Although methods such as contingent valuation have received a great deal of attention in environmental valuation literature, fewer studies have reported willingness-to-pay estimates with agribusiness applications. Because agribusinesses are increasingly interested in producing and selling differentiated goods and services whose values has not been established by well-functioning markets, we provide a short introduction to willingness-to-pay methodology and provide a discussion of several different methods used to estimate willingness-to-pay. More specifically, we discuss how much of the work in environmental and experimental valuation literature can be extended to agribusiness applications, which have their own set of unique issues.
Many producer groups and agribusinesses are interested in “adding value” to their products by differentiating generic agricultural commodities or developing alternative products or services with new technologies.1 However, research and development and new product introductions can be costly. Compounding this are the tens of thousands of new food products introduced annually with success rates often as low as 10%. Thus, market research into the viability of new products and services is critical.
When investigating the viability of a new venture, agribusiness firms are generally interested in two factors: production costs and consumer demand for the new product or service. These factors are often the primary determinants of product adoption and pricing decisions. Although costs are relatively straightforward to estimate, assessing consumer demand for novel products and attributes is often more complex. Because these firms are interested in selling a new product or promoting a novel attribute, secondary data from actual markets are unavailable. To estimate consumer demand, or willingness-to-pay (WTP), for these novel goods or services, economists must turn to hypothetical contingent or experimental markets.
In recent years, several studies have estimated consumer WTP for neoteric products or food quality enhancements (e.g., Buhr et al.; Hayes et al.; Hoffman et al.; Fox; Fox et al.; Lusk et al. 2001a, 2001b; Lusk, Roosen, and Fox; Melton et al.; Menkhaus et al.; Roosen et al.; Shogren et al. 1999; Unterschultz et al.). However, the primary focus has been on methodological or theoretical issues associated with estimating WTP or on policy issues rather than on how the estimates might be used to make product adoption or pricing decisions. Furthermore, most of the literature on WTP estimation has been in the context of contingent valuation, where the primary focus is on environmental issues and on the measurement of changes in aggregate welfare (e.g., Hanemann 1984).
Despite the rising interest in estimating consumer demand for value-added foods and services, the intersection of agribusiness economists and those familiar with issues and methods of WTP estimation appears to be small. The purpose of this paper is to narrow the gap between WTP experts and agribusiness marketing economists.2 More specifically, this paper seeks to show how much of the work in environmental and experimental valuation literature can be extended to agribusiness applications. One of the key issues is that estimates of average or median WTP, which are often the primary statistics of interest in environmental valuation, are less relevant for agribusinesses. WTP estimates are useful to agribusinesses in the sense that they can be used to estimate market demand for novel products.
In this paper, we formally define WTP and discuss conceptual issues associated with its interpretation. The following section discusses several methods used to measure WTP and outlines their advantages and disadvantages. We illustrate how WTP estimates derived from various elicitation methods can be applied to agribusiness applications. We highlight several issues that are important for agribusiness applications and offer some concluding comments about the use of WTP in agribusiness research.
WTP, in its traditional sense, is a Hicksian surplus measure. Hanemann (1991) provides a thorough discussion of the theoretical underpinnings of consumer WTP and shows that WTP can be expressed in a number of equivalent manners. First, one might consider a consumer's utility maximization problem subject to a budget constraint, where the level of a good's quality (q) is fixed exogenously. Many applications refer to q as the level of a public good provided, but for agribusiness applications, q is most applicable as a measurement or index of a good's quality (see Hanemann, 1991). The consumer chooses the level of the market good (xm) that maximizes utility, producing the traditional Marshallian demand curve, xm(p, y, q); where p is the market price of the good and y is income. The resulting indirect utility function is v(p, y, q). Now assume that an agribusiness considers an improvement in the quality of an existing product from q0 to q1. A measurement of the value the consumer places on this improvement can be derived by determining the magnitude of WTP such that the following equality holds: v(p, y − WTP, q1) = v(p, y, q0).
One can also consider the consumer's dual expenditure minimization problem subject to a given level of utility. In this case, the consumer chooses the level of consumption of the market good (xh) that minimizes expenditures, producing the familiar Hicksian demand curve, xh(p, U, q), where U is the level of utility. The associated indirect expenditure function is, m(p, U, q). In this case, the estimated value the consumer places on the change in the good's quality from q0 to q1 is, WTP = m(p, U0, q0) − m(p, U0, q1).
