Consumer preferences for alternative formats of nutrition

Consumer preferences for alternative formats of nutrition information on grocery store shelf labels Joshua P. Berning Doctoral Candidate School of Economic Sciences Washington State University Pullman, WA 99164 phone: (509) 335-8600 email: jberning@wsu.edu Hayley Chouinard Assistant Professor School of Economic Sciences Washington State University Pullman, WA 99164 phone: (509) 335-8739 email: chouinard@wsu.edu Jill J. McCluskey Professor and Chair of Graduate Studies School of Economic Sciences Washington State University Pullman, WA 99164 phone: (509) 335-2835 email: mccluskey@wsu.edu Abstract: Alternative formats for providing nutrition information are being provided by manufacturers as well as grocery stores. While the effect of nutrition labels has been well studied, few papers examine the effect of different label formats. We use choice experiments to examine preferences for two different formats of nutrition information provided on grocery store shelf labels: detailed and easy-to-use information. We find larger mean preferences for detailed nutrition information but also a greater dispersion of preferences. Given the costs of obtaining information, the easy-to-use format may benefit more shoppers. In the case of nutrition labels, detailed labels may provide too much information. We are grateful for funding from the USDA, National Research Initiative. We thank Ken Manning and David Sprott for their input and direction. We want to recognize Rosangela Bando, Sophia Villas-Boas, Grant Chen, Melissa Hidrobo, Kristin Kiesel, Quoc Luong, and Bruno Romani for their help conducting the store surveys; Steve Flaming for his help facilitating store participation; and Sophia Villas-Boas for her work organizing the survey events. We also thank Ying Hu for her research assistance and Washington State University Graduate Research seminar for outstanding feedback on this paper. Introduction It is prohibitively expensive for shoppers to identify by themselves the nutritional content of food items. Nutrition labels benefit shoppers by effectively turning nutritional content, a credence characteristic, into a search characteristic (Caswell and Majduszka 1996). Nutrition labels thereby enhance economic efficiency by directing expenditures towards products that are most wanted (Golan et al 2001). Nutrition facts panels on product packaging are the standard source of nutrition information; however many different nutrition labels have also been created by manufacturers and grocery stores that vary in terms of both the information quality and method of providing the information. These formats are generally easier to use and also easier to locate for shoppers but offer less information than nutrition facts panels. While there is an extensive literature examining the effect of nutrition information on shopper behavior (Drichoutis et al 2006), few papers have discussed how shoppers perceive different formats of nutrition information. In this article, we examine shopper preferences for different formats of nutrition information, specifically focusing on nutrition information provided on grocery store shelf labels. To this end, we develop two label formats to provide on grocery store shelf labels: a format that provides more information defined as detailed and a format that provides less information defined as easy-to-use. We then use choice experiments to elicit preferences for the different label formats. We find larger mean preference for detailed nutrition information. At the same time, however, the dispersion of preferences is larger for the detailed information. Shoppers who face greater costs to obtaining nutrition information may be less likely to use the detailed information than the easy-to- 1 use information. Stores looking to provide shoppers with their own nutrition label should consider both the presentation and the amount of nutrition information they provide. The provision of nutrition information Nutrition information is provided to shoppers in many different formats beside the standard nutrition facts panels. Several major food companies have started providing their own nutrition information using proprietary labels on their packaging to indicate healthier choices (two examples are Pepsi’s Smart Choice and Kraft Food’s Sensible Solution). Some grocery store chains have also affixed nutrition information labels on store shelves along with product price labels to provide shoppers with easier access to nutrition information. For example, grocery stores in the Washington, D.C. area affixed shelf label markers to identify foods with specific positive nutrition claims1. Analysis of sales data found that the shelf label markers affected purchasing patterns by directing shoppers to healthier choices (Levy et al 1985). A similar experiment was conducted in the northeast using shelf label nutrition information2 in which the authors (Teisl et at 2001) found that shelf label nutrition information helps consumers increase their total satisfaction from food consumption while maintaining their overall health risks. More recently, a grocery store chain has developed its own star-rating shelf label which identifies the nutritional quality of food items based on the number of stars it receives (New York Times, Nov. 6, 2006). This store-specific star-rating method is designed using proprietary methods not regulated by the Food and Drug Administration (FDA) and therefore, the quality of the star-rating system is not explicitly known. Reports by the store suggest that the star-rating 1 2 These claims included low or reduced in sodium, calories, cholesterol and fat. Similar claims were made regarding products considered to be low or reduced in fat, cholesterol, sodium or calories. 