purchased set will yield inconsistent comparisons. Procured quality normalized by expenditures provides a common basis for comparison.

The acceptance algorithm in the auction ranks offers by score and then accepts them successively until the price of the marginal offer exceeds the conservation budget.14 This approach maximizes quality/cost instead of total quality, subject to a budget constraint. Therefore, the optimal benchmark is constructed in the same way, ranking the endowed scores (qij/cij) of all conservation choices and selecting those with the highest scores (maximum of one per participant) iteratively until the next selection would exceed the budget.15 This yields a consistent percentage measure of the realized versus potential cost-effectiveness of quality procurement for evaluating our treatment effect.

Figure 1 provides a qualitative summary of the cost-effectiveness of the procured set of conservation actions across treatments and sessions. Visual inspection suggests that providing information about the quality of conservation actions generally increases the cost-effectiveness of procurement, based on the higher observed values for the Quality Value treatment overall (box plot in figure 1a) and in most sessions (scatter plot in figure 1b). To explore this relationship more formally, we specify three different regression models, including important covariates like participant experience (table 1).

Figure 1.

The percentage of optimal cost-effectiveness ratio (POCER) across treatments

Note: The box plot in figure 1a displays results for each treatment aggregated across all twelve sessions, showing that, on average, auction performance is strongest in the Quality Value treatment. The scatter plot in figure 1b shows auction performance for each treatment across sessions.

Figure 1.

The percentage of optimal cost-effectiveness ratio (POCER) across treatments

Note: The box plot in figure 1a displays results for each treatment aggregated across all twelve sessions, showing that, on average, auction performance is strongest in the Quality Value treatment. The scatter plot in figure 1b shows auction performance for each treatment across sessions.

Table 1.

Auction Performance—Percentage of Optimal Cost-Effectiveness Ratio

 Model 1 Model 2 Model 3 
Quality Rank Treatment −0.0268*** −0.0268*** −0.0541** 
(0.0068) (0.0069) (0.0193) 
No Quality Treatment −0.0355*** −0.0355*** −0.0224* 
(0.0056) (0.0056) (0.0119) 
Treatment Experience  0.0002 0.0006 
 (0.0005) (0.0008) 
Endowment  0.0004 −0.0004 
 (0.0008) (0.0007) 
Quality Rank x Experience   0.0028 
  (0.0018) 
No Quality x Experience   −0.0039** 
  (0.0013) 
Quality Rank x Endowment   0.0014 
  (0.0012) 
No Quality x Endowment   0.0019 
  (0.0019) 
Constant 0.8728*** 0.8691*** 0.8719*** 
(0.0031) (0.0067) (0.0075) 
Session Fixed Effects Yes Yes Yes 
Observations 432 432 432 
 Model 1 Model 2 Model 3 
Quality Rank Treatment −0.0268*** −0.0268*** −0.0541** 
(0.0068) (0.0069) (0.0193) 
No Quality Treatment −0.0355*** −0.0355*** −0.0224* 
(0.0056) (0.0056) (0.0119) 
Treatment Experience  0.0002 0.0006 
 (0.0005) (0.0008) 
Endowment  0.0004 −0.0004 
 (0.0008) (0.0007) 
Quality Rank x Experience   0.0028 
  (0.0018) 
No Quality x Experience   −0.0039** 
  (0.0013) 
Quality Rank x Endowment   0.0014 
  (0.0012) 
No Quality x Endowment   0.0019 
  (0.0019) 
Constant 0.8728*** 0.8691*** 0.8719*** 
(0.0031) (0.0067) (0.0075) 
Session Fixed Effects Yes Yes Yes 
Observations 432 432 432 

Note: The dependent variable is the percentage of the optimal cost-effectiveness ratio achieved. The unit of observation is an auction period. The Quality Value treatment is the base case. Robust standard errors clustered at the session level are reported in parentheses. One, two, and three asterisks indicate 10%, 5%, and 1% significance for a two-tailed hypothesis test based on a t distribution with 11 degrees of freedom, respectively. Wooldridge tests for serial correlation in panel data fail to reject the null hypothesis of no first-order autocorrelation in each model.

The estimated regressions assume a fixed-effects, session-level error structure to control for unobserved session-level heterogeneity arising from idiosyncratic experimental effects in each session (Fréchette 2012).16 The regressions are of the form  

(1)
(POCERtg=α+Xtgβ+Ztgγ+sg+νtg)
where t indexes auction periods within a session and g indexes sessions. Further, POCERtg represents our efficiency metric, the percentage of optimal cost-effectiveness ratio, and Xtg is composed of treatment indicator variables (the indicator for the Quality Value treatment is excluded from the regressions as a reference). The vector Ztg is composed of a variable denoting the period number within a given treatment, which measures participant experience, and interactions between this term and the treatment indicators.17

With the Quality Value treatment as the reference, model 1 shows that reducing the amount of quality information available results in a corresponding decrease in auction efficiency, as evidenced by the significant negative coefficients on the Quality Rank and No Quality treatment indicator variables. Relative to the performance of the auction when participants have full information about the quality of their suite of conservation actions, providing participants with only rank information about the quality of these actions reduces the cost-effectiveness of the auction by 2.7 percentage points, representing a loss in efficiency of 3.1% from the Quality Value treatment. Completely concealing quality information from auction participants reduces cost-effectiveness by 3.6 percentage points, or 4.1% relative to the Quality Value treatment. The specification in model 1 is identical to a Dunnett’s test for comparing multiple treatments to a control (Williams 1971), demonstrating significant differences in the mean efficiency of these treatments relative to the 87.3% mean efficiency observed in the Quality Value treatment. A pairwise comparison using Tukey’s HSD test supplements these results by confirming that efficiency is higher in the Quality Value treatment than in the other treatments; however, a significant difference is not observed between auction efficiency in the Quality Rank and No Quality treatments (p = 0.323).

Model 2 introduces an additional predictor for experience. Experience is quantified simply as the chronological order of a period in each treatment (1-12). The estimated effects of withholding quality information from auction participants on auction performance are nearly identical to those in model 1 - there are no observable experience effects.

Model 3 uses interactions between the explanatory variables in models 1 and 2 to explore the possibility that there are treatment-specific experience effects. Controlling for these effects reveals a wider gap in performance between the Quality Value treatment and the two reduced quality-information treatments. This specification indicates that the relative inefficiency of the No Quality treatment is partially due to decreasing auction efficiency over time. Specifically, every three additional periods in the No Quality treatment reduces the cost effectiveness of the auction by approximately 1.17 percentage points relative to the Quality Value treatment. In the 12th period, there is an additional 4.68 percentage point reduction in the percentage of optimal cost-effectiveness ratio for the No Quality treatment, leading to a cumulative reduction of 6.92 percentage points relative to the Quality Value treatment, while there is a constant 5.41 percentage point reduction for the Quality Rank treatment relative to the Quality Value treatment.

