Você está na página 1de 9

Food Quality and Preference 31 (2014) 8189

Contents lists available at ScienceDirect

Food Quality and Preference


journal homepage: www.elsevier.com/locate/foodqual

The wine headache: Consumer perceptions of sultes and willingness to pay for non-sulted wines
Marco Costanigro a,, Christopher Appleby a, Stephen D. Menke b
a b

Department of Agricultural and Resource Economics, Colorado State University, Fort Collins, CO 80523-1172, United States Department of Horticulture and Landscape Architecture, Colorado State University, United States

a r t i c l e

i n f o

a b s t r a c t
A panel of 223 alcohol consumers recruited in a liquor store participated in a survey/bestworst experiment investigating perceptions on sultes and willingness to pay for non-sulted wines. We nd that 34% of our sample experiences headaches after consuming moderate amounts of wine, and sultes are the most frequently attributed cause. Based on a rank ordered logit estimation of bestworst choices, headache syndrome sufferers are willing to pay a ceteris paribus premium of $1.23 per bottle to avoid added sultes. However, results from a (logit) model of purchase intentions suggest that quality and price are most important, with differentiating labels (no sulte added, organic) playing only a marginal role. Marketing implications for the wine industry are offered, and negative perceptions toward sultes are contextualized within the hypothesis of a lightning rod effect induced by the contains sultes warning label. 2013 Elsevier Ltd. All rights reserved.

Article history: Received 4 October 2012 Received in revised form 19 July 2013 Accepted 1 August 2013 Available online 21 August 2013 Keywords: Wine marketing Labeling Food labels Sultes Discrete choice experiments Bestworst experiments Willingness to pay

1. Introduction 1.1. Motivation and objectives Added as sulfur dioxide (SO2) or other forms, sultes are commonly used as an antioxidant and antimicrobial agent and, since 1985, the U.S. Food and Drug Administration (FDA) imposes mandatory labeling of the use of sultes as an additive for most foods and beverages. Regulatory action in the Eighties was prompted by an FDA-commissioned study nding that sultes, while safe for most people, could pose a hazard of unpredictable severity to a small (about 1%) population of sulte sensitive consumers (see Papazian, 1996, for a chronology of the events leading to mandatory labeling). Reported symptoms associated with sulte sensitivity range widely in intensity and severity and include trouble breathing, skin rashes, and stomach pain (Grotheer, Marshall, & Simonne, 2005; Vally & Thompson, 2001). As a direct result of the FDA regulation, all wines sold in the United States via interstate commerce include a warning statement if they contain more than 10 ppm of sultes (see also Alcohol & Trade Bureau., 2012). Even though the population of sulte-sensitive consumers is relatively small, the perception that sultes may cause negative health effects appears to be more common. Anecdotal evidence
Corresponding author. Tel.: +1 (970) 491 6948.
E-mail address: marco.costanigro@colostate.edu (M. Costanigro). 0950-3293/$ - see front matter 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.foodqual.2013.08.002

and articles in the popular press suggest that some consumers report experiencing headaches and migraines after consuming small amounts of certain wines, particularly the red varieties (Gaiter & Brecher, 2000; Robin, 2010). Even though the scientic debate regarding what exactly may cause these adverse effects is ongoing (several chemicals have been identied as plausible triggers, see Mauskop & Sun-Edelstein, 2009 and Millichap & Yee, 2003), consumers have been reported to associate migraines and headaches to the presence of sultes (Gaiter, 2000). Despite the widespread use of sultes in the food and beverage industry, we are not aware of any study formally investigating consumers beliefs and perceptions regarding sultes, the central theme of this article. Small amounts of sultes may form naturally in wine during fermentation (Chengchu, Ruiying, & Yi-Cheng, 2006), but vintners commonly add around 3090 ppm of additional sultes throughout production (Burgstahler & Robinson, 1997) to prevent spoilage and enhance aging potential (Goode & Harrop, 2011). While historically uncommon, winemaking without added sultes is becoming increasingly feasible due to better hygiene in the production process and technological improvements, such as refrigerated fermentation in climate-controlled facilities (Goode & Harrop, 2011) or pasteurization via ultraviolet irradiation (see Fredericks, Du Toit, & Krgel, 2011). Low-sulte winemaking in the United States has been so far predominantly synonymous with organic production, as sultes are forbidden by the organic wine production protocol (Alcohol &