2 labels have been successful at directing shoppers to healthier choices (New York Times, Sep. 7, 2007). As more manufacturers and grocery stores begin to offer their own systems for conveying nutrition information, it is relevant to examine how information quality affects shopper behavior. Persons value informational attributes differently and therefore derive different levels of utility from different information formats. As both opportunity cost and preferences for information vary across consumers, their use of information will vary as well (McCluskey and Swinnen 2004). Even if nutrition information is provided in a costless manner some consumers may have other motivations and taste preferences that preclude them from using nutrition information to make healthier diet choices (Teisl and Roe 1998). The effectiveness of nutrition labels lies in providing the appropriate nutrition label to specific consumer segments (Caswell and Padberg 1992). Additionally, nutrition information is likely to be effective when it addresses specific informational needs and can be processed and used by its target audience (Verbeke 2005). Labels that are appealing and easily processed by shoppers may be more effective at influencing consumer behavior. The demand for nutrition information People derive utility not just from the goods they consume, but also from the attributes of the goods (Lancaster 1966). In this article we separate product attributes into three groups: search, experience and credence characteristics. With search characteristics (Nelson 1970), shoppers are able to make choices based solely on appearances. For example, fruits and vegetables are often selected based on physical appearance. Other goods may benefit from experience (Nelson 1974), that is, learned knowledge of their 3 quality. For example, a shopper may develop preferences for wine after consuming it. In the case of credence characteristics, however, shoppers are essentially unable to evaluate which goods possess certain desired attributes (Darby and Karni 1973). Nutritional content is an example of a credence characteristic since it is prohibitively expensive for a person to identify nutritional content of an individual food item before or after consumption. Shoppers must therefore use nutrition information to identify preferred nutritional characteristics. A vector of goods (x) can be identified by its experience ( Qe ), credence ( Qc ) and search ( Qs ) characteristics: x = x(Qe , Qc , Qs ) . A person who consumes a vector of goods (x) receives utility: U = U [αQe , βQc (i ), δQs ] (1) The utility derived from experience ( Qe ), credence ( Qc (i ) ), and search ( Qs ) characteristics is weighted by the level of importance of each type of characteristic using α , β and δ . An example is the preferences of a person who purchases spinach because it is high in vitamins and low in fat versus a person who purchases spinach because of its color and texture. The utility derived from credence characteristics is unique in that it is a function of the product information (i) that a person has describing each products credence characteristics. In other words, a person gets utility from credence characteristics if they have information which identifies the presence of the characteristics3. In this way, nutrition information acts as a complement to the consumption of products with unknown nutritional quality similar to the way Becker and 3 In the rest of the article, we focus strictly of nutrition which is essentially a subset of credence characteristics. 4 Murphy (1993) describe advertisements acting as complements to advertised goods. The greater the consumption of nutritional quality, the greater the marginal utility and demand for the nutrition information. For certain individuals who have strong preferences for nutrition information (high β -types), the level of information acquisition will be high. They will seek out more product information to enhance their consumption of x. Alternatively, high α and δ types may derive more utility from experience and search characteristics respectively. Since the passage of the Nutrition Labeling Education Act (NLEA) in 1990, nutrition facts panels which provide a large amount of information describing nutritional content and ingredients have been required on almost all manufactured products. A relevant concern, however, is that nutrition facts panels offer too much information and are placed so that the product has