To further explore the reduction of efficiency in the No Quality treatment across auction periods, we decompose the time trend of efficiency across periods into its component parts—optimal quality per dollar and actual procured quality per dollar (figure 2; see online

for regression analysis). As cost and quality draws were constant across sessions for a given treatment and period, in each period there is only one unique “optimal quality per dollar” value per treatment, while there is one actual procured quality per dollar value for each of the twelve sessions. Clear time trends in the optimal quality per dollar across periods are visible that originate from the small sample: a decreasing trend in the Quality Rank treatment, an increasing trend in the No Quality treatment, and no trend in the Quality Value treatment. The slopes of the optimal quality per dollar and the actual quality per dollar are similar in sign across periods in the Quality Value and Quality Rank treatments, whereas the absence of a time trend in actual procured quality per dollar in the No Quality treatment does not match the positive time trend in optimal quality per dollar, resulting in decreasing efficiency over time.
Figure 2.

Time trends in optimal (grey filled circles) and actual procured (hollow circles) quality per dollar across periods for all sessions

Note: For each period, there are twelve hollow circles representing the result in each period for each session. Actual procured and optimal quality per dollar are the numerator and denominator, respectively, of the efficiency measure, POCER. The solid and dashed lines represent regressions fitting a time trend to the optimal and actual quality per dollar, respectively. Actual quality per dollar can exceed optimal in the case where accepted offers are below cost, a situation that occurred in one period in the No Quality treatment.

Figure 2.

Time trends in optimal (grey filled circles) and actual procured (hollow circles) quality per dollar across periods for all sessions

Note: For each period, there are twelve hollow circles representing the result in each period for each session. Actual procured and optimal quality per dollar are the numerator and denominator, respectively, of the efficiency measure, POCER. The solid and dashed lines represent regressions fitting a time trend to the optimal and actual quality per dollar, respectively. Actual quality per dollar can exceed optimal in the case where accepted offers are below cost, a situation that occurred in one period in the No Quality treatment.

Taken together, these results suggest that efficiency decreases with less information and that this effect may be more pronounced as the underlying unknown quality endowment changes through time. To further explore this outcome, we turn to descriptive statistics of auction outcomes in the next section, followed by regression models of seller behavior and offer formation.

Bidder Behavior

A closer look at descriptive statistics for all sessions reveals significant differences in the mean quality of submitted offers observed across treatments, with a notable and strictly decreasing level of offered quality as information decreases observed in both submitted and accepted offers (table 2).18 Offered quality in the Quality Value treatment is greater than that in the Quality Rank treatment (Tukey HSD, p < 0.001), and the offered quality in the Quality Rank treatment is greater than that in the No Quality treatment (p < 0.001).19 Prices across treatments form a less clear pattern as there is no significant difference between treatments.20

Table 2.

Descriptive Statistics

 Quality Value Quality Rank No Quality 
Mean quality 82.60 77.59 75.68 
 (0.32) (0.35) (0.35) 
Mean price 683.8 686.5 795.8 
 (3.6) (3.6) (74.6) 
Mean quality/price 0.1265 0.1191 0.1183 
 (0.001) (0.001) (0.0011) 
Mean profit 39.26 37.21 35.13 
 (1.50) (1.38) (1.25) 
Mean cost 613.5 618.9 608.6 
 (3.6) (3.6) (3.5) 
Observations 1728 1728 1728 
 Quality Value Quality Rank No Quality 
Mean quality 82.60 77.59 75.68 
 (0.32) (0.35) (0.35) 
Mean price 683.8 686.5 795.8 
 (3.6) (3.6) (74.6) 
Mean quality/price 0.1265 0.1191 0.1183 
 (0.001) (0.001) (0.0011) 
Mean profit 39.26 37.21 35.13 
 (1.50) (1.38) (1.25) 
Mean cost 613.5 618.9 608.6 
 (3.6) (3.6) (3.5) 
Observations 1728 1728 1728 

Note: All offers. Standard errors appear in parentheses.

The quality per experimental dollar offered across treatments illustrates the joint effect of these findings, with the mean value in the Quality Value treatment significantly higher than that observed in the other two treatments (p < 0.001 for both comparisons). Notably, there is no significant difference between the quality per experimental dollar in the Quality Rank treatment relative to that in the No Quality treatment. Also, as predicted by theory, an increase in the information available to participants is associated with larger profits, as reflected in the observed higher mean profit in the more informative treatments. However, while there is an ordering in these information rents established by the mean values across treatments, the only significant difference is between the Quality Value treatment and the No Quality treatment (p = 0.059). Taken together, the summary statistics across treatments demonstrate that the higher efficiency in the Quality Value treatment is likely due to the higher mean quality of submitted conservation actions.

Explaining the behavior that leads to an increase in auction efficiency with an increase in information is complicated by the nature of offer formation. The decision process for a participant in the auction consists of selecting a conservation action from the three potential choices and submitting a price for undertaking the chosen action. The order in which these two actions are performed is not restricted by the software interface, and decisions are based on both expectations of earnings and beliefs about the probability of an offer being accepted conditional on its score. In the absence of an analytical model for guidance on how to disentangle these endogenous decisions, reduced-form regression models are used to further explain price and conservation action selection.

The conservation action choice regression is a conditional logit model, also referred to as a fixed-effects logit model for panel data. Fixed effects are included to capture unobserved, participant-specific heterogeneity.21 Conservation action characteristics are used as predictors for the binary dependent variable yijt (1 if action is selected, zero otherwise). The model is  

(2)
(yijt=1[α+Xijtβ+ci+uijt0])
where uijt is a distributed extreme value conditional on Xijt and ci, i indexes experiment participants, j indexes conservation actions, and t indexes auction periods. Further, Xijt is composed of conservation action cost and quality, as well as indicator variables that take on a value of 1 if the action chosen is the minimum cost or the maximum quality of those available. This regression was run independently for each treatment. We cluster standard errors at the session level to allow for unobserved heteroskedasticity and serial correlation within each session.

The model results (table 3) report the marginal effect of each explanatory variable on the probability of selection evaluated at the mean of all explanatory variables under the assumption that the fixed effect is zero. Cost plays a significant role in which conservation choice is selected in all three treatments, and a higher cost leads to a lower probability of selection, ceteris paribus. Quality affects choice differently in the three treatments. In the Quality Value treatment, the continuous quality information is associated with a greater probability of selection when the submitted action has higher quality. In the Quality Rank treatment, participants do not observe the continuous quality values and so attention is focused on the action with the highest-ranked quality, evidenced by the 8.6 percentage point increase in the probability that the action with the highest-ranked quality (Maximum Quality) is selected. In the No Quality treatment, participants cannot condition their selection on ranked or continuous quality and therefore react entirely to the cost information, with the lowest-cost action acting as a focal point. In this treatment, an action is 22.6 percentage points more likely to be selected if it is the least-cost choice (Minimum Cost). These results indicate that by reducing the amount of information about quality, participants are forced to condition their conservation choice on a narrower set of characteristics that is less informative.

Table 3.