Alternatively, WTP can be derived by assuming the consumer is able to choose the level of quality, q, in addition to levels of consumption of market goods, x. This approach is a hypothetical maximization problem as the level of quality is actually set exogenously; however, it is useful for examining other theoretical notions of WTP. In this case, demands, xm(p, g, y) and qm(p, g, y), result from the utility maximization problem, where g is the price of q. The dual to this problem is to minimize expenditures on x and q subject to a given level of utility. The resulting Hicksian demand for quality is qh(p, g, U). The inverse-compensated demand function for quality, g(qh, p, U) is often referred to as the WTP curve. The function identifies the price an individual is willing to pay for a given level of quality, q, given specific levels of p and U. The estimated WTP3 for a change in quality from q0 to q1 is
Agribusinesses such as supermarkets, restaurants, and food processors will be interested in estimating consumer demand for a new product or attribute; however, agribusinesses such as seed and chemical companies, technology and equipment dealers, and agricultural service providers might also be interested in producer (farmer) WTP for a new product or service (e.g., Hudson and Hite). Although WTP is almost always discussed within the context of utility maximization of consumers, this concept can also be extended to producers.
In this case, we consider a producer's profit maximization decision subject to a given production function. The producer chooses the level of inputs, x, to use, but the level of one input, q, is fixed exogenously. Here q can be thought of as the level of some service provided, a new technology, or the quality of some input. Given a vector of input prices, w, and a vector of output prices, p, the producer chooses the optimal level of inputs and outputs, which yields the indirect restricted profit function, π(p, w, q).
Now assume that an agribusiness considers improving the quality of an existing product or service from q0 to q1. The WTP, or shadow price, for the change is: WTP = π(p, w, q1) − π(p, w, q0). In other words, WTP represents the maximum amount of profit a producer would be willing to forgo to obtain q1 rather than q0.
Measurement of Willingness-to-Pay
There are several methods available to estimate consumer or producer WTP for novel goods or changes in the qualities of existing goods. In outlining the advantages and disadvantages of elicitation methods, several factors are important to consider. One of the primary issues surrounding the credibility of an elicitation technique is that of incentive compatibility. An elicitation mechanism is considered incentive-compatible if an individual's dominant strategy is to truthfully reveal their preference for the good in question. A closely related issue is that of hypothetical bias: that individuals respond differently when responding to hypothetical questions than when confronted with real payment. Because many valuation questions involve asking hypothetical questions where incentives may not be properly aligned, this issue is an important consideration.
The vast majority of studies suggest that hypothetical bias is a significant problem in contingent valuation estimates. For example, List and Gallet conducted a Meta analysis of twenty-nine studies containing fifty-eight valuations and found that average subjects overstated their preferences by a factor of about three in hypothetical settings. Similar results have been observed by Cummings, Harrison, and Rutström; Cummings and Taylor; Fox et al.; List and Shogren; List; Loomis et al.; and Neill et al.4 More generally, Carson, Groves, and Machina contend that conventional contingent valuation techniques, such as the single-bounded dichotomous choice question, are not incentive-compatible when considering the provision of a new private good in a hypothetical context. The advantage of agribusiness over environmental applications is that goods are typically private and deliverable. That is, when asking an individual what they would be willing to pay for a new product, the valuation setting can almost always be made nonhypothetical, which generally makes the elicitation method incentive-compatible.5 Several different elicitation methods are discussed below (see Lee and Hatcher for discussions on other elicitation methods in food marketing applications).
Dichotomous Choice Questions
Dichotomous and double-bounded dichotomous choice questions are frequently used in environmental economics literature to estimate the value of nonmarket goods (e.g., An; Hanemann, Loomis, and Kanninen). Although dichotomous and double-bounded dichotomous choice questions have been used less frequently in valuation of novel food products, these contingent valuation techniques are easily extendable to this application (e.g., Lusk 2003b; Ready, Buzby, and Hu; Wertenbroch and Skiera). In such questions, consumers are typically confronted with the price of a new product and are asked whether they would buy the new product (YES or NO) at the stated price. In a single-bounded dichotomous choice framework, the price of the new product is varied across surveys and average WTP is estimated by examining how the proportion of YES responses varies at alternative price levels. In a double-bounded dichotomous choice question, if an individual responds NO to the first question, another dichotomous choice question is posed with a lower price; however, if an individual responds YES to the first question, a subsequent dichotomous choice question is posed with a higher price.
The single-bounded dichotomous choice question is perhaps the most frequently used method in the environmental valuation literature because of its proposed incentive-compatibility properties (Carson, Groves, and Machina). Furthermore, estimation of WTP is straightforward using existing software. Typically the probability of observing a YES response, conditional on a constant and price, is estimated with simple logit or probit models, where average WTP is given by the ratio of the constant coefficient to the price coefficient (Hanemann 1984). Another common estimation approach was proposed by Cameron (1988, 1991), where the distribution of WTP is directly modeled using interval censoring techniques. Cameron and James have also shown how this estimation technique is applicable for marketing applications.