82

M. Costanigro et al. / Food Quality and Preference 31 (2014) 8189

Tobacco Tax & Trade Bureau, 2012). Unfortunately, the few studies examining consumers attitudes toward organic wines (e.g. Olsen, Thach, & Hemphill, 2012) provide scarce information regarding perceptions of sultes, mainly because consumers attribute to organic production several other important properties (e.g. better health, taste, nutritional, environmental and social outcomes, which may confound more specic considerations regarding sultes; see Fotopoulos, Krystallis, & Ness, 2003 or Chryssohoidis & Krystallis, 2005). If consumers do in fact worry about sultes, a key aspect for entrepreneurs and winemakers is how much consumers value a minimized sulte level (independently of the other standards imposed by organic production), and what share of consumers would consider such a trait important in their buying decisions. In this article we therefore use a survey and bestworst choice experiment to (1) formally assess consumers perceptions and beliefs toward sultes in wine; (2) quantify willingness to pay (WTP) for non-sulted wines; and (3) identify consumer segments receptive to low-sulte wine marketing. 1.2. Background on choice experiments The basic idea underlying choice experiments is that the process of choosing between alternatives is an obvious and familiar way for consumers to manifest their preferences. When several possible product formulations exist, observing which product proles are preferred implicitly reveals consumers preferences for specic product attributes. Thus, a common approach in choice experiments requires participants to select the best option out of a small number of alternatives listed in several choice sets. By including price as one of the varying product attributes, choice experiments allow estimating the rate at which participants are willing to trade money for the inclusion of one or more attributes (i.e. WTP). The hypothetical nature of discrete choice experiments has limitations (see Lusk & Schroeder, 2004 on hypothetical bias), but the major advantage is allowing the study of products or attributes not yet available in the market. The adoption of specic experimental designs can ensure that the effect of each attribute on preferences can be identied, while simultaneously minimizing the sample size needed to obtain acceptable (i.e., low variance) estimates. This experimental approach also allows researchers to control for exogenous factors that may otherwise invalidate or weaken results (Kroes & Sheldon, 1988), as is often the case for studies using real-world observational data. Both discrete choice experiments and conjoint analyses have been used in studies evaluating the relative importance of product attributes (e.g. Gil & Snchez, 1997; Hu, Batte, Woods, & Ernst, 2012; Mtimet & Albisu, 2006; Onozaka & Thilmany McFadden, 2011), but, as Louviere, Flynn, and Carson (2010) point out, choice experiments have their foundation within the economic random utility framework, while classical conjoint analysis lacks a behaviorally meaningful error term, and is generally inconsistent with economic demand theory. Bestworst methods (Finn & Louviere, 1992), another class of choice experiments, request participants to indicate the most and least preferred items within each choice set, thereby providing a more efcient mean for recovering preferences than a traditional pick one choice experiment (Flynn, Louviere, Peters, & Coast, 2007). In bestworst experiments, researchers can either ask participants to rank products attributes (to study how attributes are ranked and which ones are most important, as in Flynn et al., 2007); or rank multi-attribute options/product proles (to recover the contribution of each individual attribute to the overall utility and WTP for a multi-attribute prole, as in this article and Scarpa, Notaro, Louviere, & Raffaelli, 2011). The interested reader is referred to Marley, Flynn, and Louviere (2008) and Marley and

Pihlens (2012) for an exhaustive treatment of the theoretical and probabilistic properties of the two approaches. 1.3. Wine marketing studies The United States is the largest wine market by sales revenue in the world, representing nearly $32 billion in total retail value (Wine Institute., 2012). In the last 15 years, American wine production has increased 55%, and both total and per-capita wine consumption has expanded every year since 2001 (Wine Institute., 2011a; Wine Institute., 2011b). Though wine remains a highly diversied product, the growing domestic demand for U.S. wines has incentivized industry consolidation (Goodhue, Hein, Green, & Martin, 2008) and a greater degree of uniform production practices within wellknown geographical areas (e.g. Napa Valley). Countering this trend, some producers have begun differentiating their products by focusing on more natural and sustainable production practices (see Goode & Harrop, 2011, on organic and biodynamic wines). Consumer preferences for various intrinsic and extrinsic wine attributes have been investigated in several previous marketing studies, often using choice experiments. For example, Gil and Snchez (1997) varied price, age, and origin and found that, in the absence of other quality cues, origin is the most important wine attribute. Lockshin, Jarvis, dHauteville, and Perrouty (2006) and Mtimet and Albisu (2006) examined how market involvement inuences the valuation of wine attributes such as brand, region of production, quality medals, and aging. In addition to variety and region of production, Jarvis, Mueller, and Chiong (2010) studied the effect of label images and slogans. Experiments considering health-related claims are particularly relevant to our research objectives. This literature generally supports the hypothesis that consumers will pay a premium for wines perceived as being healthier than others. For example, BarreiroHurl, Colombo, and Cantos-Villar (2008) estimate a positive valuation for resveratrol-enriched wine, a health-promoting ingredient. Organic wines are also often perceived as being health-promoting (Barreiro-Hurl et al., 2008; Fotopoulos et al., 2003), and healthconscious consumers are particularly receptive to marketing campaigns promoting natural (and organic) wines (Goode & Harrop, 2011). However, higher valuation for organic wines also comes from environmental concerns, and, as Olsen et al. (2012) argue, the premium for organic wine may be viewed as the nancial self-sacrice made in order to protect the environment. In summary, organic is a multifaceted attribute encompassing numerous consumer values, and consumers may even have difculty explaining why they value organic wine over other varieties (e.g., Barreiro-Hurl et al., 2008). 2. Methods and data 2.1. Sample characteristics Subscribers (older than 21 years of age) to the email list of a large beer, wine and spirits retailer in northern Colorado were contacted via email and offered a $20 wine voucher to be redeemed at the collaborating wine retail store in exchange for their participation in an online survey. The online, anonymous survey was conducted between March 8, 2012 and March 31, 2012, and a total of 223 participants completed the survey. Demographic characteristics of the sample and participants involvement in the wine market are reported in the rst column of Table 1. While the sample may not be representative of the entire U.S. population (participants have higher incomes and education than the national average, which is typical of U.S. college towns), the recruiting strategy was highly successful in targeting

M. Costanigro et al. / Food Quality and Preference 31 (2014) 8189 Table 1 Summary statistics for complete (n = 223) and headache syndrome sufferer (HSS) subsample (n = 76). Characteristic Gender Age Range Male Female 2130 3140 4150 5160 6170 Over 70 Under $25,000 $26,00050,000 $51,00075,000 $76,000100,000 $101,000200,000 Above $200,000 Less than High School High School Some College Bachelors Degree Masters Degree Doctorate/Professional Degree 0 bottles 13 bottles 46 bottles 79 bottles 10 or more bottles 0 bottles 13 bottles 46 bottles 79 bottles 10 or more bottles Complete sample 47.98 52.02 17.49 19.73 14.80 33.18 13.45 1.35 8.52 19.73 16.59 23.77 27.80 3.59 0.00 1.35 15.70 43.95 25.56 13.45 2.24 27.80 32.29 17.49 20.18 2.69 24.22 14.35 8.97 49.78 HSS cohort 32.71 35.34 28.21 27.27 45.45 33.78 40.00 33.33 36.84 40.91 45.95 32.08 25.81 12.50 0.00 66.67 42.86 34.69 33.33 20.00 60.00 33.87 33.33 38.46 28.89 50.00 37.04 40.63 25.00 31.53