Behavior—Conservation Action Selection

 Quality Value Quality Rank No Quality 
Item Cost −0.0003*** −0.0002*** −0.0002*** 
 (0.0000) (0.0000) (0.0000) 
Item Quality 0.0024*** 0.0005 −0.0001 
 (0.0003) (0.0003) (0.0004) 
Minimum Cost 0.1311*** 0.1460*** 0.2256*** 
 (0.0115) (0.0170) (0.0241) 
Maximum Quality 0.2359*** 0.0862*** 0.0064 
 (0.0099) (0.0222) (0.0131) 
Participant Fixed Effects Yes Yes Yes 
Observations 5,184 5,184 5,184 
 Quality Value Quality Rank No Quality 
Item Cost −0.0003*** −0.0002*** −0.0002*** 
 (0.0000) (0.0000) (0.0000) 
Item Quality 0.0024*** 0.0005 −0.0001 
 (0.0003) (0.0003) (0.0004) 
Minimum Cost 0.1311*** 0.1460*** 0.2256*** 
 (0.0115) (0.0170) (0.0241) 
Maximum Quality 0.2359*** 0.0862*** 0.0064 
 (0.0099) (0.0222) (0.0131) 
Participant Fixed Effects Yes Yes Yes 
Observations 5,184 5,184 5,184 

Note: Results are based on conditional logit models with participant-fixed effects in which the dependent variable is an indicator variable denoting whether or not the action was selected. The unit of observation is a conservation action available to an auction participant in a single auction period. Standard errors clustered at the session level are reported in parentheses. One, two, and three asterisks indicate 10%, 5%, and 1% significance for a two-tailed hypothesis based on a t distribution with eleven degrees of freedom, respectively.

Determinants of the submitted price across conservation choices and participants are identified using a model with participant fixed-effects and standard errors clustered at the session level.22 The estimated model is specified as follows:  

(3)
(Priceit=α+Xitβ+Zitγ+ci+νit)
where i indexes experiment participants and t indexes auction periods. Further, Xit is comprised of submitted conservation characteristics including cost, quality, and minimum cost and maximum quality indicators, as given in the selection regression. The vector Zit is comprised of participant characteristics. These characteristics include the following: experience (which measures the number of auction periods in which the individual has participated in a given treatment and can vary from 0 to 11); cumulative profit, which measures the individual’s earnings from all previous auction periods; and an interaction term between a risk-aversion indicator variable and the cumulative profit variable.23 The cumulative profit variable and its interaction with the risk-aversion indicator variable are included to explore the relative importance of risk aversion and biased subjective probabilities of offer acceptance in determining auction behavior, an unresolved issue in the auction literature (Armantier and Treich 2009).

Table 4 presents the results of running the above regression model independently for each quality information treatment.24 Cost plays a significant role in determining prices across all three treatments, with an increase in price of approximately 0.90 experimental dollars for each 1 experimental dollar increase in cost of the selected conservation choice across treatments. While prices are found to significantly increase with quality in the Quality Value treatment, participants are unable to condition their offered price on their unobserved quality endowment in the No Quality and Quality Rank treatments. However, in the Quality Rank treatment, participants do react to their ranked quality information. The maximum quality choice of the three potential selections significantly increases the offered price as participants seek profit based on this uncertain signal of quality. In the No Quality treatment, cost is the sole focal characteristic. In addition to the positive observed relationship between cost and price, the average price in this treatment decreases by 17.06 experimental dollars if the selected action has the lowest cost of the three potential conservation actions. This is an interesting result indicating that, in the absence of quality information, participants are not simply basing their price on the nominal cost value but are also reacting to qualitative comparisons of cost between conservation choices. This nuanced view of offer formation is supported by the observation that 32.2% of the time, participants do not select the lowest-cost option in the No Quality treatment, an unexpected result given that cost and quality are drawn randomly and independently in our experiment.25

Table 4.

Behavior—Offered Price

 Quality Value Quality Rank No Quality 
Selected Item Cost 0.8904*** 0.9182*** 0.9087*** 
(0.0147) (0.0187) (0.0173) 
Selected Item Quality 0.8296*** −0.1146 0.1367 
(0.1904) (0.0971) (0.1482) 
Minimum Cost 1.1355 2.1769 −17.0604*** 
(3.4846) (3.1240) (5.1266) 
Maximum Quality 1.5954 11.5297*** 0.3498 
(3.2004) (2.5481) (4.6256) 
Period 0.6815 −0.7597 0.6259 
(0.9705) (0.8697) (1.3298) 
Cumulative Profit −0.0478 0.0203 −0.0054 
(0.0393) (0.0160) (0.0302) 
Risk Averse x Cumulative Profit 0.0281 −0.0171 0.0280 
(0.0344) (0.0167) (0.0175) 
Constant 68.5745*** 123.7729*** 120.2973*** 
(16.2220) (17.8997) (18.0501) 
Participant Fixed Effects Yes Yes Yes 
Observations 1,728 1,728 1,724 
 Quality Value Quality Rank No Quality 
Selected Item Cost 0.8904*** 0.9182*** 0.9087*** 
(0.0147) (0.0187) (0.0173) 
Selected Item Quality 0.8296*** −0.1146 0.1367 
(0.1904) (0.0971) (0.1482) 
Minimum Cost 1.1355 2.1769 −17.0604*** 
(3.4846) (3.1240) (5.1266) 
Maximum Quality 1.5954 11.5297*** 0.3498 
(3.2004) (2.5481) (4.6256) 
Period 0.6815 −0.7597 0.6259 
(0.9705) (0.8697) (1.3298) 
Cumulative Profit −0.0478 0.0203 −0.0054 
(0.0393) (0.0160) (0.0302) 
Risk Averse x Cumulative Profit 0.0281 −0.0171 0.0280 
(0.0344) (0.0167) (0.0175) 
Constant 68.5745*** 123.7729*** 120.2973*** 
(16.2220) (17.8997) (18.0501) 
Participant Fixed Effects Yes Yes Yes 
Observations 1,728 1,728 1,724 

Note: The dependent variable is the price submitted by an auction participant as a part of her offer. The unit of observation is an auction period. Robust standard errors clustered at the session level are reported in parentheses. The models include participant fixed effects. One, two, and three asterisks indicate 10%, 5%, and 1% significance for a two-tailed hypothesis test based on a t distribution with eleven degrees of freedom, respectively.

Participant characteristics are consistently found to have no impact on the offered price across treatments. Within our chosen participant-level fixed effects specification, none of the time-varying participant characteristics have an impact on price, and we are unable to contribute to the recent literature that subjective beliefs about offer acceptance may be an important determinant of auction behavior (Armantier and Treich 2009).

Discussion

Asymmetric information in procuring conservation raises both challenges and potential opportunities. While the cost of provision may only be known to landowners, a procuring agency typically has an informational advantage in that it sets the environmental quality scoring rule and conducts assessments of ecosystem service values of interest. Our study finds that a conservation buyer may be able to affect auction outcomes through the strategic control of this information about potential conservation actions, which is consistent with existing work. The important contribution of this article is the finding that withholding quality information introduces an efficiency-eroding identification effect based on sellers’ inability to identify high-quality conservation actions. The decrease in efficiency from this effect is only observable when quality is treated as a stochastic choice variable, and is generally opposed by a concomitant reduction in information rents.