The advantage of the double-bounded dichotomous choice framework is statistical efficiency. Hanemann, Loomis, and Kanninen have shown that the double-bounded approach yields more efficient estimates of mean WTP than the single-bounded approach, primarily because the double-bounded approach incorporates more information about an individual's WTP than a single dichotomous choice question. Despite this advantage, there are several potential drawbacks to the double-bounded dichotomous choice method. First, Carson, Groves, and Machina contend the method may not be incentive-compatible in a hypothetical context. Second, Cameron and Quiggin have shown that responses to the first and second dichotomous choice questions may not be perfectly correlated, bringing into question which WTP estimate is most relevant. Finally, several authors have suggested that the double-bounded approach may suffer from starting point biases: responses to the second question depend on the price offered in the first (e.g., Shogren and Herriges).
One disadvantage with both single- and double-bounded approaches is that they only elicit discrete choices. That is, researchers only observe whether an individual would pay more or less than a particular price level. Thus, parametric assumptions must be made about the distribution of WTP in a sample. A more informationally efficient approach might involve eliciting each individual's exact WTP. A final disadvantage of both approaches is that they only consider WTP for a single good and estimation of cross-price effects between new and competing products requires modification of conventional survey design.
Choice-based Conjoint Analysis
Choice-based conjoint (CBC) analysis is another method that can be used to elicit WTP for novel goods or services. CBC has been frequently used in marketing, transportation, and environmental valuation literature (Adamowicz et al. 1998; Louviere, Hensher, and Swait; Lusk, Roosen, and Fox; Unterschultz et al.). In a CBC framework, consumers are typically confronted with a choice between alternative products, defined by several attributes, such as price and quality. Consumers are typically asked to choose which product they would purchase, given several product descriptions. For example, an individual might be asked to choose between three orange juice options, where each differs by brand, price, and sweetness. In turn, each attribute is typically varied at several levels. For example, orange juice price might be between $0.75, $1.00, and $1.75 and sweetness might be varied between “unsweetened” and “sweetened.” Through focus groups, pretesting, and available market data, a researcher selects the attributes to include in a CBC application and determines the number and magnitude of levels in each attribute. Experimental design procedures are used to vary product descriptions across several repeated choice sets.
CBC has several advantages over other contingent valuation techniques. First, CBC is based on random utility theory and is consistent with Lancaster's theory of utility maximization, where consumers demand attributes embodied in a good (Louviere, Hensher, and Swait). Second, CBC closely mimics consumers' typical shopping experiences—choosing one product from several competing options. Third, CBC allows a researcher to investigate trade-offs between several competing product attributes, such as price, package size, nutrition, etc., which is not easily done with traditional contingent valuation techniques. Finally, CBC can be readily used to estimate cross-price elasticities between novel and existing products, a task more difficult with other techniques such as dichotomous choice questions.
Another advantage of the CBC framework is that hypothetical responses to CBC questions have been found to be similar to revealed preferences (e.g., Adamowicz et al. 1997, Carlsson and Martinsson). Lusk and Schroeder (in press) found that hypothetical CBC responses were statistically different from nonhypothetical responses, but differences were generally small especially when determining marginal WTP for a change in product quality. Even if CBC is prone to hypothetical bias, this technique can readily be used to construct simulated retail settings, which have the advantage of increased realism for participants. In a simulated retail setting, a researcher could create a store shelf similar to one found in a traditional agribusiness. The shelf could be stocked with alternative products, including pre-existing and neoteric products. Participants in the simulated retail setting can indicate which product they will purchase in several repeated pricing scenarios or trials. At the completion of all trials, a random number could be drawn to determine the binding trial. Once the binding trial is determined, participants could make the purchase(s) they originally indicated on their bid sheet for the specified trial (see Lusk and Schroeder 2003).6
With CBC data, the probability of choice is typically modeled using a multinomial logit (MNL) model, as a function of the attributes of the good, including price (see Louviere, Hensher, and Swait). Alternative estimation approaches, such as random parameters logit, multinomial probit, heteroskedastic extreme value, hierarchical Bayes, or latent class estimator are also available to relax the restrictive independence of irrelevant alternatives assumption associated with the MNL.