83

Income

contain small levels of sultes naturally, usually less than 10 ppm (e.g., Yang et al., 2012). Participants were also informed that sultes, when added in typical quantities (e.g., up to 90 ppm, see Burgstahler & Robinson, 1997) serve as a preservation agent, enhance the aging process, and prevent unwanted fermentation (Goode & Harrop, 2011). Regarding the health impacts and perceptions of sultes, we summarized the ndings of the FDA-commissioned study estimating that 1 in 100 people suffer from a sulte allergy where common side effects include trouble breathing and skin rashes (e.g. Grotheer et al., 2005; Papazian, 1996). Finally, we informed participants that, while some consumers attribute a headache-triggering role to sultes, scientic evidence is inconclusive on whether a link actually exists (Mauskop & Sun-Edelstein, 2009). Following the information points, participants read a standard cheap-talk script (Carlsson, Frykblom, & Lagerkvist, 2005), a brief explanation of the choice tasks, and then began the choice experiment, which took approximately 1525 min to complete. 2.3. Experimental design The selection of an experimental design in choice experiments always implies a tradeoff between the inclusion of all the product attributes of interest to the researcher, and the number of product formulations to be evaluated, which grows exponentially for each attribute/level added. As our main interest lies in estimating WTP for no sultes added, we chose to include the following attributes (see also Table 2): quality (four levels), no sultes added (binary), organic (binary), and price (four levels). To reduce the dimensionality of the design, random blocking was used to account for two other variables likely to interfere with the valuation of sultes: wine variety (red vs. white) and price range. Quality levels were communicated using Wine Spectator (2012) scores, whose expert ratings are commonly reported in wine stores across the United States and have been shown to strongly correlate with wine prices in hedonic studies using observational market data (e.g., Costanigro, McCluskey, & Goemans, 2010). Denitions of scores were communicated (according to Wine Spectators own descriptions) as follows: 95100, Classic: a great wine. 9094, Outstanding: a wine of superior character and style. 8589, Very good: a wine with special qualities. 8084, Good: a solid, well-made wine. 75 79, Mediocre: a drinkable wine that may have minor aws. 5074, Not recommended.

Education

Purchases per month

Bottles of wine currently at home

wine consumers: more than 97% of the sample reported purchasing at least one bottle of wine in a typical month, with the majority of participants (32%) purchasing between four and six bottles. 2.2. Survey and choice experiment layout In the survey, participants were rst asked to report information on demographics and purchasing habits. Then, without directly referencing sultes, participants were asked if they ever experience headaches even after drinking moderate amounts of certain types of wines. Moderate consumption was emphasized to ensure that participants did not confuse their post-consumption headache with the alcohol hangover. While studies have linked light alcohol consumption to hangover headaches and migraines (e.g., Panconesi, 2008), the exact quantity consumed was intentionally left unspecied, as alcohol tolerance is highly weight, gender and race specic (National Institute on Alcohol Abuse, 1995). In cases of afrmative responses to the headache question, participants were asked to choose one or more perceived causes of their headaches from a randomly ordered list of factors (dehydration, histamines, tannins, organic wines, red wines, tyramine, white wine, sultes and other). After this task, participants read stylized information (similarly to the study on WTP for resveratrol in Barreiro-Hurl et al., 2008) regarding the role of sultes in winemaking, and the current medical evidence regarding the effects of sultes on human health.1 Specically, we indicated that all wines
1 There are pros and cons to revealing information before the choice experiment. One objection is that information may bias consumer choices in the following experiment. However, it seems reasonable that, if no sulte added wines were actively marketed; more information would be made available. Our approach here was to rst collect participants original beliefs, and then provide a balanced overview of the existing scientic data currently available on the use and effects of sultes.

Quality levels represent our proxy for a wide range of attributes used by consumers as quality cues (such as designation of origin and/or brand names) excluded from the experimental design to limit the number of products to evaluate. No sultes added and organic were both included to allow a direct comparison of the valuation for conventional low-sulte wine versus full-organic. As valuation of certain wine attributes may vary across price ranges (see Costanigro, McCluskey, & Mittelhammer, 2007), wine prices were made to vary in three increments of $1.50 from three base levels: $10.49, $20.49 and $30.49. A similar random blocking strategy was used to account for the eventuality that attributes, and especially the lack of sultes, might be valued differently in red versus white wines. That is, each participant was randomly assigned to a cabernet sauvignon scenario, or to a chardonnay scenario. This blocking strategy also accounts for the fact that attribute valuation may differ depending on the consumption context (Hall, OMahony, & Lockshin, 2001), at least to the extent that the combination of price ranges and wine varietals identify different occasions of consumption. To summarize, price levels, quality, sultes and organic labeling varied for each participant, while price

84

M. Costanigro et al. / Food Quality and Preference 31 (2014) 8189

Table 2 Experimental design: attributes, levels and coding (in parenthesis) for the discrete choice experiment. Attribute USDA-certied organic label No Sultes added label Quality score Price
a

Level 0 No (0) No (0) 80 (0) Floor price (FPa) (0)

Level 1 Yes (1) Yes (1) 84 (1) FPa + $1.50 (1)

Level 2 88 (2) FPa + $3.00 (2)

Level 3 92 (3) FPa + $4.50 (3)

Floor price is $10.49, $20.49 or $30.49 depending on random block assignment.