to be removed from the shelf to be used. Given the time and effort required, the nutrition facts panels may are not always being used by shoppers. Nutrition information is being provided to shoppers using alternative formats as well which may be easier to use and process. Different formats of nutrition information can be distinguished by the amount of information they provide. For example, a star-rating label is easy to use and requires little mental processing. Comparison between products with star-ratings can be reduced to a single metric: the number of stars. The star-rating label, however, provides a small amount of nutrition information. A detailed nutrition label, on the other hand, provides more information regarding nutritional quality. At the same time, however, detailed nutrition labels require a greater amount of time and processing and are therefore more costly. 5 Consider a person who purchases a vector of consumptive goods (x) at prices (p) and faces a budget constraint x ⋅ p = w ⋅ Tw while earning wages w working Tw hours. Each person derives utility from the consumptive goods according to some function f(x). At the same time, the person acquires information (i) which describes the nutritional quality of the vector of consumptive goods. The amount of time available to obtain nutrition information ( ci ), where c is measured as time per units of information, is limited as is leisure time and work time such that 1 = Tw + TL + ci . A person chooses to consume goods and information to maximize their utility subject to their budget and time constraints: max U [αQe , βQc (i ), δQs , f ( x )] i, x s.t. x ⋅ p = w(1 − TL − ci ) (2) Without specifying a functional form, the first order conditions imply that the optimal acquisition of nutrition information is then: i (β , c, TL , w, p, x* ) where x* is the optimal vector of consumptive goods. In equation 2, nutrition information acts as a complement to consumptive goods. Consumers derive utility from the provision and presentation of nutrition information according to how it describes preferred nutritional attributes. Accordingly, the marginal utility of nutrition information is positive such that ∂U x ∂i > 0 , where U x is the marginal utility from consumption. A detailed nutrition label provides more information and therefore has a larger benefit per unit of consumption than an easy-to-use label. At the same time, there is a clear tradeoff between acquiring nutrition information, leisure and work such that ∂i ∂TL , ∂i ∂w , ∂i ∂c < 0 . A nutrition label providing detailed nutrition information may require more time to process than an easy-to-use label. A person facing 6 higher costs of time may consume less of the detailed nutrition information than the easyto-use information. The preferences for credence characteristics ( β ) includes preferences for nutritional characteristics. Shoppers who are high β -types are more likely to benefit from nutrition information and nutrition labels. Put another way, a shopper who has preferences for low fat foods is more likely to prefer nutrition information that identifies this nutritional characteristic. The purpose of this article is to estimate shopper preferences for the presentation and the type of nutrition information being provided. Specifically, we focus on nutrition information provided on grocery store shelf labels. We describe the vector of preferences for nutrition information using a random utility model (RUM). The non-stochastic utility is therefore a function of the attributes of the shelf label. While we are interested in preferences for the provision of nutrition information on grocery store shelf labels, stores also provide price information, and in most cases, unit price information. The price and unit price information may impact the preferences for nutrition information. As such, non-stochastic utility is specified as: V = β1 price label + β 2unit price label + β 3nutrition label (3) The term price label refers to the display of price information, unit price label refers to the display of unit price information, and nutrition label refers to the presentation of nutrition information. Data Choice experiments were conducted to examine shopper preferences for different formats of nutrition information provided on shelf labels. For the choice experiments, we created 7 grocery store shelf labels using varying displays of price and unit price, a written description of nutrition information and a star-rating of nutritional quality. Price information is varied on 2-levels (low and high prominence), unit price on 2-levels (low and high prominence), written nutrition information on 3-levels (not present, low and high prominence), and star-rating nutrition information on 2-levels (not present and present). High prominence refers to bold font; low prominence refers to standard font. All of the labels were produced as color images and displayed below a picture of a can of tomato soup to provide shoppers with a context for examining the labels4. Three label examples are provided (Figure 1). The written nutrition information describes the soup using the terms