Despite the large number of advantages of CBC, disadvantages also exist. First, as with dichotomous choice questions, only discrete choices are observed from respondents, making identification of WTP and market demand relatively complex. Second, it is difficult to incorporate consumer demographics and other explanatory variables into CBC models. Third, estimating models that relax the assumptions of the traditional MNL can be difficult and are not generally supported by existing software. Fourth, experimental design of choice sets can be confusing for beginning practitioners (see Louviere, Hensher, and Swait or Lusk (2003a) for discussions on this issue). Finally, several studies have found that subjects' responses may be inconsistent across choice questions or are influenced by the complexity of the choice task (e.g., Johnson and Desvousges, DeShazo and Fermo, Swait and Adamowicz).
Experimental auctions are becoming a popular method of nonmarket valuation because of evidence that consumers respond differently in hypothetical and real environments. Experimental auctions are generally conducted in one of two ways. First, consumers can be provided with an endowed good (typically a pre-existing substitute) and then are asked to bid to exchange their endowed good for a novel good (e.g., Fox, Lusk et al. 2001a). Secondly, consumers can bid directly on several competing goods and a random drawing can be used to determine which good is binding so that demand for a single unit can be elicited (e.g., List and Shogren; Lusk, Feldkamp, and Schroeder). In the former case, the auction elicits demand for the novel good relative to a pre-existing substitute. The advantage of the endowment approach is that it forces subjects to focus on the auction, because they will leave the experiment with at least one good regardless of their actions. The disadvantage of the approach is a potential problem with the “endowment effect,” where subjects place greater value on a good simply because they possess it (Kahneman, Knetsch, and Thaler).7
Using a demand-revealing auction, such as a Vickrey second price auction, each consumer bids against other auction participants for the novel good. In the case of the second price auction, the individual with the highest bid pays the second highest amount and either exchanges their pre-existing product for the novel product or obtains the randomly selected good, depending upon the chosen method. All other auction participants either retain their pre-existing product or obtain nothing. Bid amounts in the auction reflect consumer's WTP. Other auctions commonly employed in the literature include the English, BDM, and random nth price (Lusk et al. 2001a, Rutström, Shogren et al. 2001). Becker, DeGroot, and Marschak; Irwin et al.; Shogren et al. (2001); and Vickrey provide formal discussions of incentive compatibility properties of the each of these auctions. Auction types vary by the level of market interaction and feedback, but all theoretically yield the same value assuming independently distributed values. Shogren discusses the advantages and disadvantages of employing auctions that incorporate market feedback. And Lusk (2003c) discusses practical advantages and disadvantages associated with implementing different incentive compatible auctions.
There are several advantages of using experimental auctions to elicit consumer WTP for novel goods or services. First, one obtains a bid (WTP value) from each individual precluding the need to make parametric assumptions about the shape of the market demand curve. Second, experimental auctions involve the exchange of real goods and real money with properly aligned incentives. Third, modeling determinants of WTP is straightforward given the continuous nature of the dependent variable. Fourth, subjects can incorporate feedback from the experimental market into their bids as they might in an actual market setting. Finally, there is a wealth of theoretical literature on auctions that can aid researchers in designing appropriate experiments as well as providing interesting hypotheses to test.
Some drawbacks to experiments include: (a) subjects must be recruited and paid participatory fees to attend laboratory sessions, which potentially introduces bias into resulting bids (Rutström) and limits sample sizes; (b) bids might be truncated or censored by outside alternatives (substitutes) not available in the experiment (Harrison, Harstad, and Rutström);8 (c) bidder values may be or become affiliated, which degrades the incentive compatibility of an auction (Milgrom and Weber); and (d) depending upon procedure, it is not uncommon to observe a large frequency of zero bidding, potentially because of participant disinterest.
Fortunately, some of these drawbacks can be mitigated by conducting experiments in a field rather than lab setting. Some recent experimental work has begun to elicit values in a field or retail setting (e.g., Bohm, List, List and Lucking-Reily, Lusk et al. 2001a). By moving the valuation setting to more familiar territory for the respondent, the researcher might: (a) better target the population of interest (thereby reducing sample selection bias); (b) reduce costs of experimental work because of lower compensatory fees; (c) decrease bias associated with high or nonuniform compensatory fees; and (d) most importantly, put subjects in a context where typical purchasing decisions are made. In a laboratory, subjects have limited opportunity to purchase substitute goods and may have little knowledge of competing prices that can appropriately anchor valuation and promote consistency with retail behavior. Eliciting values in a field setting relaxes this restriction. In fact, Lusk and Fox found that experimental setting (lab vs. store) can have an influence on WTP bids.
