Experimental Design
We design our experiment to incorporate the private-information-revealing feature of experimental auction markets (Smith 1976; Fox et al. 2001) and the rigorous randomized treatment effects of statistical experimental design. The experimental design consisted of six biotech information-labeling treatments with two replications. The treatments are randomly assigned to twelve experimental units, each consisting of 13 to 16 consumers drawn from the households of two major urban areas and who are paid to participate. Each participant participated in two trials.i Using randomly chosen consumers from the population of an urban area, rather than undergraduate college students at a university, is a advantage when it comes to drawing inferences from the experiments or generalizing to the Midwest or whole U.S. population.
Consumers might react differently to GM content in different types of food or they may have no demand for some food products. Using only one food item seemed unlikely to reveal enough information, given the sizeable fixed cost of conducting the experiment. Three food items were chosen: vegetable oil (made from soybeans), tortilla chips (made from yellow corn), and Russet potatoes. In the distilling and refining process for vegetable oils, essentially all of the proteins (which are the components of DNA and the source of genetic modification) are removed leaving pure lipids. Minimal human health concerns should arise from GM oil, but consumers may either worry that GM soybeans affect the environment or lack adequate information on the distilling process. Tortilla chips are highly processed foods that may be made from GM or non-GM corn, and consumers might have human health and environmental concerns. Russet potatoes are purchased as a fresh product and are generally baked or fried before eating. Similar to tortilla chips, consumers might see both human health and environmental risks from eating GM-Russet potatoes.
Auctions were conducted at two Midwestern U.S. cities: Des Moines, IA, and St. Paul, MN. Participants in the auctions were consumers in these two areas who were contacted by the Iowa State University Statistics Laboratory and agreed to participate in the study. The Statistics Laboratory obtained 1,200 to 1,500 randomly selected residence telephone numbers from each of the metropolitan areas. Employees of the ISU Statistics Laboratory called these numbers to make sure that the phone number was for a residence. The employees then asked to speak to an adult in the household (individual who was 18 years of age or older). They were told that Iowa State University was looking for people who were willing to participate in a group session in Des Moines (St. Paul) that relates to how people select food and household products. The sessions were held on Saturday, April 7th (April 21st) and participants were informed that the session would last about 90 minutes. Each participant was told that they would receive $40 in cash for their time. The sessions were held at the Iowa State University Learning Connection, 7th and Locust, Des Moines (lower level of the Classroom Office Building, University of Minnesota, St. Paul). Three different times were available each auction day, 9 a.m., 11:30 p.m., and 2 p.m., and willing participants were asked to choose a time that best fit their schedule. Participation per household was limited to two adult individuals, and they were assigned to different groups.ii To willing participants, the Statistics Laboratory followed up by sending a letter containing more information, including a map and instructions on when and where the meeting would be held, how to get there, and a telephone number to contact for more information.
There were twelve experimental units, six in Des Moines, and six in Minneapolis. Twelve hundred people in Des Moines were called and 99 of them agreed to participate. Of those 99 people who agreed to participate, 77 did indeed attend. For the Minneapolis experiments, 1,500 people were called and 118 people agreed to participate. Of those 118, we had 95 participants in the Minneapolis experiments. The total sample size is 172, which is large compared to most experimental auctions.
Each auction had ten steps, which are summarized in figure 1.iii When participants arrived at the lab, they signed a consent form to agree to participate in the auction. After they signed this form, they were given $40 for participating and an ID number to preserve their anonymity. The participants then read a brief set of instructions and filled out a questionnaire.
Step 2 introduced the auction. We used a random nth price auction in this experiment (Shogren et al., 2001). The advantages of the random nth price auction are that it is demand revealing in theory and the auction attempts to engage bidders at all locations along the demand curve.iv The random nth-price works as follows: Each of k bidders submits a bid for one unit of a good; then each of the bids is rank-ordered from highest to lowest. The auction monitor then selects a random number which is drawn from a uniform distribution between 2 and k; and the monitor sells one unit of the good to each of the (n-1) highest bidders at the nth-price. For instance, if the monitor randomly selects n = 5, the four highest bidders each purchase one unit of the good priced at the fifth-highest bid. Ex ante, bidders who have low or moderate valuations now have a nontrivial chance to buy the good because the price is determined randomly. This auction increases the odds that insincere bidding will lead to a loss. Participants were given detailed instructions on the random nth-price auction, including an example written on the board. After the participants learned about the auction, a short quiz was given to ensure that everyone understood how the auction worked.