range and wine variety varied only across participants, but not within. A main effects, fractional factorial optimal in the differences orthogonal design (OOD, see Street, Burgess, & Louviere, 2005) was used to obtain 36 unique product formulations to be compared in 12 choice sets of 3 alternative wines each (Ngene was the software of choice). Orthogonal designs eliminate collinearity issues in linear regression models, and fractional factorial designs are often preferred to full factorial designs as they substantially reduce the number of products to evaluate. OOD designs simultaneously impose orthogonality between attribute levels and the additional constraint that attribute levels are never the same for alternative products within the same choice set, forcing participants to make tradeoffs between attributes. In each choice set, participants were asked to select their most preferred and least preferred wine label out of the three. To avoid any possible path dependency or the effect of survey fatigue, the order of the choice sets was randomly determined for each participant (Qualtrics was the adopted online survey tool). Many choice experiments also include opt-out options where participants can decide not to select any of the alternatives (e.g. Barreiro-Hurl et al., 2008; Hu et al., 2012; Onozaka & Thilmany McFadden, 2011), allowing participants to signal a would not purchase intention or perhaps indifference between choices. As the adopted OOD design maximizes the differences in attributes across the products within each choice set (and price and quality are included as product attributes), participants could reasonably be expected to have ordinal preferences over the wines, but may still not want to purchase any of them. A reasonable objection, therefore, is that preferences are relatively meaningless if they do not correspond to real demand (i.e. product purchase) or at least to consumers who actively participate in a given market. In our case, study participants were recruited among the customers of a wine store and, since the sample is drawn from the relevant consumer population, all preferences expressed in the bestworst task are relevant. Nonetheless, after each best worst task and with the products still displayed on the screen, participants were asked if they would actually purchase their most preferred wine2 (a screenshot showing a representative choice set is shown in Fig. 1). This strategy allows recovering information on relative preferences (from the bestworst) and purchasing intentions (from the buy-no buy question) while limiting survey length. The reader is referred to Carlsson, Frykblom, and Lagerkvist (2007) for a more general discussion of the pros and cons of including opt-out options in choice experiments. In short, the experimental setup allows investigating how preferences and purchasing intentions vary depending on wine quality, price, and organic/non-sulted production. In addition, formal testing of structural changes in consumer behavior induced by

Fig. 1. Screenshot of a choice set in the $10.4914.99 price block.

different price ranges and wine variety (red vs. white) is made possible via random blocking. 3. Theory/estimation procedure 3.1. Introduction to random utility theory and choice models The random utility model assumes that the total utility (an ordinal, rather than cardinal concept) individual i obtains from choosing product j is divided in a deterministic part inuenced by factors known to the researcher, Vij, and a random or stochastic component eij. In most empirical applications, it is assumed that Vij is well approximated by a linear (in parameters) function of known product characteristics xij (here determined by the experimental design) and a conforming vector of unknown parameters b (i.e. the attributes weights or marginal utilities) to be estimated. That is, the utility from option j is U ij V ij eij x0ij b eij ; and decisions arise from the process of weighting the pros and cons bundled within each alternative product. As choice experiments reveal only which products are preferred, rather than their utility (utility is a latent, unobservable construct), maximum likelihood estimation of the model parameters requires establishing a plausible mathematical relationship between observable product attributes and the probability that a given product is preferred over the others. Such expressions always imply specic distributional assumptions about eij, and assumptions vary depending on the empirical application. In this article, we use the rank-ordered logit (Beggs, Cardell, & Hausman, 1981) to model participants choices in the bestworst tasks, while a logit model was used to model purchasing intentions. 3.2. Rank ordered logit, logit, and panel models

2 Note that the data obtained from the second choice task is not equivalent to a typical choice experiment, as the buy/no buy choices are conditioned on the previous sorting. As an anonymous reviewer pointed out, this may lead to collinearity issues in the second stage. While this did not seem to be a problem in our specic application, we cannot exclude this possibility for other applications and/or experimental designs.

Within each choice set, the bestworst task requires that participants identify an ordinal ranking of preferences between three options. That is, if item A is selected as best, and item B as worst, the implied ranking will be U ijA > U ijC > U ijB , which can be

M. Costanigro et al. / Food Quality and Preference 31 (2014) 8189

85

represented by triplet ACB. Beggs et al. (1981) show that, under the assumption that eij are iid draws from the extreme value distribution, the probability of observing a specic triplet can be expressed in the form probA; C ; B P
expx0 b ijA
jABC

4. Results 4.1. Headache syndrome sufferers sulte perceptions More than a third of the total sample (34.08%), surprisingly, reported experiencing headaches after consuming even moderate amounts of certain wines. We refer to this group as the headache syndrome sufferers, or HSS. The second column of Table 1 illustrates how headache experiences vary across different demographic and market involvement segments. Two patterns are worth noting: rst, the perception becomes less common as education level increases. Second, HSS appears more common with consumers who buy and store wine less often, suggesting that HSS may discourage some alcohol consumers from purchasing wine. In support of our motivating hypothesis, the perception that sultes are the main culprit for the headache syndrome is rather widespread. When asked to choose the three most important triggers of headaches, HSS participants included sultes 63.16% of the times, followed by dehydration (57.89%), red wine (32.89%) and Tannins (20%).

expx0 b ij

expx0 b ijC
jBC

expx0 b ij

. This probabi-

listic statement provides the template to derive the likelihood function of the whole observed sample and implement estimations of the model parameters. Similarly, a participant will be willing to purchase his/her preferred product in a given choice set (A in the previous example) if U ijA > 0, i.e. the utility from purchasing the wine (and paying the associated price) is greater than zero. Analogously to its rank-ordered version, the logit model of a discrete choice between two alternatives hinges on the expression probPurchase yes 1expij . x0 b
ij

expx0 b

When participants complete multiple choice tasks (as in this case) over the course of an experiment, sequential choices from the same individuals cannot be reasonably assumed to be independent of each other. Indeed, assuming independence would lead to overstating the information content of the dataset and produce (downward) biased MLE estimates of the standard errors. To account for this lack of independence, we estimate a clustered (on each participant i) variancecovariance matrix for the rank ordered logit, and a (population averaged) panel logit (see Rabe-Hesketh & Skrondal, 2008, pp. 249249 and pp. 300301) to model purchasing intents.