Fat Free, Saturated Fat Free, and Cholesterol Free and represents detailed nutrition information. These claims are based on NLEA standards for characterizing levels of total fat, saturated fat, calories, cholesterol, sugar and sodium however survey participants were not explicitly informed of the source of the nutrition claims5. The star rating system is a fictitious 4-star scale and is intended to represent easy-to-use information. The formulation of the star-rating is not provided; therefore shoppers are not made explicitly aware of what information the star-rating is representing. Selecting from 24 possible labels (full factorial design: 2x2x3x2), the final Tomato soup was selected to be innocuous. However, there were a handful of survey participants who voiced distaste for tomato soup. 5 Since survey participants were not explicitly informed of the source or reliability of the nutrition claim, preferences may include individual levels of trust. These choice experiments are designed to test preferences for nutrition information provided on shelf label information, and not a specific source of information. If these labels cause survey participants to be skeptical, then preferences may be biased downward. 8 4 experimental design has 16 choice sets comprised of 4 labels per set6. Survey respondents were shown each choice set and asked to select the most preferred label. The choice experiments were administered using surveys at a major retail chain in November 2006 over a three-day period in the East San Francisco Bay area of California7. A total of 600 surveys were administered. Upon completing the survey, participants were given a $10 grocery store gift card. After eliminating incomplete surveys, 410 surveys remained for a total of 6560 choice experiments8. Summary statistics for the demographic variables are presented in Table 1. In addition to demographic variables, each respondent’s level of nutrition consciousness is calculated using the summed value of 9 seven-point Likert scale questions (see the Appendix). We define nutrition consciousness in terms of how likely a person is to pursue a healthy diet9. Methods Based on pre-study trials, 16 choice sets seemed short enough to avoid fatiguing participants while still providing an adequate number of observations. The fourth choice is the same in all of the choice sets: a label with high prominence price and unit price information and no nutrition information. The constant choice label closely resembles the current label of the store where the research was conducted and provides survey participants the option of selecting no label change. An optimal choice experiment was created by maximizing the D-efficiency which improves efficiency by minimizing variance. Survey participants were asked to view each set of four labels and select the one label they prefer most from each set. 7 Interviewers took care to prevent participants from taking the survey more than once. The full survey included more than just the choice experiments and therefore took about 15 minutes to complete. The time requirement also acted as a deterrent to shoppers from taking multiple surveys. The choice experiment alone took less than 5 minutes. 8 Only 13 choice experiments were not fully completed. However, far more surveys had incomplete demographic information. These experiments were removed to avoid biasing the results. 9 If more than two nutrition questions were not answered, the nutrition score was not included. If two or fewer questions were unanswered, the mean value of the answered questions was used to fill in the blank responses. The Cronbach alpha score, which identifies how well a set of questions identify a latent construct was 0.93 indicating that the 9 questions offer a fairly consistent measure of the survey participants perceptions of their own nutrition consciousness. 9 6 The results of the choice experiments are analyzed using a random parameters logit model (RPL)10. Similar to standard logit, RPL models identify non-stochastic utility of person n for choice i at time t ( Vnit ) as being linear in the attributes of any choice xnit such that Vnit = βxnit and ε as a random component of utility such that U nit = Vnit + ε nit . The probability that consumer n chooses alternative j out of choice set Ci at time t, V where Ci = {A, B, C , D} is then Pr{ j} = Pr{ njt + ε njt ≥ Vnkt + ε nkt ; ∀k ∈ Ci }. Assuming that the errors are independently and identically distributed with a type 1 extreme value distribution, then the probability that consumer n choose shelf label j is given by: Pr{ j} = k ∈C ∑ eβ e βx njt x nkt . (4) In an RPL model the vector of preferences β are allowed to vary over some density f β | Ω* where Ω* are the true parameters of the distribution. To estimate β , a distribution of the parameters is designated. Based on this distributional assumption, two estimates of preferences are produced: an estimate of the population mean b and an estimate of stochastic differences in taste η n such that: β n = b + η n . Inserting β n into the random utility function, the random parameters logit model can be usefully expressed as: U nit = bxnit + η n xnit + ε nit . (5) ( ) On the right hand side of equation (5) utility is composed of mean preferences, differences in taste preferences and random error. 