The design of this experiment resulted in a competitive auction, as roughly half the offers were accepted in any given period. The level of competition impacts the rent-earning potential: all else being equal, the introduction of more sellers into our auction would lower prices as sellers compete to raise their scores. Fewer sellers (or a higher budget—see Messer, Duke, and Lynch 2014) may yield higher rents to sellers to such an extent that it dominates the positive efficiency effects of providing information. As such, while we observe a net increase in efficiency from providing quality information, our results should not be construed to mean that efficiency always increases with the provision of quality information when quality is a choice variable; rather, information revelation produces the countervailing efficiency effects of higher-quality submissions and more information rents. The Conservation Reserve Program accepted nearly 90% of offers in a recent year (USDA 2012), significantly reducing the incentives for offer competition and resulting in the use of bid caps to avoid overpayment. Addressing the effect of information revelation in a low-competition environment is a potential avenue for further research, given the importance of the Conservation Reserve Program auction in landscape conservation.

We chose to draw values for the cost and quality of the conservation actions from independent distributions, a decision that has been made previously in the literature (e.g., Banerjee, Kwasnica, and Shortle 2014Cason, Gangadharan, and Duke 2003), because the relationship between productivity (opportunity cost of conservation) and ecosystem service provision is not straightforward and may feature negative, positive, or no correlation, especially when multiple services are aggregated into a composite index. When cost and quality are uncorrelated random variables, offering higher quality is a costless strategy for improving the likelihood of offer acceptance in a competitive auction. Withholding quality information from auction participants necessarily makes cost the focal point of offer formation, a potential issue for conservation auctions if cost and quality are positively correlated random variables. If low-cost parcels are systematically biased toward a specific type of ecosystem service provision, or are associated with low service provision overall, then withholding quality information may provide an inferior portfolio of services. However, if sellers know or can infer the nature of the correlation between cost and quality through the auction process, they may be able to adjust their bidding strategy to field higher quality submissions. We leave this line of inquiry for future research. Ultimately, the identification effect demonstrated here is related to the degree of uncertainty regarding quality, not the nature of the correlation between cost and quality.

Participant experience did not play a large role in the outcome of these auctions overall, contrary to the findings of Schilizzi and Latacz-Lohmann (2007). The information environment in that experiment differed considerably from this study, however, in that participants knew the budget and had information about how their opportunity costs related to others’ via quartile information. Even our most information-rich treatment did not provide this information. In our efficiency analysis we show that less information makes it more difficult to adjust over time to exogenously changing market conditions introduced via the differential time trends in endowments across treatments (as reflected by the “optimal quality per dollar” metric). While we did not endeavor to test experience directly, our results and the comparison of our information environment against that of Schilizzi and Latacz-Lohmann (2007) suggest that more information provides sellers with an opportunity to extract more rents over time in discriminatory price, budget-constrained auctions.

Conclusion

Much of the previous research regarding conservation auction design has either focused on one-dimensional procurement auctions or auctions in which both cost and quality determine acceptance, but are exogenously endowed. Our focus on auction outcomes within the context of endogenous conservation action selection sheds light on a previously-unremarked-upon aspect of such actions: in the absence of information about the quality of given conservation actions, sellers may be unable to identify and submit valued conservation actions. In our experimental context, the magnitudes of the demonstrated efficiency gain (between five and seven percentage points) due to access to conservation action quality information would seem to be noteworthy to policy makers, particularly in this time of heightened calls for government austerity. For example, a 6.92 percentage-point increase in the efficiency of $200 million in expenditures in the Conservation Reserve Program’s 43rd general signup would have saved $13.84 million dollars for the same level of ecosystem service provision.

The inability to identify high-quality conservation actions can be resolved by the provision of such information, at the cost of increased information rents to sellers. In situations where there are a few well-defined conservation actions (or locations, which also influence quality) for each seller, a potential solution to this trade-off is to employ the bid menu approach shown in Cason, Gangadharan, and Duke (2003). By withholding quality but evaluating all submitted actions, this format strips the ability of sellers to condition their bids on quality and extract information rents, while overcoming the identification effect introduced in this article. The bid menu solution works to remove information rents and the identification effect in the set of offered actions; however, the potential for the bid menu approach to render quality exogenous to the bidding process is not certain as the number of discrete conservation options increases. Because preparing offers is costly, sellers may only select a subset of their potential options. There is a chance that this subset contains only low-quality conservation outcomes because sellers do not have quality information to direct their efforts in offer creation. Therefore, the practicality of this bid menu solution rests on the feasibility of sellers submitting all potential offers, or the buyer’s identification of high-quality actions for sellers, without introducing the potential for information rent. The buyer’s transaction costs associated with evaluating the ecosystem service quality for each of these potential actions across all sellers also will play a large role in the cost-effectiveness of using a bid menu for conservation procurement.

Supplementary Material

are available at American Journal of Agricultural Economics online.