4.2. Preferences for non-sulted and organic wines: rank-ordered logit results Estimated parameters for the aggregated rank ordered logit in Eq. (1) are presented in Table 3. All coefcient signs conform to a priori expectations: since the dependent variable was coded as one for the most preferred and three for the least preferred wine in any given choice set, negative coefcients imply a decrease in ranking (e.g. from 2 to 1) or, equivalently, an increase in utility associated with the presence of an attribute. Estimated standard errors are quite small, and all coefcients are signicant at the 1% level, perhaps validating the choice of attributes, experimental design, and sample size. Based on these aggregate estimates, the per bottle premium for non-sulted wine is WTPsulfite bsulfite =bprice $1:5 0:2748=0:6409 $1:5 $0:64 per bottle, while the added value of organic winemaking is $1.22, a nding consistent with the idea that organic and non-sulted wines are differentiated attributes, but organic evokes a richer and more complex set of values. Even though the magnitude of these estimates is economically signicant, we also nd that a 4-point increase in the quality score induces a $2.84 increase in WTP, indicating that quality differences inuence wine choices much more than the presence of organic or no added sultes labels. Structural changes of parameter estimates based on random blocking were investigated rst. While some changes in parameters are detectable across price ranges and variety, differences in estimates are not statistically signicant from zero in joint tests (W = 6.23, p = 0.622; and W = 2.65, p = 0.618), indicating that model parameters are stable and can be estimated jointly regardless of blocks. Several other segmented models including demographic variables were estimated and tested for signicance,3 but for the sake of brevity we focus our exposition to the comparison of WTP conditional on headache experience (Eq. (2)). Statistical testing revealed that preferences of HSS and non-HSS participants differ across the two consumer segments (W = 14.09,
3 Estimates and tests are compiled in an appendix, available from the authors upon request. Many comparisons (gender and many other variables measuring wine purchasing habits) yielded no signicant differences in preferences. Two prominent exceptions, where parameters differ signicantly, are education and income levels. WTP for non-sulted wines appears lower for high-education participants, while no clear trend could be established for income. These dimensions could be further investigated in a larger study including a sample more representative of the general US demographics.

3.3. Model parameterizations and WTP estimates The simplest parameterization of our model assumes that a participants utility from selecting a wine depends solely on the attributes and their levels, that is:

x0j b b1 Organicj b2 Sulfitej b3 Qualityj b4 Pricej

Average willingness to pay for each attribute can be calculated as a function of model parameters according to WTP ba =b3 ; where a = 1, 2, 3 depending on the attribute of interest, and b3 is the estimated price coefcient (Barreiro-Hurl et al., 2008; Louviere & Islam, 2008), or the marginal disutility attributable to an increase in product price. Our interest also lies in assessing whether the random blocks (red vs. white variety, price range) affect valuation estimates, or if people who consider wine to be a trigger of headache are willing to pay more for non-sulted wines. These factors are categorical and invariant across alternative choices for each participant, and their effect can be measured by specifying appropriately interacted models, or, equivalently, by estimating separate model parameters for each segmenting factor. For the simple case of a dichotomous variable, say headache experience, one can   1 if headache dene the two indicator variables IHEAD , 0 if no headache   0 if headache and estimate the partitioned and INOHEAD 1 if no headache model

V ij x0j b1 IHEAD x0j b2 INOHEAD : i i

Under this framework, the hypothesis that preferences of headache sufferers and general population are the same can be tested post-estimation via a simple Wald test (implied null hypothesis H0 : b1 b2 ; is often referred to as structural change or Chow test).

86

M. Costanigro et al. / Food Quality and Preference 31 (2014) 8189

Table 3 Estimates for rank-ordered logit, aggregated model (Eq. (1)). Variable Organic Sulte Quality Price Coefcient 0.5196 0.2748 1.2139 0.6409 Robust st. error 0.0538 0.0578 0.0531 0.0344 Z 9.65 4.75 22.86 18.65 p-Value 0.000 0.000 0.000 0.000 95% condence interval 0.6252 to 0.4141 0.3882 to 0.1614 1.3180 to 1.1098 0.5735 to 0.7083

Table 4 Estimates for rank-ordered logit, segmented model (HSS vs. non HSS). Consumer segment Headache Variable Organic Sulte Quality Price Organic Sulte Quality Price Coefcient 0.399 0.502 1.156 0.6121 0.593 0.146 1.254 0.6631 Robust st. err. 0.0805 0.126 0.091 0.0604 0.0695 0.0542 0.065 0.0418 Z 4.96 3.98 12.71 10.14 8.54 2.7 19.3 15.86 p-Value 0.000 0.000 0.000 0.000 0.000 0.007 0.000 0.000 95% condence interval 0.5569 to 0.2413 0.7487 to 0.2546 1.3347 to 0.9779 0.4938 to 0.7305 0.7297 to 0.4572 0.2527 to 0.0402 1.3810 to 1.1264 0.5812 to 0.7450

No headache

Wald test for structural break headache vs. no headache: W = 14.09, p = 0.0070.