10 Random parameter’s logit are a common method of estimation and we do not perform any modifications to the basic simulation methods. As such, a thorough description of RPL is not included here. Interested readers are referred to Revelt and Train (2000) for a detailed description. 10 In our analyses, we first specify a normal distribution for all parameter estimates and then a triangular distribution. The normal distribution is a useful and common distribution and the triangular distribution is use to identify maximum and minimum values. To investigate possible correlation among preferences for attributes we specify β to be correlated between parameter estimates. Accounting for any correlation should improve the estimates and provide additional insight into shopper preferences. Individual differences may also affect preferences for the provision of the nutrition information as well. To identify differences in preferences according to demographic differences, a model is estimated that includes interaction terms. In this article we only focus on the interaction of demographic variables with the preferences for the nutrition labels and not price and unit price labels. Estimation We first estimate a base model with no demographic interaction terms included11. The estimation results of the model (Table 2, columns 1 and 2) provide evidence that the 11 For this article, we rely on Limdep’s NLOGIT 3.0 to estimate our models. Alternative specific constants (ASC’s) are designated for each of the four choices in a set to capture the average effect on utility of factors not explicitly included in the model and to ensure that the model error, ε njt , has a mean of zero. Since only differences in utility matter in a random utility framework, the three ASC’s are calculated relative to one constant choice, in this case, the fourth label. The estimated coefficients of the three ASC’s represent the average affect of utility of a given choice relative to the fourth label. Systematic difference in the unexplained utility between the locations can be accounted for using a scaling parameter. The scaling parameter is estimated relative to a base location (Swait and Louviere) by setting the scaling parameter for the base location to 1 and estimating the remaining scaling parameters relative to the base group using an artificial nested logit (Adamowicz 1998). Location 1 is designated as the base location and has a scaling parameter of 1. Location 2 has a scaling parameter of .99, suggesting that the unexplained variance in location 1 and 2 are fairly similar. Location 3, the largest sample population, had more unexplained variance than location 1 as indicated by the scaling parameter of .84. The scaling parameters are used to scale the data from each location before all of the data are pooled together. 11 display of the label information affects shopper preferences. Low prominence price, unit price and nutrition information all have a negative affect on utility. Alternatively, high prominence nutrition information is significantly positive. Based on a normal distribution, the estimates of the mean and the standard error of the preferences for the high prominence nutrition information suggest that roughly 75 percent of survey respondents have a positive preference for the label. The estimates of the mean and the standard error of the preferences for the star-rating information suggest that roughly 64 percent of survey respondents have a positive preference for the label. As such, we may expect that there is a larger acceptance of the high prominence nutrition information. Using the mean, standard deviation and distributional assumptions we calculate an arc elasticity of demand for the two different formats of shelf label information. While the mean value of the high prominence nutrition label (1.82) is not a true willingness to pay estimate, it does represent the perceived value of the label, which incorporates individual opportunity cost. With a normal distribution, 50 percent of the survey respondents have higher valuation of the high prominence nutrition label. At a one percent increase in the valuation of the label, 49.7 percent of the survey respondents have a higher valuation of the label. That represents an arc elasticity of 0.52. By itself, this elasticity is vague because it represents a change in demand in response to a change in personal opportunity costs as opposed to actual prices. It may be useful, however to compare to the measure of elasticity for the star-rating label. At a one percent increase in the valuation of the star-rating label, 49.9 percent of the respondents have a higher valuation, resulting in an arc elasticity of 0.29. Comparing the two elasticities suggests that given an overall increase in the opportunity cost of time, we might expect that greater 12 numbers would value the high prominence (i.e. detailed information) less. That is, with an increase in the cost to acquire