Step 3 was the first practice round of bidding, where participants bid on a brand-name candy bar. The participants were asked to examine the product and then place a (sealed) bid on the candy bar. The bids were collected and the first round of practice bidding was over. Throughout the auction, when the participants were bidding on items in a particular round, they had no indication of what other items they may be bidding on in future rounds or if additional rounds would occur.
Step 4 was the second practice round of bidding. In this round the participants bid separately on three different items. The three products were the same brand-name candy bar, a deck of playing cards and a box of pens. The consumers were asked to examine the three products in practice round two and make bids on the products. Then the bids were collected. Only one of the two rounds was chosen as binding (valid), so that participants would not take home more than one of any product. The reason was to eliminate price reduction due to the consumer buying a larger quantity because of diminishing marginal utility of these products (i.e., lower prices due to a consumer’s negatively sloped demand curve).v Participants were informed that only one of the two rounds would bind before step 3 and were reminded of this again before step 4.
After the two practice auction rounds were completed, the binding round and the binding nth-prices were revealed in step 5. All of the bids were written on the blackboard, and the nth-prices were circled for each of the three products. Participants could see what items they won immediately, and the market-clearing price. The participants were notified that all purchases of goods would take place after the experiment was over, so that all exchanges of money for goods would take place at the end of the session.
In step 6, information about biotechnology was released to the participants. The possible types of information a participant could receive were: (1) the industry perspective—a collection of statements and information on genetic modification provided by a group of leading biotechnology companies, including Monsanto and Syngenta; (2) the environmental group perspective—a collection of statements and information on genetic modification from Greenpeace, a leading environmental group; and (3) the third-party, verifiable perspective—a statement on genetic modification approved by a third-party group, consisting of a variety of individuals knowledgeable about genetically modified goods, including scientists, professionals, religious leaders, and academics, none of whom have a financial stake in genetically modified foods. To assist the participants process these different sources of information, the volume of information released of each type was limited to one 8 1/2" x 11" page, and it was organized into five categories: general information, scientific impact, human impact, financial impact, and environmental impact, to ease the information processing load on participants. Figures 3, 4, and 5 show the exact wording of the three types of information about genetically modified food.
The information was randomized to create six treatments of information combinations: pro-biotechnology information; anti-biotechnology information; both pro and anti-biotechnology information;vi pro-biotechnology and third party, verifiable information; anti-biotechnology and third-party,vii verifiable information; and pro-biotechnology, anti-biotechnology, and third-party, verifiable information. These six combinations were randomized among all twelve experimental units, with each information combination going to two experimental units.
Two auction rounds followed the distribution of information. One of the two rounds had the participants bid on food products with just a standard food label. The other round had participants bid on the same food products with the same label, except there was a sentence added indicating that the food had been genetically engineered. These labels were made as plain as possible to avoid any influence on the bids from the label design (see Figure 2). The sequencing of GM labels was randomized across experimental units. Each combination of information was given to two experimental units. One of these experimental units bid on food with the standard label in round one, and the food with the label indicating genetic modification in round two. The other experimental unit bid on food with the label indicating genetic modification in round one, and the standard label in round two. For each experimental unit, only one of the two food rounds was chosen as the binding (valid) round. This avoided the problem of bid prices being reduced as consumers moved along their demand curve.
In step 7, participants bid on three different food products: a bag of potatoes, a bottle of vegetable oil, and a bag of tortilla chips. The participants were instructed to examine the three products and then write down their (sealed) bid for each of the three goods. Participants bid on each good separately. Then the bids were collected from the individuals, and the participants were informed that they were about to look at another group of food items.
Step 8 had participants come examine the same three food products, but with the different labels (the second trial). After the participants examined the products, they were instructed to bid on the three products. Each good was bid on separately. The bids were then collected from all of the participants. Once again, consumers were informed that only one of the two trials or bidding rounds would bind before step 7, and they were told this again before step 8.viii
Step 9 consisted of the selection of which of the two trials would be chosen as binding, along with the binding nth-prices. After the binding round and binding nth prices were revealed, the winners were notified and the participants were asked to complete a brief post-auction questionnaire. In step 10, the participants who did not win any products were informed that they were free to leave, and the participants who won products exchanged money for their goods, and then they were free to leave.