Comparison of Valuations across Methods
A number of studies have compared WTP estimates across elicitation methods. A short review of these findings is presented here. By gaining insight into the relative magnitude of WTP across methods, researchers might ascertain how results might differ if an alternative elicitation technique were employed.9 First, research suggests that WTP from double-bounded dichotomous choice questions is generally greater than WTP from single-bounded dichotomous choice questions (Hanemann, Loomis, and Kanninen; Cameron and Quiggin). Second, WTP from open-ended questions and experimental auctions is typically less than WTP from single-bounded dichotomous choice questions (Frykblom and Shogren; McFadden; Ready, Buzby, and Hu). Third, WTP from single-bounded dichotomous choice questions is generally equivalent to that from CBC (Adamowicz et al. 1998). Fourth, bids from 2nd price, BDM, English, and random nth price auctions are lower than WTP from CBC (Lusk and Schroeder 2003c). Fifth, 2nd price auction bids are greater than English and BDM bids after multiple bidding rounds (Lusk, Feldkamp, and Schroeder; Rutström).
Applying Willingness-to-Pay to Agribusiness Applications
Theoretically, WTP measures the maximum amount of money an individual is willing to give up to either: (a) obtain a product with quality q or (b) exchange a product with quality q0 for a product with quality q1 as discussed in the second section of the paper. Practically, how can agribusiness use these measures? At this point, an important distinction must be made. The discussion in the second section of the paper was related to measurement of an individual's WTP. However, agribusinesses will typically be interested in the distribution of WTP in a particular market.
We contend that agribusinesses are interested in WTP to the extent that these measures can be used to construct inverse-compensated demand curves for a novel product in a particular market. Provided certain assumptions about the number of units purchased per individual within a particular time period, inverse market demand curves can be constructed by horizontally summing individual demand curves (WTP) over the entire market. When the application elicits WTP for one unit of a novel good, the individual demand curves consist of a single point (e.g., price = WTP, quantity = 1). When the application is extended to elicit WTP for multiple units per person, more developed inverse demand curves for each individual can be constructed prior to horizontal summation to the market level. Even if WTP for multiple units is not elicited, assumptions can be made about the degree of diminishing marginal WTP within a particular time period to derive more complete individual demands. Once demand schedules are identified, agribusinesses can locate the position on the demand curve that maximizes profit.
Constructing inverse demand curves is relatively straightforward, depending upon the particular methodology employed to estimate WTP. If, for example, WTP is estimated using a discrete choice method (i.e., dichotomous choice or CBC), one can employ a probabilistic model (logit, probit, MNL, etc.) to calculate the percentage of individuals with WTP greater than particular price levels. These probabilities can be converted to quantities by making certain assumptions about the number of individuals in a particular market and the number of units purchased per unit time. Although predictions from these models are often interpreted as the probability that an individual has WTP greater than a particular price, an equivalent interpretation is that the prediction gives the frequency of individuals (e.g., market share) with WTP greater than a particular price (e.g., Louviere, Hensher, and Swait).
It is also straightforward to estimate an own-price elasticity of market share to determine price sensitivity of the novel product. The own-price elasticity is simply (∂prob/∂price) × (prob/price), where prob is the market share from the probabilistic model. (Louviere, Hensher, and Swait). Predictions from the probabilistic model can also be used to make revenue projections. Projected monthly revenue, for example, might be: monthly revenue = (proposed price of novel good) × (market size) × (projected market share at proposed price) × (average monthly consumption of close substitute).
One drawback to this approach is that certain assumptions must be made about the parametric distribution of WTP. If WTP is assumed normally distributed in the population, then the estimated mean WTP and standard deviation are the only statistics needed to construct an inverse demand curve. However, when normality is assumed (as opposed to having the entire distribution of WTP), identification of viable niche markets may be impossible.
If WTP is estimated using a direct method (open-ended contingent valuation question, experimental auction, etc.), a slightly different approach must be employed to construct inverse demand curves. Under the assumption that only one-unit is purchased per person per unit of time, WTP values can simply be sorted from highest to lowest and plotted against a linear time trend to identify the inverse demand curve for the sample (see Wertenbroch and Skiera for one example). To extend the demand curve to a larger market, the sorted WTP values can be plotted against the frequency of the sample with WTP greater than particular price levels (i.e., construct a frequency table or histogram). The advantage of this approach is that no assumption must be made about the distribution of WTP or the shape of the demand curve, unless one wishes to parameterize the degree of price sensitivity.