Table 5 Willingness to pay estimates from aggregate and headache-segmented models (parameter estimates are in Tables 4 and 5, respectively). MVariable Aggregate Variable Organic Sulte Quality Organic Sulte Quality Organic Sulte Quality WTP ($) 1.2162 0.6431 2.841 0.978 1.2292 2.8334 1.3424 0.3313 2.836 St. error 0.1465 0.1394 0.2091 0.2153 0.3193 0.373 0.1898 0.1242 0.2511 Z 8.3 4.61 13.59 4.54 3.85 7.6 7.07 2.67 11.3 p-Value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.008 0.000 95% condence interval 0.92911.5032 0.36990.9164 2.43123.2508 0.55601.4000 0.60341.8550 2.10263.5643 0.97041.7144 0.08780.5748 2.34393.3281

Segmented Headache

No Headache

Table 6 Effects of wine attributes on purchasing intentions. Marginal effects estimated from population-averaged binomial logit model. Variable Organic Sulte Quality Price Marginal effects 0.0186 0.0172 0.0571 0.0510 St. error 0.0116 0.0114 0.0071 0.0052 Z 1.61 1.51 8.08 9.87 p-Value 0.108 0.131 0.000 0.000 95% condence interval 0.0041 to 0.0414 0.0051 to 0.0394 0.0432 to 0.0709 0.0612 to 0.0409 Avg. 0.6431 0.5960 2.4563 0.9294

p = 0.007). Parameter estimates are presented in Table 4 (Eq. (2)), while Table 5 summarizes WTP estimates for aggregate and headache segmented models. Results indicate that HSS consumers have a willingness to pay for non-sulted wines of $1.23, considerably higher than the $0.33 WTP from non-HSS participants. Curiously, headache sufferers are willing to pay more for wines without added sultes than for organic wine, which is at odds with the fact organic wines are necessarily sulte-free (of course, some consumers may not be aware of this). 4.3. Models of purchasing decision: population-averaged logit results Table 6 presents the estimates for the binomial logit model (with would buy/would not buy designated as the dependent variable) analyzing the determinant of purchasing decisions based on the aggregate model parameterization in Eq. (1). For simplicity, we report marginal effects (rather than the parameter estimates), which can be immediately interpreted as the effect of a marginal increase in a given attribute on the probability of purchasing a

wine. For the whole sample and ceteris paribus, the organic label is found to increase the probability of purchase by 1.86%, and the no added sultes similarly increases the probability of purchase by 1.72%. While these effects have the expected sign (a positive effect on purchasing intentions), they are rather small in magnitude and neither of them is statistically signicant at conventional levels. On the contrary, a 4-point increase in the quality score increases the probability of purchase by 5.71%, and a $1.50 increase in price decreases the probability of purchase by 5.10% (both statistically signicant at 1%). The prominence of price and quality in inuencing purchasing decisions (and therefore demand) becomes even more evident when the effect of the price range blocking is analyzed in more detail. Table 7 presents the results of a model similar to the previous, but also including dummy variables for the $2025 range as well as the $3035 range. Participants in the $2025 price range block were 22% less likely (than the $1015 block) to indicate they would purchase their preferred wines, while the $3035 block was 33% less likely. When comparing the HSS and non HSS consumer segment (Table 8), we still nd

M. Costanigro et al. / Food Quality and Preference 31 (2014) 8189 Table 7 Effects of wine attributes and price range blocks on purchasing intentions. Marginal effects estimated from population-average binomial logit model (aggregate). Variable Organic Sulte Quality Price PRANGE 20a PRANGE 30b
a b

87

Marginal effects 0.0189 0.0187 0.0635 0.0541 0.2206 0.3340

St. error 0.0127 0.0125 0.0079 0.0057 0.0676 0.0648

Z 1.49 1.50 8.06 9.46 3.27 5.15

p-Value 0.137 0.134 0.000 0.000 0.001 0.000

95% condence interval 0.0060 to 0.0438 0.0057 to 0.0431 0.0481 to 0.0790 0.0654 to 0.0429 0.3530 to 0.0882 0.4611 to 0.2070

Avg. 0.6431 0.5960 2.4563 0.9294 0.3318 0.3363

PRANGE 20 = 1 if $20.4924.99 price group, 0 otherwise. PRANGE 30 = 1 if $20.4924.99 price group, 0 otherwise.

Table 8 Estimates for logit model of purchasing decision, HSS vs. non-HSS segmentation.a Segment Headache Variable Organic Sulte Quality Price Organic Sulte Quality Price Marginal effects 0.0218 0.0341 0.0641 0.0555 0.0172 0.0097 0.0542 0.049 St. error 0.0193 0.0195 0.0117 0.0088 0.0139 0.0135 0.0083 0.0062 Z 1.13 1.74 5.48 6.35 1.24 0.72 6.53 7.86 p-Value 0.257 0.081 0.000 0.000 0.217 0.472 0.000 0.000 95% condence interval 0.0159 to 0.0596 0.0042 to 0.0724 0.0412 to 0.0870 0.0726 to 0.0384 0.0101 to 0.0445 0.0168 to 0.0363 0.0380 to 0.0705 0.0612 to 0.0368 Avg. 0.216 0.212 0.833 0.323 0.427 0.383 1.622 0.605

No Headache

Wald test for structural break headache vs. no headache: W = 1.98, p = 0.7391.