1 Withholding information from auction participants is also demonstrated to increase procurement auction performance in a more general context (Haruvy and Katok 2013).
2 The “benchmark” auction model assumess that bidders are risk-neutral and symmetric, that their values are independent, and that the payment is a function of the cash bid alone (McAfee and McMillan 1987). Conservation procurement auctions generally consider parcels with differentiated quality and opportunity costs, based on asymmetric land endowments across landowners, which violate the assumption of symmetric bidders. Farmers, the primary landowner type typically participating in these auctions, are generally believed to be risk-averse (Chavas, Chambers, and Pops 2010), which introduces analytical challenges to optimal auction design (McAfee and McMillan 1987). Conservation procurement auctions are typically designed to accept multiple contracts under a budget constraint or acreage target that may be unknown to participants, removing a key piece of information that forms the basis of participants’ beliefs about their offer’s acceptance probability. Additionally, as in Cason, Gangadharan, and Duke (2003) and Banerjee, Kwasnica, and Shortle (2014), our study assumes that the scoring rule is non-linear in price and quality, whereas quasi-linearity provides analytical tractability for the theoretical studies mentioned. Taken together, capturing the salient features of the field context for conservation procurement extends beyond the bounds of tractable analytical modeling of Nash equilibrium bidding strategies and optimal auction design (Glebe 2013Schilizzi and Latacz-Lohmann 2007).
3 A fundamental information issue in theoretical modeling of conservation auctions is how participants form probability expectations of winning, conditional on their parcels’ quality and price. The complexity of the auction design, largely attributable to endowment asymmetry in a context where multiple parcels are accepted under an unknown budget, has led most modeling efforts to treat parcel quality and the probability of winning (conditional on score) as exogenously entering the objective function of the bidder (Jacobs, Thurman, and Marra 2014; Vukina et al.  2008; Kirwan, Lubowski, and Roberts 2005; Latacz-Lohmann and Van der Hamsvoort 1997). This departure from strategic Nash equilibrium modeling simplifies the decision process for the sake of tractability and is useful in retrospective empirical modeling or for projecting outcomes under the same auction rules. However, the predictions from these models are less applicable to studying changes in strategic behavior under alternative auction designs and information regimes with the goal of improving the allocation and efficiency of procurement.
4 Quality is introduced as a costly choice variable as a treatment in Hellerstein and Higgins (2010), with the main finding that competition drives bidders to incur costly quality improvements to improve their expected returns. Their study did not employ a quality information treatment.
5 We choose a single-round format here for consistency with the design of the Conservation Reserve Program, the largest conservation program of our format worldwide. However, recent work has shown that information revelation of prior single-round auction results can increase the rent premiums commanded by sellers in subsequent auctions (Messer et al. 2016), a situation that is more characteristic of the information-gathering process in a multi-round auction. This is an important avenue of work as conservation programs continue to explore how much information to release after the completion of an auction.
6 Conversion rates were established during pilot testing of the experiment. Note that a constant exchange rate between heterogeneously-endowed participants would typically leave some participants with less earning potential than others; however, participants rotated roles so that they each spent an equal number of periods as each type, thus equalizing earning potential.
7 Scientific limitations in monetizing the value of environmental services from parcels has commonly resulted in the use of scientifically informed indices of ecosystem service value (Ribaudo et al. 2001), an approach adopted here to value the quality of parcels.
8 A fourth treatment was included to assess the impact of heterogeneous bidder types on auction performance. This analysis will be presented in a separate manuscript.
9 The “buyer” is represented by this deterministic scoring and ranking algorithm that determines bid acceptance. Our choice of a scoring auction format is based on evidence that benefit-cost targeting selection mechanisms like the one used in our design tend to fare well from an efficiency perspective in the presence of positive correlation between costs and benefits, while negative correlation leads to no worse performance than other targeting procedures (Babcock et al. 1997).
10 We employed a budget-constrained versus a target-constrained (such as acreage) auction format based on prior findings that budget-constrained conservation procurement auctions are more robust to efficiency losses from repetition (Schilizzi and Latacz-Lohmann 2007).
11 As random cost and quality draws introduce variation that may confound hypothesis testing across treatments, twelve periods (one treatment worth) of cost and quality endowments for all participants were drawn and reused across treatments. These cost and quality endowments were rearranged between treatments by first reassigning them to new players within a period, and then reordering periods. This ensured that no participant ever held the same endowment twice in a session and that endowment sets within periods never occurred in the same order across treatments.
12 The additive shift in cost distributions captures important participant- and conservation-level cost asymmetries observed in field auctions (Vukina et al. 2008). This cost heterogeneity was also introduced to assess the extent to which participants gravitated towards a conservation option that provided positive private (or on-site) benefits, such as erosion prevention or pollination services. Hypotheses related to the impact of bidder heterogeneity and private benefits from conservation actions are explored in a manuscript in preparation titled “Private Benefits of Conservation and Procurement Auction Performance.” This issue is a critical concern regarding the true benefits of payment-for-ecosystem-services programs (Ferraro 2008) and is also relevant to the Conservation Reserve Program, which uses an Environmental Benefits Index that rewards actions with private benefits, such as improved agricultural outcomes due to reduced soil erosion, to determine the quality value of conservation actions.
13 display results for our regression analyses using a counterbalanced latin square comprised of four of the twelve experimental sessions (four, eight, eleven, and twelve), for evidence that our findings are robust to the presence of potential order effects.
14 The algorithm does not attempt to exhaust the budget by accepting inferior-score offers with prices that are less than the remaining budget.
15 Because this design results in variation in total expenditures per auction and there exists a dependency between total expenditures and offer variables, the potential exists for treatment comparisons to be confounded by the way information relates to the marginal acceptance criteria. However, the budget is constant, binding, and unknown to participants across all treatments, and can only be inferred via experience in the auction. A one-way ANOVA demonstrates no significant difference in mean expenditures across treatments (p = 0.692). Prices exhibit no correlation with experience across all treatments (table 4), while controlling for other attributes of the conservation choice and the individual. Together, these results provide compelling evidence that expenditures are stationary across treatments and have no systematic effect on prices across periods or treatments.
16Stock and Watson (2008) demonstrate that the traditional sandwich variance-covariance matrix is inconsistent in the fixed-effects context when there are more than two time periods, even without serially-correlated errors, which can lead to incorrect inference. The standard errors in our regression analyses are clustered at the session level and are robust to the concerns put forth in Stock and Watson (2008). Clustered standard errors have desirable large-sample properties (Wooldridge 2003). The significance of the coefficients estimated in the different models is based on a t distribution with Gl degrees of freedom, where G represents the number of clusters (12) and l represents the number of cluster-level coefficients estimated in the model (1), resulting in a critical t value of 2.201 at the 95% level, an adjustment made to acknowledge that the group-level error follows a t-distribution with small samples (Donald and Lang 2007).
17 Standard errors are clustered at the session level as opposed to the use of a multilevel model because the research question at hand is not interested with session-level effects and because the data is balanced, with the same number of auction periods in each treatment across sessions (Gelman 2006).
18 The statistics in table 2 are based on all submitted offers. Our primary goal is to understand how access to information about conservation-action quality impacts bidder behavior, which includes the item-selection and bid submission processes for all offers, whether they are ultimately accepted or rejected. Presenting data for all submitted bids prevents any mischaracterization of bidder behavior across treatments due to the acceptance criteria. , based only on accepted offers, shows that there are relatively few differences in trends across treatments between the two samples.
19 The experimental design features repeated measures across treatments on the same pool of participants in each session and therefore the pair-wise comparisons in this section violate the assumption of independence between groups. This interdependence is addressed by preceding a post hoc Tukey test with a repeated-measures ANOVA to obtain accurate p-values.
20 The presented statistics include results from all experimental sessions. In five of twelve periods in the first treatment (No Quality) of session five, a participant submitted prices that were several times greater than the budget for the auction, which were orders of magnitude higher than any observed prices across all the other sessions. As high prices are simply non-competitive, it is unlikely that this behavior meaningfully impacted auction outcomes for other participants. We have retained this data in the analysis to allay concerns that this aberrant behavior could meaningfully influence our results, as this type of behavior may occur in the field with landowners new to conservation auctions. This handful of very large data points drives the large variation and mean value for the price variable in the No Quality treatment. Dropping this sessions’ data and conducting a Tukey HSD test, we find that the only significant difference in the average price across treatments is that prices in the No Quality treatment are, on average, less than the prices in the other treatments (Quality Value - p = 0.039, Quality Rank - p = 0.006). Table 6 in the online supplemental appendix presents descriptive statistics excluding session five.
21 Maximum likelihood estimation of fixed-effects models can lead to inconsistent estimates of structural parameters (Neyman and Scott 1948). Greene (2004) suggests that there are certain conditions under which this model would be preferred to the pooled logit estimator, utilizing Monte Carlo methods to explore the impacts of incidental parameters on small-sample results and finding that the fixed effects estimator is biased upward in small samples, increasing the possibility of type I error, while the pooled estimator is biased downward in small samples. The magnitude of the bias is shown to be greater for the fixed-effects estimator with a limited number of time periods, with the biases of the estimators comparable once there are at least eight time periods. Given these results, we present results of the pooled logit model in table 7 of the online supplemental appendix to ensure the robustness of our results. depicts the mean values of the conservation action characteristics across the selected and non-selected items.
22 To demonstrate the robustness of our findings related to prices, online presents the results of model 3 using Wild bootstrapped standard errors, using a methodology described in Cameron, Gelbach, and Miller (2008).
23 The risk aversion variable is represented by an indicator variable that takes on a value of one if the individual switched from lottery A to lottery B after paired lottery six in the Holt and Loury risk attitude assessment test (Holt and Laury 2002), indicating risk aversion. The risk aversion indicator is time-invariant and is not identified in the model with participant fixed effects.
24 Online presents the results of running the above regression model independently for each quality information treatment using the subset of accepted offers. The results from this subset of the data are generally similar to the overall results; however, it is important to note that when the data are filtered on accepted offers, a correlation is introduced between quality and the offer amount, as higher-quality submissions can support higher prices while maintaining an acceptable score.
25 The No Quality treatment leaves each seller in a position where improving the competitiveness of their score can only be achieved via lowering their offer, as they are unaware of the quality of any of their actions. In this information environment, the lowest-cost action is equally likely to have the highest quality given the known independent nature of cost and quality draws, so selecting this action has the highest expected value for a given offer level. The binding budget also reinforces the lowest-cost action as the optimal selection. Nonetheless, roughly one-third of the time sellers do not choose the lowest-cost action. In looking at the data in more depth, we find that: (a) there is a strong tendency towards selecting the lowest-cost action in the No Quality treatment versus other treatments, (b) the more lowest-cost actions chosen, the higher a seller’s profits, and (c) sellers who do not select the lowest-cost actions offer statistically indistinguishable quality levels compared to those who submit their lowest cost action, ex post. These findings suggest that a subset of sellers were not able to identify the optimal strategy in this treatment. Responses to a post-experiment questionnaire about strategies in the experiment anecdotally support this conclusion. Of those responses that directly addressed strategy in the No Quality treatment, twelve said they focused on the lowest cost and six had strategies that were either random (five) or focused on beliefs about correlation (one), reflecting the proportions observed in the treatment. Similar difficulties in identifying bidding strategies were not expressed for the Quality Value treatment in the questionnaire, suggesting that withholding information may inhibit the formation of bidding strategies to the detriment of auction performance. This is in addition to the identification effect issue highlighted in this article.