Special Issues for Agribusiness Applications
Most of the developments of contingent valuation methodology have come from environmental literature. However, most environmental applications are interested in valuing public goods with few substitutes or are interested in policy implications associated with changes in appropriations of public goods. As a result, there are a few issues which agribusiness researchers might consider when performing a WTP study that are perhaps less relevant for environmental applications.
A common trend among current studies estimating WTP for agribusiness or environmental applications is that they focus almost exclusively on estimating demand for a single new product or the premium for a new product relative to a pre-existing substitute. However, WTP estimates for a single good may not be adequate for a multiproduct firm or one with many direct competitors to determine whether new product adoption will be profitable. The relationship between sales of pre-existing products and introduction of a neoteric product must be known to make effective product adoption and pricing decisions. An estimated WTP premium above costs is not sufficient to ensure enhanced profitability because introduction of the neoteric product will likely affect sales of a firm's other goods or may cause competitors to lower the price of substitute goods. In essence, agribusinesses must consider the cross-price elasticity between a novel product and pre-existing ones before an accurate product adoption decision can be made.10
To illustrate this issue, we conducted a small pilot study with sixty undergraduate students enrolled in an agricultural economics course. The students were asked a hypothetical open-ended question regarding how much they would be willing to pay for steam-pasteurized (SP) ground beef, which contains lower levels of bacteria than typical ground beef (Nutsch et al. 1997, 1998; Phebus et al.). Half the students were told that regular ground beef sold for $1.00/lb in the local grocery store while the other half were told that regular ground beef sold for $2.00/lb.11 We found that average WTP for SP ground beef was $1.72/lb and $2.77/lb when non-SP (regular) ground beef was $1.00/lb and $2.00/lb, respectively.
Figure 1 displays the demand schedules for SP ground beef when the price of non-SP was $1.00/lb and $2.00/lb. The key point here is that the position of the demand curve for SP beef is heavily dependent on the price of non-SP beef. For example, when the price of SP ground beef was $2.50/lb, expected market share for SP ground beef was 14% if non-SP ground beef price was $1.00/lb and 59% if non-SP ground beef price was $2.00/lb. Obviously, such a large difference in expected market share for a new product could generate different adoption decisions. The degree of substitutability is clearly an important component that should be measured prior to making sales forecasts or product adoption decisions.
Although previous literature has focused heavily on the intricacies of estimating single product WTP, incorporating methods of estimating cross-price elasticities in nonmarket valuation is straightforward. In the following discussion, we outline some potential methods to incorporate estimates of cross-price effects into traditional nonmarket valuation techniques.12
In a dichotomous choice framework, cross-price elasticities can be estimated by simply incorporating the price of a pre-existing product into the survey design. In the initial question, the price of a pre-existing product can simply be stated and the consumer can be asked the dichotomous YES/NO question(s). In addition to varying the price of the novel good across surveys, the price of the pre-existing good can be systematically varied across surveys using a predefined experimental design. Following such a procedure, the probability of observing a YES can now be defined as a function of own- and cross-prices. Because two variables are allowed to vary, the survey design grows in size, thus increasing the cost of the survey. Decisions must be made, then, comparing the marginal benefit of the added information from cross-price effects with the added cost of the survey.
CBC most readily lends itself to estimating cross-price elasticities because researchers can incorporate pre-existing products into the choice set with the novel good. For example, a consumer can be asked to choose between products A, B, and C, where A is new and B and C are pre-existing products. The prices of the three goods can be altered, according to a predefined experimental design, and the consumer can be asked again to choose between A, B, and C. Importantly, one should estimate a model that relaxes the independence of irrelevant alternatives assumption of the MNL when attempting to identify cross-price effects (see Louviere, Hensher, and Swait).
In addition to choosing a different estimator, Pilon suggests an alternative method of estimating cross-price elasticities using CBC. After making a product choice, a consumer can indicate the number of units they wish purchase. Then, in a traditional fashion, the level of consumption can be modeled as a function of price and cross-prices. While incorporation of cross-price effects necessarily makes estimation and interpretation of results more complicated, it should not significantly increase the cost or change experimental design for the researcher.
Cross-price effects might also be estimated with experimental auctions; however, the procedure is less straightforward. First, auction participants could be asked to bid full value for a novel good. An approach similar to that described by Cummings, Ganderton, and McGuckin might be followed, where consumers are asked to bid on a novel product in the presence and absence of substitute/compliment products. Substitute/compliment relationships can be estimated by examining how the bid price for the novel good changes when consumers are asked to value multiple goods.