that for both groups price and quality are of utmost importance. HSS participants are 3.41% more likely to purchase a wine made without added sultes (p = 0.081), while the analogous effect for non-HSS consumers is essentially zero. 5. Discussion and conclusions 5.1. Main ndings Thirty-four percent of a sample of 223 participants recruited in a wine, beer and spirits store in northern Colorado reported experiencing headaches after consuming moderate amounts of certain wines, with the majority of HSS consumers identifying sultes as one of the principal causes. A bestworst choice experiment was conducted to establish how quality, prices and organic/no added sultes labeling inuence consumers wine preferences. Econometric analysis shows that HSS consumers, after being informed about sultes in wine, are willing to pay a ceteris paribus premium of $1.23 to avoid them. However, we also nd that the same group is only slightly (3.4%) more likely to purchase a wine if it does not contain added sultes. More generally, the decision to purchase a wine is largely determined by price range and quality, with little inuence from organic and sultes labeling. Based on these ndings, one hypothesis worthy of future research is that consumers might rst establish desirable quality and price levels, and then evaluate tradeoffs between more marginal attributes (such as organics and/or no added sultes). 5.2. Consumer perceptions of sultes A clear result of this study is that a sizable number of participants experience negative health effects after consuming moderate amounts of wine. While the survey shows that HSS consumers commonly attribute headaches to the presence of sultes, the etiology of these symptoms may or may not be sulte sensitivity, as, at the present moment, we were not able to nd a medical study providing conclusive evidence on the matter. Several hypotheses and considerations can be made to contextualize and understand consumers negative perception of sultes.

A rst possibility is that consumers were primed to think that sultes are generally dangerous by the early media attention to cases of allergic reactions occurred during the Eighties (when sultes were unlabeled and even used in raw fruits and vegetables). This interpretation, however, counters other existing studies showing how media scare effects, while signicant in the short period, tend to vanish over time (see Piggott & Marsh, 2004, on food safety events and meat demand). Another hypothesis, perhaps more fascinating, is that the contains sultes warning may itself induce or reinforce negative perceptions. That is, the presence of a warning label may act as catalyst for all negative experiences, a lighting rod effect where all attention is directed toward sultes because of the mandatory labeling. Analogous effects have been documented for positive health claims inducing a halo or silver bullet effect, a behavioral phenomenon where people attribute excessive or inappropriate health benets to a product because of the presence of an attribute (see Roe, Levy, & Derby, 1999). The implications of such effects for labeling policy are noteworthy and deserving of further research. Can truthful labels induce unsubstantiated beliefs? As of 2013, nearly half U.S. states have or are considering bills to require labeling of GM foods. If propositions are passed into law, is the choice between labeling genetically modied food as GM-free versus contains GM an important one? The lightning rod hypothesis seems worthy of further testing, but we also note that sultes are present (and labeled) in many other foods and beverages. For example, one wonders why there does not seem to be much consumer or research attention directed toward sultes in, say, dried fruit. One explanation is that consumers may be more likely to notice sultes warnings when purchasing high involvement products, such as wine, which are likely to induce a special focus on product labels. Thus, the interaction between warning labels, possible adverse effects, and varying level of consumer involvement might deserve future consideration. Explicit modeling of unknown heterogeneity in beliefs and motivations regarding sultes via random parameters or other methods could yield more clarity, but such methods necessitate signicantly larger sample sizes (see the results of Chang & Lusk, 2011, in Monte Carlo simulations).

88

M. Costanigro et al. / Food Quality and Preference 31 (2014) 8189

5.3. Wine marketing opportunities The HSS cohort comprises a surprisingly large share of the consumer population, and therefore represents an obvious, potentially important market segment for wine producers interested in product differentiation. Based on our ndings, one recommendation is that organic producers should emphasize more how the organic protocol forbids the use sultes, as participants did not seem to be aware of this requirement. However, budget constraints imposed a main-effects-only experimental design, so the nuances of how organic and no sultes added labels interact remain unexplored. Regardless of these possible interactions, non-sulted wines may enable wineries to target the natural and preservative free consumer movements without having to abide to the full, more costly organic certication protocol. Before investing signicant resources in such initiatives, wineries should carefully consider several caveats. First and foremost, non-sulted wines will be unlikely to succeed if the production method somehow compromises quality. In fact, consumers may be deterred from purchasing a low-sulte wine if they associate them with lower quality. Offering a money-back guarantee would reassure consumers but also shift the risk to the producer. In the end, alcohol consumers who feel strongly about sultes always have the option of drinking beer and/or spirits (which rarely contain added sultes), and eliminate wine altogether. To determine protability, producers should also consider the additional costs (if any) involved with alternative preservation technologies, and compare it with the premium for non-sulted wine. Our estimates, it is important to note, relate to a sample of consumers in Colorado and may not be applicable to different markets. Another long-term factor to consider is that valuation of nonsulted wines is, based on our results, conditional on attributing negative health effects to sulte residues. If sultes are not at the root of the problem (as it may well be), the premium for non-sulted wines may vanish, at least in the long run. Based on these considerations, no sultes added marketing may be more appropriate for wines not necessitating extended aging (which increase the chances of spoilage), and/or less likely to contain other chemicals currently shortlisted as possible causes of the headache syndrome. Perhaps more importantly, we nd clear evidence that good quality wines with low potential to induce headache syndrome will grant access to a substantial niche of consumers. Investment in medical research to establish the chemical root of the problem promises signicant rewards for the wine industry. Disclosure statement We acknowledge funding from Surepure ($5000), a producer of equipment for turbid liquid photopurication, to partially cover participants incentives and expenses, and from Wilburs Total Beverage for providing a 10% discount for the survey participation vouchers, in addition to access to their customer email list. The authors have no afliation nor were compensated in any other way by the mentioned companies, nor were the companies directly involved in conducting the study, analyzing the data and the writing of this article. IRB The research was approved on January 20, 2012 by the IRB Coordinator of the Research Integrity & Compliance Review Ofce, Colorado State University. IRB ID: 13112H.