References

Armantier
O.
Treich
N.
.
2009
.
Subjective Probabilities in GamesAn Application to the Overbidding Puzzle
.
International Economic Review
 
50
(
4
):
1079
102
.
Arnold
M.A.
Duke
J.M.
Messer
K.D.
.
2013
.
Adverse Selection in Reverse Auctions for Environmental Services
.
Land Economics
 
89
(
3
):
387
412
.
Asker
J.
Cantillon
E.
.
2008
.
Properties of Scoring Auctions
.
The RAND Journal of Economics
 
39
(
1
):
69
85
.
Asker
J.
Cantillon
E.
.
2010
.
Procurement when Price and Quality Matter
.
The RAND Journal of Economics
 
41
(
1
):
1
34
.
Babcock
B.
Lakshminarayan
P.
Wu
J.
Zilberman
D.
.
1997
.
Targeting Tools for the Purchase of Environmental Amenities
.
Land Economics
 
73
(
3
):
325
39
.
Babcock
B.A.
Lakshminarayan
P.G.
Wu
J.
Zilberman
D.
.
1996
.
The Economics of a Public Fund for Environmental AmenitiesA Case Study of CRP Contracts
.
American Journal of Agricultural Economics
 
78
:
961
71
.
Banerjee
S.
Kwasnica
A.M.
Shortle
J.S
.
2014
.
Information and Auction Performance: A Laboratory Study of Conservation Auctions for Spatially Contiguous Land Management
.
Environmental and Resource Economics
 
61
(
3
):
1
23
.
Banks
J.
Olson
M.
Porter
D.
Rassenti
S.
Smith
V.
.
2003
.
Theory, Experiment and the Federal Communications Commission Spectrum Auctions
.
Journal of Economic Behavior and Organization
 
51
(
3
):
303
50
.
Branco
F.
1997
.
The Design of Multidimensional Auctions
.
The RAND Journal of Economics
 
28
(
1
):
63
81
.
Cameron
A.C.
Gelbach
J.B.
Miller
D.L.
.
2008
.
BootstrapBased Improvements for Inference with Clustered Errors
.
The Review of Economics and Statistics
 
90
(
3
):
414
27
.
Cason
T.N.
Gangadharan
L.
.
2005
.
A Laboratory Comparison of Uniform and Discriminative Price Auctions for Reducing NonPoint Source Pollution
.
Land Economics
 
81
(
1
):
51
70
.
Cason
T.N.
Gangadharan
L.
Duke
C.
.
2003
.
A Laboratory Study of Auctions for Reducing NonPoint Source Pollution
.
Journal of Environmental Economics and Management
 
46
:
446
71
.
Cattaneo
A.
Claassen
R.
Johansson
R.
Weinberg
M.
.
2005
.
Flexible Conservation Measures on Working Land: What Challenges Lie ahead?
Economic Research Report
 
vol. 5
.
United Stated Department of Agriculture, Economic Research Service
.
Washington, DC
.
52
pp.
Charness
G.
Gneezy
U.
Kuhn
M.A.
.
2012
.
Experimental MethodsBetweenSubject and WithinSubject Design
.
Journal of Economic Behavior
 
81
:
1
8
.
Chavas
J.-P.
Chambers
R.G.
Pope
R.D.
.
2010
.
Production Economics and Farm Management: A Century of Contributions
.
American Journal of Agricultural Economics
 
92
(
2
):
356
75
.
Che
Y.K.
1993
.
Competition through Multidimensional Auctions
.
RAND Journal of Economics
 
24
:
668
80
.
Claassen
R.
Cattaneo
A.
Johansson
R.
.
2008
.
Cost-Effective Design of Agri-Environmental Payment Programs: U.S. Experience in Theory and Practice
.
Ecological Economics
 
65
(
4
):
737
52
.
Conte
M.N.
2013
. Valuing Ecosystem Services. In
Encyclopedia of Biodiversity
 , ed.
Levin
S.
,
314
326
,
Princeton University Press
.
Donald
S.G.
Lang
K.
.
2007
.
Inference with DifferenceinDifferences and other Panel Data
.
The Review of Economics and Statistics
 
89
(
2
):
221
33
.
Ferraro
P.F.
2008
.
Asymmetric Information and Contract Design for Payments for Environmental Services
.
Ecological Economics
 
65
:
811
22
.
Fischbacher
U.
2007
.
z-tree: Zurich Toolbox for Ready-Made Economic Experiments
.
Experimental Economics
 
10
(
2
):
171
8
.
Fréchette
G.
2012
.
Session-Effects in the Laboratory
.
Experimental Economics
 
15
(
3
):
485
98
.
Friesen
L.
Gangadharan
L.
.
2013
.
Environmental Markets: What Do We Learn from the Lab?
Journal of Economic Surveys
 
27
(
3
):
515
35
.
Garboua
L.L.
Maafi
H.
Masclet
D.
Terracol
A.
.
2012
.
Risk Aversion and Framing Effects
.
Experimental Economics
 
15
(
1
):
128
44
.
Gelman
A.
2006
.
Multilevel (Hierarchical) Modeling: What It Can and Cannot Do
.
Technometrics
 
48
(
3
):
432
5
.
Glebe
T.
2013
.
Conservation Auctions: Should Information about Environmental Benefits Be Made Public?
American Journal of Agricultural Economics
 
95
(
3
):
590
605
.
Goeree
J.K.
Holt
C.A.
Palfrey
T.R.
.
2002
.
Quantal Response Equilibrium and Overbidding in PrivateValue Auctions
.
Journal of Economic Theory
 
104
(
1
):
247
72
.
Greene
W.
2004
.
Incentives in Teams
.
The Econometrics Journal
 