the information, more shoppers will continue have positive preferences for the easy-to-use star-rating label. Intuitively, this makes sense since the more detailed high prominence nutrition label takes more effort to process the information. Although the mean valuation of the high prominence nutrition label is higher than the star-rating label, shoppers may be quicker to move away from using more detailed nutrition information if search becomes more costly. The parameter estimates from the triangular distribution (Table 2, columns 3 and 4) provide a picture of the maximum and minimum values of the preferences. While the mean estimates of the high prominence nutrition label and the star-rating label do not change drastically, it is clear from the estimates of the standard error that preferences are more greatly dispersed for the high prominence nutrition label. While there may be a greater preference for the detailed information, there is also more disparity amongst the population of preferences. We next estimate models using the normal and triangular distribution in which we specify the interaction of demographic variables with the preferences for the nutrition labels (Table 3). Both models have similar performance in terms of their LRI score. The main effect of the high prominence nutrition label is negative in both models, however, adjusting for the nutrition consciousness score results in a positive effect suggesting that those persons who identify themselves as being nutritionally conscious prefer to have the shelf label nutrition information. The significance of the nutrition consciousness interaction, however, may also indicate a social desirability bias. Survey respondents who 13 identify themselves as being nutritionally conscious (whether they really are or not) may also select the shelf label that they expect to be the right choice. While demographic characteristics should influence preferences for the label attributes, in these models few of the other demographic interactions have a significant effect or are inconsistent based on the distributional assumptions. Changes in sign between different distributional assumptions are not uncommon with RPL models and warrant caution in interpreting the results. The shape of the normal distribution relative to the triangular may effect the signing of the interaction terms, calling into question which distribution is most appropriate. If the preferences of the population tend to gather around the mean then the normal distribution may be more appropriate, whereas a larger spread of preferences may suggest the triangular distribution is a better choice. The correlation matrix is calculated for the model using the normal distribution.12 There is significant interaction between preferences for the different label formats (see Table 4). Preferences for the high prominence and low prominence nutrition label are positively correlated. The correlation of the high prominence nutrition label and the star rating label is negative however, which suggests that too much information may be overwhelming for shoppers. Providing the high prominence nutrition label and the starrating label detracts from the label presentation. The low prominence nutrition label and the star rating label, however, are positively correlated. Again, this indicates that the presentation of the nutrition labels impacts shopper preferences and possibly their usage of the labels. Conclusions 12 The correlation matrix was also estimated for the triangular distribution, but the results for the nutrition labels are similar so it is left out. 14 The results of the choice experiments suggest that shoppers have positive preferences for nutrition information provided on shelf labels, with the caveat that presentation also appears to have an effect. The current nutrition facts panels may be greatly underused because they offer too much information and therefore require too much time to use. Other formats for providing nutrition information that require less effort may be more effective. This result is clearly important for the prospect of nutrition labels in other markets as well, such as the food away from home category. While there is an overall stronger mean response for the high prominence nutrition label the star-rating may be all around more effective. The preferences for the star-rating are less elastic suggesting that their use may be less sensitive to changes in opportunity costs. Although this is not covered in this article, the star-rating may also effectively act as a sorting mechanism for food products. That is, a shopper facing a large shopping list or a large set of choices may prefer to use the easier to use star-rating label to initially sort through potential purchases. More detailed nutrition facts panels may then be used to select from a smaller subset of choices. This multi-step process, however, needs to be further examined. Grocery stores can choose to meet shopper demand for nutrition information by providing their own nutrition label format. Stores may gain some level of price control by offering product information to differentiate