We can examine the bids from the experimental auctions to see how information influences whether or not an auction participant was “out of the market” for GM-labeled foods. This is defined as a demand so low that the participant would not demand GM-labeled foods at any price. This will give insights into how the diffusion of technology is impacted by information.
Although we follow standard experimental auction valuation procedures (e.g., Smith 1976; Shogren et al., 1994), we make several refinements to our experimental design to better reflect consumer purchases. First, our subjects submitted only one bid per product instead of using multiple repeated trials and posted market-clearing prices to avoid affiliated values which can affect the demand-revealing nature of a laboratory auction (see e.g., List and Shogren 1999). Second, we do not endow our subjects with any food item and then ask them to “upgrade” to another food item as that can cause distorted bid prices (e.g., Lusk and Shroeder, 2002). Third, each consumer bid on three unrelated food items, such that if he or she did not have positive demand for one or two products, we could still obtain information from them on their taste for genetic modification based on the second and (or) third product. Fourth, we randomly assigned treatments to the experimental units – now estimation of treatment effect is simply the difference in means across treatments (see Wooldridge, 2001). Fifth, we use adult consumers over 18 years of age from two different Mid-western metropolitan areas that were chosen from a random digit dialing method. See figure 1 for a summary of the demographic characteristics of our sample. The demographics of our sample do not perfectly match the U.S. census demographic characteristics for these regions, but they are similar and provide a sufficient representation for our initial probe into labeling and information for GM products (see Appendix table 1for the 2002 Census of Population demographic characteristics of the areas). Although our participants are slightly skewed toward women, Katsara, et al. (2001) showed that women make up a disproportional share of grocery shoppers—83 percent of shoppers versus 52 percent in the U.S. Census of Population. Finally, information from our laboratory experiments is complimented by information obtained from pre- and post-experiment surveys administered to participants. The pre-auction survey allowed us to obtain socio-demographic information and information on participants’ beliefs about GM and other technologies before treatment, which is useful to help explain bidder behavior.
The Model
Economic theory suggests that as a consumer receives commutative negative information about a product or process, their demand schedule shifts down, causing them to consume a lower quantity of a good at a given price. If a consumer hears enough negative information about a product or process – their demand for a product or process could be so low that they would not be willing to purchase one unit at a positive price!
Objective, verifiable information is likely to dampen the effectiveness of negative information at pushing down a consumer’s demand for a product. If consumers are given verifiable information on GM foods, it is likely that some consumers who would otherwise be “out of the market” would now purchase GM-labeled foods if it was priced appropriately. Probit models are used to examine what characteristics help push a consumer out of the market, and what helps keep consumers in the market for GM foods.
We use two tests to examine the probability a consumer is “out of the market” for GM foods. The strong test assumes that a consumer is out of the market if that consumer bids zero for one unit of the GM food when they bid a positive amount for one unit of the unlabeled product. If a consumer does not have a positive willingness to pay (WTP) for one unit of a GM food when they have a positive WTP for their non-GM counterpart, this suggests they would not consume the GM variety of the product at any price. Determining what impacts the probability of being out of the market is important, because if a large share of consumers will not buy a particular food product, the grocery store will discontinue displaying/stocking/supplying the food. The weak test assumes that a consumer is “out of the market” if the price an individual bids for the GM-labeled food is less than or equal to 2/3 of the price they bid for the plain-labeled food. This weak test captures the reality that most premiums for non-GM foods do not exceed 20 percent (Kiesel, Buschena, and Smith, 2002).
We fit probit models explaining the probability that a consumer is out of the market for each of the three products used in our economics experiments – Russet potatoes, vegetable oil, and tortilla chips using both the strong test and the weak test. In this paper we summarize the results from probit models to show how anti-GM information from environmental groups and verifiable information from a third party affect the probability that a consumer is out of the market for each of the 3 GM food products. In addition, we estimate the probability that a consumer is out of the market for all 3 food products jointly. This is the probability that a consumer bid zero for every GM-labeled food product for the strong test, or bid 2/3’s or less of their bid for each plain labeled product in the weak test.
Results
Part A of table 2 shows the mean bid prices for all participants. Consumers, on average, discounted GM-labeled foods by fourteen percent. Part B shows that participants who received only positive information actually put a premium on the GM-labeled food for two of the three products. This was despite the fact that the genetic modification was only used to enhance the production process, and did not give the foods any enhanced attributes. Part C shows that when consumers received only negative information, they discount the GM-labeled foods by an average of approximately thirty-five percent. Part D shows that consumers who received both positive and negative information discount the GM-labeled foods by an average of seventeen to twenty-nine percent, depending on the food product.