7
(
1
):
98
119
.
Guerry
A.
Polasky
S.
Lubchenco
J.
Chaplin-Kramer
R.
Daily
G.
Griffin
R.
Ruckelshaus
M.
, et al.  .
2015
.
Natural Capital and Ecosystem Services Informing DecisionsFrom Promise to Practice
.
Proceedings of the National Academy of Sciences
 
112
(
24
):
7348
55
.
Hajkowicz
S.
Higgins
A.
Miller
C.
Marinoni
O.
.
2008
.
Targeting Conservation Payments to Achieve Multiple Outcomes
.
Biological Conservation
 
141
:
2368
75
.
Haruvy
E.
Katok
E.
.
2013
.
Increasing Revenue by Decreasing Information in Procurement Auctions
.
Production and Operations Management
 
22
(
1
):
19
35
.
Heimlich
R.
1989
.
Productivity and Erodibility of U.S. Cropland
.
Agricultural Economics Report
 
604
.
United States Department of Agriculture, Economic Research Service
.
Washington, DC
. 56 pp.
Hellerstein
D.
2010
.
Challenges Facing USDA’s Conservation Reserve Program
.
Amber Waves
 
8
(
2
):
28
33
.
Hellerstein
D.
Higgins
N.
.
2010
.
The Effective Use of Limited Information: Do Bid Maximums Reduce Procurement Cost in Asymmetric Auctions?
Agriculture and Resource Economics Review
 
39
(
2
):
288
304
.
Hellerstein
D.
Higgins
N.I.
Roberts
M.
.
2015
.
Options for Improving Conservation Programs: Insights from Auction Theory and Economic Experiments
.
Amber Waves
 .
United States Department of Agriculture, Economic Research Service
.
Washington, DC
. 53 pp.
Holt
C.A.
Laury
S.K.
.
2002
.
Risk Aversion and Incentive Effects
.
American Economic Review
 
92
(
5
):
1644
55
.
Jacobs
K.L.
Thurman
W.N.
Marra
M.C.
.
2014
.
The Effect of Conservation Priority Areas on Bidding Behavior in the Conservation Reserve Program
.
Land Economics
 
90
(
1
):
1
25
.
Johnson
K.A.
Dalzell
B.J.
Donahue
M.
Gourevitch
J.
Johnson
D.L.
Karlovits
G.S.
Keeler
B.
Smith
J.T.
.
2016
.
Conservation Reserve Program CRPLands Provide Ecosystem Service Benefits that Exceed Land Rental Payment Costs
.
Ecosystem Services
 
18
:
175
85
.
Kagel
J.H.
Roth
A.E.
.
2002
.
Theory and Misbehavior in First-Price Auctions: Comment
.
American Economic Review
 
82
(
5
):
1380
91
.
Kareiva
P.K.
Tallis
H.
Daily
G.
Ricketts
T.
Polasky
S.
, eds.
2011
.
Natural Capital: Theory and Practice of Mapping Ecosystem Services
 .
New York: Oxford University Press
.
Kirwan
B.
Lubowski
R.N.
Roberts
M.J.
.
2005
.
How Cost-Effective Are Land Retirement Auctions? Estimating the Difference between Payments and Willingness to Accept in the Conservation Reserve Program
.
American Journal of Agricultural Economics
 
87
(
5
):
1239
47
.
Latacz-Lohmann
U.
Van der Hamsvoort
C.P.
.
1997
.
Auctioning Conservation Contracts: A Theoretical Analysis and Application
.
American Journal of Agricultural Economics
 
79
:
407
18
.
McAfee
R.P.
McMillan
J.
.
1987
.
Auctions and Bidding
.
Journal of Economic Literature
 
25
(
2
):
699
738
.
Messer
K.
Duke
J.
Lynch
L.
.
2014
. Applying Experiments to Land Economics: Public Information and Auction Efficiency in Ecosystem Service Markets. In
Oxford Handbook of Land Economics
 , ed.
Duke
J.
Wu
J.
,
1
23
.
Oxford University Press
.
Messer
K.
Duke
J.
Lynch
L.
Li
T.
.
2016
.
When Does Public Information Undermine the Efficiency of Reverse Auctions for the Purchase of Ecosystem Services?
Applied Economics and Statistics Research Report, University of Delaware
 ,
RR16
04
.
Neyman
J.
Scott
E.L.
.
1948
.
Consistent Estimation from Partially Consistent Observations
.
Econometrica
 
16
:
1
32
.
Reeson
A.F.
Rodriguez
L.C.
Whitten
S.M.
Williams
K.
Nolles
K.
Windle
J.
Rolfe
J.
.
2011
.
Adapting Auctions for the Provision of Ecosystem Services at the Landscape Scale
.
Ecological Economics
 
70
(
9
):
1621
7
.
Rezende
L.
2009
.
Biased Procurement Auctions
.
Economic Theory
 
38
(
1
):
169
85
.
Ribaudo
M.O.
Hoag
D.L.
Smith
M.E.
Heimlich
R.
.
2001
.
Environmental Indices and the Politics of the Conservation Reserve Program
.
Ecological Indicators
 
1
(
1
):
11
20
.
Rolfe
J.
Windle
J.
McCosker
J.
.
2009
.
Testing and Implementing the Use of Multiple Bidding Rounds in Conservation Auctions: A Case Study Application
.
Canadian Journal of Agricultural Economics/Revue Canadienne d’Agroeconomie
 
57
(
3
):
287
303
.
Roth
A.E.
2002
.
The Economist as Engineer: Game Theory, Experimentation, and Computation as Tools for Design Economics
.
Econometrica
 
70
(
4
):
1341
78
.
Schilizzi
S.
Latacz-Lohmann
U.
.
2007
.
Quantifying the Benefits of Conservation Auctions
.
Eurochoices
 
6
(
3
):
32
9
.
Short
C.
Winter
M.
Crabb
J.
Temple
M.
Dauven
A.
Augustin
B.
.
2000
. Economic Evaluation of the Countryside Stewardship Scheme. Cheltenham and Gloucester College of Higher Education and ADAS.
Stock
J.H.
Watson
M.W.
.
2008
.
HeteroskedasticityRobust Standard Errors for Fixed Effects Panel Data Regression
.
Econometrica
 
76
(
1
):
155
74
.
Stoneham
G.
Chaudhri
V.
Ha
A.
Strappazzon
L.
.
2003
.
Auctions for Conservation ContractsAn Empirical Examination of Victoria’s Bushtender Trial
.
Australian Journal of Agricultural and Resource Economics
 
47
(
4
):
477
500
.
Ulber
L.
Klimek
S.
Steinmann
H.
Isselstein
J.
.
2010
.
A Market-Based Payment Scheme for Plant Diversity in Farming Systems
.
Aspects of Applied Biology
 
100
:
319
26
.
Ulber
L.
Klimek
S.
Steinmann
H.
Isselstein
J.
Groth
M.
.
2011
.
Implementing and Evaluating the Effectiveness of a Payment Scheme for Environmental Services from Agricultural Land
.
Environmental Conservation
 
38
(
4
):
464
72
.
U.S. Department of Agriculture
.
2012
. The Conservation Reserve Program: 43rd signup results. Washington DC.
Vickrey
W.
1961
.
Counter speculation Auctions and Competitive Sealed Tenders
.
Journal of Finance
 
16
(
1
):
8
37
.
Vukina
T.
Zheng
X.
Marra
M.
Levy
A.
.
2008
.
Do Farmers Value the Environment? Evidence from a Conservation Reserve Program Auction
.
International Journal of Industrial Organization
 
26
:
1323
32
.
Williams
J.D.
1971
.
A Multiple Regression Approach to Multiple Comparisons for Comparing Several Treatments with a Control
.
The Journal of Experimental Educational
 
39
(
3
):
93
6
.
Wooldridge
J.M.
2003
.
ClusterSample Methods in Applied Econometrics
.
American Economic Review
 
93
(
2
):
133
8
.