their products. That is, nutrition labels complement the product being described in such a way that there is perceived higher quality. As such, grocery stores can maximize profits by selecting both the optimal price and the level of information detail to provide describing nutritional quality. If a store provides labels that offer a large amount of detailed nutrition information to shoppers this 15 requires that the nutrition label be very explicit about nutritional content, identifying specific details about nutritional quality similar to a nutrition facts panel. As a consequence, only high quality products are identified with nutrition information since negative nutrition information is generally not used. Put another way, the detailed description only highlights high quality products. This results in a market defined by high quality and low quality products. If a store provides labels that offer low amounts of nutrition information to shoppers the labels are less useful for discriminating between products. This is similar to a store in which all food items are labeled with a star rating to indicate nutritional quality. While the star rating may identify foods of higher overall quality, the rating does not necessarily discriminate between, for example high fat, low sodium foods and high sugar, low fat foods. As such, differences in product quality are less perceptible. The quality discrimination that occurs as a result of providing labels with a large amount of nutrition information may be a less than desirable outcome. With only high quality products having detailed nutrition labels, only shoppers with strong preferences for nutrition information or low opportunity costs are using the nutrition information. These are the same people, however, who are more likely to seek out nutritional information to begin with. As such, the provision of the detailed in-store nutrition information does not affect people who normally do not seek out nutritional information to being with. This is similar to the current marketplace. Nutrition facts panels offer a large amount of detailed information, but only those motivated shoppers are using this information to seek out products with higher nutritional content. 16 Labels offering less information may direct more people to products with higher nutrition quality. Less motivated shoppers may be more likely to use labels that require less time to process and will therefore be more capable of identifying foods with better nutritional quality. If stores provide labels with less information, however, shoppers who prefer a large amount of nutrition information are worse off because they will be underprovided nutrition information by the store. As such, the existence of nutrition facts panels provides an important function for nutritionally conscious shoppers. In terms of public policy, although it often seems that more information is better, too much information may neutralize the intended effect of the information. If the goal is to direct shoppers to healthier purchases, then it is important to consider what format (or amount) of information is likely to direct those shoppers if they are currently not making healthy choices. References Becker, G.S. and K.M. Murphy. 1993. “A simple theory of advertising as a good or bad.” The Quarterly Journal of Economics 108 (4): 941-964. Caswell, J.A. and E.M. Mojduszka. 1996. “Using information labeling to influence the market for quality in food products.” American Journal of Agricultural Economics 78(5): 1248-1253. Caswell, J.A. and D.I. Padberg. 1992. “Toward a more comprehensive theory of food labels.” American Journal of Agricultural Economics 74(May): 460-468. Darby, M. R. and Karni, E. (1973). Free competition and the optimal amount of fraud. Journal of Law and Economics 16: 67–88. 17 Drichoutis, A.C., Lazaridis, P., Nayga, R.M. Jr. 2006. “Consumers’ use of nutritional labels: a review of research studies and issues.” Academy of Marketing Science Review 9:1-22. Golan, E., Kuchler, F., Mitchell, L., Greene, C. and Jessup, A. (2001). Economics of food labelling. Journal of Consumer Policy 24: 117–184. Lancaster, K.J. 1966. “A new approach to consumer theory.” The Journal of Political Economy 74(2): 132-157. Levy, Alan S., Odonna Mathews, Marilyn Stephenson, Janet E. Tenney, and Raymond E. Schucker. 1985. “The Impact of a Nutrition Information Program on Food Purchases.” Journal of Public Policy & Marketing 4 (Spring), 1-13. Martin, Andrew. “The package may say healthy, but this grocer begs to differ.” New York Times, November 6, 2006. Martin, Andrew. “Store chain’s test concludes that nutrition sells.” New York Times, September 6, 2007. McCluskey, J.J. and J.F.M. Swinnen. 2004. “Political economy of the media and consumer perceptions of biotechnology.” American Journal of Agricultural Economics 86(5): 1230-1237. Nelson, P. 1970. “Information and consumer behavior.” The Journal of Political Economy 78(2): 311-329. Nelson, P. 1974. “Advertising as information.” The Journal of Political Economy 82(4): 729-754. 18 Revelt, D. and Train, K. 1998. Mixed logit with repeated choices: Households’ choices of appliance efficiency level. The Review of Economics and Statistics. 