Third-party information has an impact on the willingness to pay for GM-labeled foods. Part E shows that consumers who received positive and third-party information discounted GM-labeled foods slightly. This is in contrast to the consumers who received only positive information who valued the GM-labeled foods more than their plain-labeled counterpart on average. Part F shows that participants who received negative and third-party information still discounted the GM-labeled foods, but by a smaller amount than the participants who received only negative information. Part G shows that participants who received negative and third-party information discounted the GM-labeled foods by an average of seventeen to twenty-two percent, depending on the product. Participants who received positive, negative and third-party information were more accepting of the GM-labeled foods than those who received only positive and negative information. The participants who received positive, negative and third-party information discounted the GM-labeled food by an average of zero to eleven percent, depending on the product.
Our results are consistent with Viscusi (1997) who found that individuals placed a slightly greater weight on negative information than positive information. In our auction, participants who received only positive information did not discount the GM-labeled food, while those who received only negative information discounted the GM-labeled food by an average of 35 percent. Those who received both positive and negative information put slightly more weight on the negative information, discounting the GM-labeled foods by 20 percent. In addition, one explanation for a moderated negative auction outcome in our experiments where participants received positive and negative GM-product information is that some individuals have an asymmetric value function giving greater weight to marginal losses than to marginal gains (Kahneman and Tvesrky 1979), for example, they do not want to bear losses.
Also, our results are in contrast to Fox et al.’s (2001) who obtained the result that negative information dominated positive information. They argued that one reason could be due to a “status quo bias,” (or endowment effect) where participants were originally endowed with a regular pork sandwich and could bid to upgrade to an irradiated pork sandwich. Participants may have their bids biased due to being endowed with one type of sandwich.ix Our auction had participants bid on items in separate rounds (trials), thus our results are not influenced by a “status quo bias.”
Tables 3 and 4 show the percentage of participants who are out of the market for the GM-labeled commodities using both the strong and weak test. The number of observations differs, because if an individual bids zero for both the GM-labeled and plain-labeled versions of a commodity, they are not included in the analysis (they did not demand the product, so we cannot determine their taste for genetic modification). Similarly, when reporting on who is out of the market for all GM-labeled foods, those who bid zero for all the food products are not included.
A summary of 8 sets of probit results explaining the probability of a participant being out of the market for GM-labeled foods is presented in table 5, and the actual estimates of the coefficients in the probit models are reported in Appendix table 2 to 9 Several different specifications were used, to examine the robustness of the results. Negative information has a positive coefficient, indicating negative information increases the probability that a consumer is out of the market. The coefficients for these effects were consistently statistically significant, for all of the products under both the strong test and weak tests.
This result has important implications. If an anti-technology group wishes to slow scientific progress and asymmetric information exists, they could disseminate large amounts of negative information – even if the information is highly biased. They could even disguise their true intentions by telling consumers they want to keep everybody “fully informed” of the consequences of a product or technology. Yet, their negative information might help push demand down to zero for enough people that suppliers would not find it profitable to invest in the new technology. Also, even if firms do not fully believe the information, negative information will increase the uncertainty about genetic modification, which has been shown to decrease the likelihood of adoption (Purvis et al. 1995). Given that technological change is one of the driving forces behind the rising standard of living enjoyed in many developed countries, stalled progress could decrease welfare significantly over time.
This result also presents an alternative explanation for why Europeans demand for GM foods is so small. Many have hypothesized that Europeans dislike GM foods because of recent food scares like BSE, dioxin, and foot and mouth disease. Our results present an alternative explanation for the low Europeans demand for GM-labeled foods – environmental groups are more prevalent and have disseminated larger volumes of negative information about GM-technology and GM-foods and they do not have a trusted source of verifiable information.
Table 5 also shows that third party, verifiable information decreases the probability that an individual is out of the market, as these coefficients are negative under all of the model specifications, and are statistically significant at the 5 and 10 percent level in many of the models. This provides evidence that a third-party source that provides neutral, verifiable information on genetically modified foods could help prevent the market from disappearing due to lack of demand. Hence, in addition to value that verifiable information may have by providing consumers accurate information on the risks and benefits of genetic modification (estimated in Rousu et al. at $2.6 billion), verifiable information also may have value by keeping new GM food products in supermarkets, increasing the range of choices, which has a real option value.
|