Appendix

Lottery Instructions

Welcome to the experiment. This is an experiment in market decision making. If you follow the instructions carefully and make good decisions you will be well-prepared to succeed in today’s experiment. In today’s session you will participate in a lottery and a series of auctions. Your cash earnings today will consist of a $10 show-up payment, and payments based on your performance in the lottery and auctions.

We will first start with the lottery experiment. We will read through the instructions for the lottery together, and then proceed to the software interface.

In this part of today’s session you are asked to make a choice in 11 different paired lotteries. Each lottery has different possible combinations of payoffs. Your task will be to consider each lottery and select A or B using the scroll bar to indicate a preference for taking part in sub-lottery A or sub-lottery B. Consider the payoffs associated with selecting A or B for each of the 10 choices and pick accordingly, as your selection will affect your payoff for completing this task.

After you are finished selecting A or B for the 11 choices, please press ”click here to continue.” At the end of the experiment when you come to collect your earnings, one of the lotteries will be selected at random to determine your payoff. These payoffs are in $, not experimental dollars like the rest of today’s session.

To determine your payoff, when you come to collect your earnings for the experiment at the conclusion of the session, you will:

  1. Select a card from shuffled cards numbered 1–11 to select the lottery that you will receive your payoff from.

  2. Select another card from shuffled cards numbered 1–10 to determine which of the two payoffs you receive from the chosen sub-lottery.

The payoff from the lottery will be added to your show-up payment and your earnings from the rest of the experiment.

Auction Instructions

We will now proceed into the next phase of today’s session, where you will make decisions in an auction environment. During this part of the experiment, you will earn money in experimental dollars. At the end of the experiment, these will be converted to real dollars at a rate of 120 experimental dollars per $1 and you will be paid as you leave. This is in addition to the $10 show-up payment and your lottery earnings.

How You Make Money

In today’s experiment, you will participate in a series of auctions. In each auction period, you will have three types of items to sell: Red, Green, and Blue items. Each item has a cost and quality, which will vary from period to period and across participants. Your values in one period are in no way linked to values in other periods. Your cost and quality values are also not linked. As we move through the instructions, you can refer to the supplemental handout for an example screenshot of the auction software interface.

In each auction period, you must choose one item that you would like to sell and the price (your “offer”) that you would like to sell the item at. Your choice of item and your offer are collected via the software interface and are not known to other participants. Do not use a dollar sign when entering your offer through the software interface.

The experimenter (who is the buyer) has a limited budget and likely cannot purchase all items offered by all participants in each auction period. You can sell only one item per period, and if you sell that item, then you must pay that item’s cost. If you are able to successfully sell an item in a given period, your earnings in that auction period are equal to the value of your offer minus the cost of the item sold.  

PeriodEarnings=Offercost

Consider the following illustrative example: if your offer of 220 for your Red item is accepted in a given period, and it has a cost of 200, your earnings that period would be 220 – 200 = 20. These period earnings are recorded and added to your “session earnings” before you move on to the next auction period. If you do not sell an item in a period, your earnings are zero for that period; you only pay an item’s cost if you are able to sell that item.

Quality

The experimenter, who is the buyer, values higher quality items and uses a scoring rule to help ensure the budget is spent on high quality items. The probability your offer will be accepted is based on both the offer and the quality of the item, in addition to the quality and offers of other auction participants. To rank offers for acceptance, the experimenter turns your offer and quality information into a score using the following rule:  

Score=Quality/Offer

Consider the following illustrative example: If your offer is 50 and the quality of the item is 10, then your score is 10/50 = 0.2. Others’ items are scored similarly and then they are ranked by their score. Offers are accepted at their offer value until the budget is exhausted.

Participant ID Rank Score Quality Offer Accepted Budget (100 to start) 
0.49 27 55 45 
0.43 21 24 
0.29 20 70 −54 
0.21 38 −92 
Participant ID Rank Score Quality Offer Accepted Budget (100 to start) 
0.49 27 55 45 
0.43 21 24 
0.29 20 70 −54 
0.21 38 −92 

In the above example table, you will see that offers are ranked not on quality or offer, but instead on their score. Items are accepted in order starting from the highest score to the lowest score, until the budget of 100 experimental dollars is exhausted. In the experiment you will not know the budget level or anything about the quality, offers, scores, and acceptance decisions of other participants.

A key thing to understand is that the numbers used in the example are for illustration only and the values and your offer choice may be completely different in the actual experiment.

Changes from the Basic Setup

You are now familiar with the basic design of an auction period. During the course of the experiment, some of the information on your screen may change.

The change relates to your knowledge of the quality of each of your items. In every auction period, each of your items has a quality assigned to it; however, in some periods the quality value may not be displayed even though it is known to the experimenter. In these periods, the experimenter still assigns a score based on your offer and quality and uses the score to rank bids. However, you will have to make your item choice and offer without this information.

In other periods, each item’s quality will be displayed as a rank. As before, the exact number is known to the experimenter and scoring is based on the exact number. However, you will only see the quality rank of the three items.

Item Quality  Item Quality 
Red 27 ⇒ Red 
Green Green 
Blue 20 Blue 
Item Quality  Item Quality 
Red 27 ⇒ Red 
Green Green 
Blue 20 Blue 

In this example, Red has the highest quality of your three items, so it is ranked first. Blue is second highest, so it gets a rank of two. Green is the lowest quality, so it is ranked third. Again, these numbers are for example only and the values in the experiment may be different.

Summary

A period has the following order

  1. Period begins.

  2. You select an item to sell and enter an offer price to sell that item for.

  3. All offers are submitted and ranked according to the scoring rule.

  4. Offers are accepted according to their ranked score until the budget is exhausted.

  5. You are notified if your offer was accepted and your earnings are calculated automatically and added to your session total.

  6. Period ends.

Questions

How well you understand these rules and procedures are an important determinant of how much you earn in today’s session. Think back over the instructions, and if you have any questions, please raise your hand now. We will conduct a practice auction next to give you an opportunity to familiarize yourself with the auction interface. None of the earnings in the practice auction will influence your cash payment today. Once the auctions begin, no talking among participants will be permitted.

Author notes

Marc Conte is an assistant professor in the Department of Economics, Fordham University. Robert Griffin is a post-doctoral research fellow at the Woods Institute for the Environment, Stanford University. This research was funded by a Fordham University faculty research grant. Robert Griffin is a post-doctoral research fellow at the Woods Institute for the Environment, Stanford University and acknowledges financial support from the Gordon and Betty Moore Foundation. Correspondence to be sent to: mconte7@fordham.edu.

Supplementary data