80(4): 647657. Teisl, M.F. and B. Roe. 1998. “The economics of labeling: An overview of issues for health and environmental disclosure.” Agricultural and Resource Economics Review 27(2): 140-150. Teisl, M.F., Bockstael, N.E., and Levy, A.S. 2001. “Measuring the welfare effects of nutrition information.” American Journal of Agricultural Economics 83(1): 133149. Verbeke, W. 2005. “Agriculture and the food industry in the information age.” European Review of Agricltural Economics 32(3): 347-368. 19 Table 1. Characteristics of Survey Respondents demographic mean st. deviation age (years) 42.4 16.6 gender (female) 62.0% -household size 3.26 1.56 household shopping performed 69.6% 30% nutrition conciousness score 44.3 12.9 educational level grade school some high school graduated from high school some college annual household income (gross) $0-5,000 $5,001-10,000 $10,001-15000 $15,001-20,000 $20,001-25,000 $25,001-30,000 $30,001-40,000 $40,001-50,000 educational percentage level 2.2% 2-year associate degree 5.6% 4-year bachelor degree 22.9% some graduate school 28.8% graduate degree annual household percentage income (gross) 4.9% $50,001-60,000 4.1% $61,001-70,000 4.6% $70,001-80,000 3.2% $80,001-90,000 5.9% $90,001-100,000 6.6% $100,001-111,000 7.1% $110,001-120,000 11.2% over $120,000 percentage 8.5% 13.2% 5.1% 13.7% percentage 8.3% 9.5% 8.8% 5.6% 3.7% 4.6% 2.4% 9.5% Table 2. Base Model Parameter Estimates Normal Distribution variable mean standard error price: low prominence -0.67** 1.21** unit price: low prominence -1.16** 1.60** low prominence nutrition label -0.62** 1.06** high prominence nutrition label 1.82** 2.76** star rating label 0.51** 1.43** number of surveys 410 LRI 0.474 Note: * and ** indicate significance at .05 and .01 levels respectively Triangular Distribution mean standard error -0.40** 3.2** -1.25** 3.8** -0.67** 2.4** 1.84** 6.22** 0.58** 3.78** 410 0.48 20 Table 3. Parameter Estimates of Model Including Interaction Terms Full Model Parameter Estimates Normal Distribution Triangular Distribution variable mean standard error mean standard error price: low prominence -0.36** 1.42** -0.47** 3.14** unit price: low prominence -0.94** 1.60** -1.04** 3.60** low prominence nutrition label 0.29 0.78** 0.36 2.30** -1.57** 2.14** -1.72** 5.68** high prominence nutrition label -0.42 1.39** -0.86* 3.65** star rating label -0.62** --0.46* -low prom nutr x shopping % 0.77** -0.29 -high prom nutr x shopping % 0.33 -0.12 -star label x shopping % 0.04 -0.01 -low prom nutr x household size -0.03 -0.07 -high prom nutr x household size star label x household size -0.01 -0.06 ---0.03 -low prom nutr x expenditures per cap -0.005 -0.09** -high prom nutr x expenditures per cap 0.03 -0.04* -star label x expenditues per cap 0.01 0.17 --0.1 -low prom nutr x gender (female) high prom nutr x gender (female) -0.42** -0.57** --0.39* -star label x gender (female) -0.32* -0.001 --0.0005 -low prom nutr x age -0.01 --0.01 -high prom nutr x age -0.0005 --0.002 -star label x age low prom nutr x nutrition score -0.01** --0.01* --0.04** -high prom nutr x nutrition score 0.07** 0.02** -0.01 -star label x nutrition score 0.01 -0.02 -low prom nutr x education level 0.05 -0.03 -high prom nutr x education level star label x education level 0.02 -0.06 -number of surveys 410 410 0.48 0.48 LRI Note: * and ** indicate significance at .05 and .01 levels respectively Table 4. Correlation of Parameter Estimates (normal distribution) 1: price 2: unit price 3: low prominence nutrition 4: high prominence nutrition 1 -0.05 0.11 -0.41** 2 0.05 -0.02 3 0.59** 4 5: star rating -0.37** 0.08 0.12* -0.012** Note: * and ** indicate significance at .05 and .01 levels respectively 21 Figure 1. Examples of shelf labels used in conjoint experiments 02 10.7 OZ 119 846 02 119 846 Nutritional Value Rating 75 39024654 0 6554 654 75 39024654 0 6554 654 Campbell’s Soup- Tomato Fat Free Saturated Fat Free Cholesterol Free 99 10.7 OZ Campbell’s Soup- Tomato 45345 ¢ 329.25¢ PER OZ Fat Free Saturated Fat Free Cholesterol Free 99 10.7 OZ ¢ Unit Price 9.25¢ PER OZ 02 Campbell’s Soup- Tomato 45345 ¢ 32PER OZ 9.25¢ 119 846 3496954]]9439 939] BEST 75 39024654 0 6554654 99 BETTER GOOD High price prominence, low unit price prominence, high nutrition information prominence High price prominence, high unit price prominence, high nutrition information prominence Low price prominence, low unit price prominence, star rating label Appendix. Survey questions used to calculate nutrition consciousness scores (all questions bounded by strongly disagree and strongly agree) 1. My diet is nutritionally balanced. 2. I try to monitor the number of calories I consume daily. 3. I try to consume a healthy amount of calories each day. 4. I try to avoid high levels of fat in my diet. 5. I try to avoid high levels of saturated fat in my diet. 6. I try to avoid high levels of cholesterol in my diet. 7. I try to avoid high levels of sodium in my diet. 8. I try to avoid high levels of sugar in my diet. 9. I am interested in nutritional information about the food I eat. 22