Aside from this procedure, cross-price elasticities might be estimated by auctioning one good (e.g., a novel good) while another (e.g., a pre-existing substitute) is offered for sale at a posted price. By varying the posted price of the pre-existing good and examining how WTP bids for the novel good change, cross-price elasticities might be estimated, either for individuals through repeated trials or on the aggregate through repeated sessions with different participants. Unfortunately, it is unclear how such a procedural change might interfere with the incentive-compatibility auction.
It is an important point to emphasize that agribusinesses are interested in WTP to the extent that inferences can be made about market demand. That is, agribusinesses are not likely to be able to accurately predict profitability of a novel product using only mean WTP. Although the distribution of WTP is important for agribusinesses, in traditional environmental valuation studies, the focus is on estimating mean WTP and aggregate welfare changes.
In environmental applications, mean WTP may be the only statistic needed to carry out cost-benefit analysis. For example, estimated mean WTP can simply be multiplied by the number of individuals affected to derive an approximate value of a particular policy, which can then be compared with aggregate costs. For agribusinesses, however, knowledge of the distribution of WTP is more relevant. For example, consider the extreme case where mean WTP for a novel product is relatively low, but a few individuals have very high WTP. Such a result for an environmental application, where funding for amenities must be derived approximately uniformly from each individual (i.e., a tax), would likely preclude adoption of a particular policy. For an agribusiness, however, a very profitable niche market may exist where the novel product can be highly priced. Regardless of whether a niche market exists, the profit-maximizing price level may be very different from mean WTP—and the knowledge of mean WTP alone does little to indicate what the profit-maximizing level might be.
To illustrate this issue, we refer back to the student pilot study. Figure 1 also illustrates demand curves constructed using only information provided by the mean and standard deviation, assuming normality. This might mimic the situation in which these are the only statistics reported by a WTP study. If we draw inferences about potential demand for SP ground beef based only on these assumptions, several biases might exist. As clearly shown in figure 1, actual market share for SP ground beef (computed using empirical WTP values) is greater than the simulated market share for SP ground beef (computed using normality assumption) at relatively high prices. For example, if non-SP ground beef was $2.00/lb and SP ground beef were priced at $3.50/lb, actual WTP data suggest a market share of almost 25%, whereas the simulated demand predicted market share would be only 17%. Thus, using only mean WTP data could potentially misidentify an important niche market, or at the least, underpredict the profitability of a niche market.13 In addition, it is relatively straightforward to determine that, given the price of non-SP ground beef is $1.00/lb, the profit-maximizing price to set for SP ground beef is $1.98/lb. given a marginal cost of $0.75/lb. In this case, the profit-maximizing price is over $0.25/lb higher than average WTP.
To assist firms in product adoption decisions, future research might provide illustrations of inverse demand curves under several distributional assumptions when discrete choice methods are employed and report frequency distributions of WTP when a direct elicitation approach is employed. In either case, statistics on the consumption habits of the sample should be reported to assist in construction of a market demand.
Another important issue for agribusinesses is identification of heterogeneity in consumer segments. Although environmental applications are primarily interested in aggregate welfare changes, agribusinesses might serve specialized niche markets, where consumers' preferences are quite different from aggregate markets.
A number of approaches exist to identify market segments. In CBC applications, one can estimate a random parameters logit model (e.g., Layton and Brown, Revelt and Train). With this specification, marginal utilities for attributes are allowed to vary in the population according to some specified distribution. With a random parameters logit, researchers can identify how preferences for various attributes might vary in a population. In CBC, another helpful model is a latent class estimator (Dillon and Kumar, Swait and Adamowicz). With a latent class estimator, discrete market segments can be identified. Each market segment is composed of different preference estimates for the attributes incorporated into the CBC. For example, one segment might be extremely price sensitive, while another is less price sensitive. Agribusiness can make appropriate pricing decisions depending upon the sizes of the segments.
Another way to assess heterogeneity in consumers is to survey subjects about their knowledge, attitudes, and perceptions about the product or service prior to posing WTP questions. Survey questions might also be constructed to determine consumer lifestyles, values, and socioeconomic characteristics. Techniques such as structural equations modeling have been used in agricultural economics literature to provide formal structure to map consumer perceptions and values to behavior intentions (e.g., Hensen and Northen, Hensen and Traill). By using multiple-scale items to measure theoretical constructs, structural equations modeling can be used to determine how various factors (and in turn, different consumer segments) might influence WTP. Most marketing applications use structural equations modeling to determine how various factors influence behavioral intentions (as measured by a scale) as opposed to WTP. By combining the rigor of WTP analysis with marketing techniques, such as structural equations modeling, agribusinesses may be able to gain valuable information about important market segments.
Lastly, there are also clustering methods that can be used to identify different market segments (e.g., Baker and Burnham). With these methods, the researcher identifies key variables or attributes of interest (that might be elicited in a survey) and creates discrete clusters (segments) of respondents based on these variables. Again, this technique might be useful to identify whether there are relatively price elastic and inelastic segments in a market.
As agribusinesses cope with the shift toward a more consumer- and demand-driven marketplace, estimates of the value of novel products are becoming important instruments guiding decision making. In an effort to assist agribusinesses in this task, a number of recent studies have estimated and reported WTP values for novel foods and technologies that have potential application for many agribusinesses. In this paper, we aimed to provide a guide for improving future studies estimating consumer WTP when the goal is to assist agribusinesses in product adoption decisions.
First, one must recognize that the objectives of WTP elicitation are different when the application is agribusiness-related versus environmental policy. Environmental policy is primarily concerned with estimating mean WTP and aggregate welfare changes. In contrast, agribusinesses are interested in WTP measures that can be used to derive compensated market demand curves for novel products. When new products are developed, an agribusiness may be able to exercise some degree of market power, and as such, they are likely interested in identifying the position on the demand curve that maximizes profit, which may be very different from the mean value. Future work in this area should attempt to provide as much detail as possible about the distribution of WTP within the sample so that proper inferences about the shape of the demand curve can be made.
We also suggest that future studies attempt to incorporate cross-price elasticity estimates in WTP studies, as these measures may be equally important as single WTP estimates. Omission of cross-price effects can significantly diminish business decision-making power. Incorporating cross-price effects into existing valuation methods is relatively straightforward, but can increase the complexity of experimental design. Thus, researchers should compare the marginal benefit of the added information from cross-price effects with the added cost of survey design, administration, and analysis.
With the proliferation of novel food products, agricultural economists have directed their attention toward estimating the value of these new products to assist agribusinesses in their adoption decision. Rigorous contingent and experimental valuation techniques have been developed to elicit consumer WTP for these novel foods. Despite recent advances, existing work is still in need of refinement to be useful to agribusinesses.
Throughout the paper, we refer to agribusiness as any firm involved in the production, processing, and selling of agricultural goods and services. Agribusinesses include food retailers, food processors, farmers, chemical manufacturers, seed companies, etc.
We do not intend to provide a comprehensive literature review of the contingent valuation literature. Interested readers are directed to Hanemann (1994) or Carson, Flores, and Meade. For a good discussion on use of experiments to address nonmarket valuation issues see Shogren.
WTP can also be derived from a consumer's maximization of expected utility when choosing amongst risky prospects (see Hayes et al.).
A few studies have found that hypothetical bias might not exist in all settings and with all elicitation methods (e.g., Haab, Haung, and Whitehead; Smith and Mansfield). In the context of contingent valuation, Carson, Groves, and Machina contend that incentive compatibility in hypothetical questions is only achieved with single-bounded dichotomous choice questions used in a referendum format with public goods.
Another problem with all field-based methods for eliciting WTP is that they might not be incentive-compatible if subjects consider their responses consequential beyond the immediate survey context (Carson, Groves, and Machina). For example, subjects might think they can affect subsequent market prices by strategically offering low WTP in a premarket study.
As discussed in Wertenbroch and Skiera, AC Nielsen's test market simulation follows a similar protocol.
Lusk, Feldkamp, and Schroeder found that the endowment effect had little influence on bids in the context presented here.
Harrison, Harstad, and Rutström outline methods to deal with the problem of censoring in experiments. The procedures include controlling for outside field prices in statistical estimation of WTP.
We note that not all research is consistent with these generalizations; however, the preponderance of evidence is supportive of the statements.
A somewhat similar argument has been made in nonmarket environmental valuation literature. Most research in this area has focused on the bias in WTP estimates when substitute/complement environmental programs are not considered (Boxall et al.; Cummings, Ganderton, and McGuckin; Loomis, Gonzalez-Caban, and Gregory).
More detail about the information provided to subjects as well as exact instructions and WTP questions can be obtained from the authors upon request. Obviously, there are contingent valuation techniques that are likely less problematic than a hypothetical open-ended question. Our goal is only to provide a simple illustration to emphasize the potential effect of including substitute prices in a typical WTP study.
It is important to note that the cross-price elasticity of WTP is not the same measure as the cross-price elasticity of demand (see Flores and Carson for a related discussion on the income elasticity of demand and WTP). Thus, care must be taken in interpreting cross-price effects in models of WTP.
The potential for this sort of bias can also be seen in the figures reported by Wertenbroch and Skiera, who conducted a more rigorous study with actual market participants.