References
Alcohol and Tobacco Tax and Trade Bureau. (2012). Wine labeling resources. Last accessed June 2, 2012 from <http://www.ttb.gov/wine/wine-labeling.shtml>. Barreiro-Hurl, J., Colombo, S., & Cantos-Villar, E. (2008). Is there a market for functional wines? Consumer preferences and willingness to pay for resveratrolenriched red wine. Food Quality and Preference, 19(2008), 360371. Beggs, S., Cardell, S., & Hausman, J. (1981). Assessing the potential demand for electric cars. Journal of Econometrics, 17(1), 119. Burgstahler, A. W., & Robinson, M. A. (1997). Fluoride in California wines and raisins. Fluoride, 30(3), 142146. Carlsson, F., Frykblom, P., & Lagerkvist, C. J. (2005). Using cheap talk as a test of validity in choice experiments. Economic Letters, 89(2005), 147152. Carlsson, F., Frykblom, P., & Lagerkvist, C. J. (2007). Consumer Benets of Labels and Bans on GM FoodsChoice Experiments with Swedish Consumers. American Journal of Agricultural Economics, 89(1), 152161. Chang, J. B., & Lusk, J. L. (2011). Mixed logit models: Accuracy and software choice. Journal of Applied Econometrics, 26(1), 167172. Chengchu, L., Ruiying, C., & Yi-Cheng, S. (2006). Bactericidal effects of wine on vibrio parahaemolyticus in oysters. Journal of Food Protection, 69(8), 18231828. Chryssohoidis, G. M., & Krystallis, A. (2005). Organic consumers personal values research: Testing and validating the list of values (LOV) scale and implementing a value-based segmentation task. Food Quality and Preference, 16(7), 585599. Costanigro, M., McCluskey, J., & Goemans, C. (2010). The economics of nested names: Name specicity, reputations, and price premia. American Journal of Agricultural Economics, 92(5), 13391350. Costanigro, M., McCluskey, J., & Mittelhammer, R. (2007). Segmenting the wine market based on price: Hedonic regression when different prices mean different products. Journal of Agricultural Economics, 58(3), 454466. Finn, A., & Louviere, J. J. (1992). Determining the appropriate response to evidence of public concern: The case of food safety. Journal of Public Policy & Marketing, 11(2), 1225. Flynn, T., Louviere, J., Peters, T., & Coast, J. (2007). Bestworst scaling: What it can do for health care research and how to do it. Journal of Health Economics, 26(2007), 171189. Fotopoulos, C., Krystallis, A., & Ness, M. (2003). Wine produced by organic grapes in Greece: Using means-end chain analysis to reveal organic buyers purchasing motives in comparison to the non-buyers. Food Quality and Preference, 14(2003), 549566. Fredericks, I. N., Du Toit, M., & Krgel, M. (2011). Efcacy of ultraviolet radiation as an alternative technology to inactivate microorganisms in grape juices and wines. Food Microbiology, 28(3), 510517. Gaiter, D.J., & Brecher, J. (2000). Why do I get headaches from wine? Wall Street Journal. Last accessed June 2, 2012 from <http://guides.wsj.com/wine/winetips-and-tricks/why-do-i-get-headaches-from-wine/>. Gil, J. M., & Snchez, M. (1997). Consumer preferences for wine attributes: A conjoint approach. British Food Journal, 99(1), 311. Goode, J., & Harrop, S. (2011). Authentic wine: Toward natural and sustainable winemaking. Berkeley, CA: The Regents of the University of California. Goodhue, R., Hein, D., Green, R., & Martin, P. (2008). California wine industry evolving to compete in 21st century. California Agriculture, 62(1), 12. Grotheer, P., Marshall, M., & Simonne, A. (2005). Sultes: Separating fact from ction. Department of Family, Youth, and Community Sciences, University of Florida, publication number FCS8787. Last accessed June 2, 2012 from <http:// edis.ifas.u.edu/fy731>. Hall, J., OMahony, B., & Lockshin, L. (2001). Exploring the links between wine choice and dining occasion: Factors of inuence. International Journal of Wine Marketing, 13(1), 3649. Hu, W., Batte, M. T., Woods, T., & Ernst, S. (2012). Consumer preferences for local production and other value-added label claims for a processed food product. European Review of Agricultural Economics, 39(3), 489510. Jarvis, W., Mueller, S., & Chiong, K. (2010). A latent analysis of images and words in wine choice. Australasian Marketing Journal, 18(2010), 138144. Kroes, E. P., & Sheldon, R. J. (1988). Stated preference methods: An introduction. Journal of Transport Economics and Policy, 22(1), 1125. Lockshin, L., Jarvis, W., dHauteville, F., & Perrouty, J. (2006). Using simulations from discrete choice experiments to measure consumer sensitivity to brand, region, price, and awards in wine choice. Food Quality and Preference, 17(2006), 166178. Louviere, J., Flynn, T., & Carson, R. (2010). Discrete choice experiments are not conjoint analysis. Journal of Choice Modelling, 3(3), 5772. Louviere, J., & Islam, T. (2008). A comparison of importance weights and willingness-to-pay measures dried from choice-based conjoint, constant sum scales and bestworst scaling. Journal of Business Research, 61(2008), 903911. Lusk, J. L., & Schroeder, T. C. (2004). Are choice experiments incentive compatible? A test with quality differentiated beef steaks. American Journal of Agricultural Economics, 86(2), 467482. Marley, A. A. J., Flynn, T. N., & Louviere, J. J. (2008). Probabilistic models of setdependent and attribute-level bestworst choice. Journal of Mathematical Psychology, 52(5), 281296. Marley, A. A. J., & Pihlens, D. (2012). Models of bestworst choice and ranking among multiattribute options (proles). Journal of Mathematical Psychology, 56(1), 2434. Mauskop, A., & Sun-Edelstein, C. (2009). Food and supplements in the management of migraine headaches. The Clinical Journal of Pain, 25(5), 446452.

M. Costanigro et al. / Food Quality and Preference 31 (2014) 8189 Millichap, J. G., & Yee, M. M. (2003). The diet factor in pediatric and adolescent migraine. Pediatric Neurology, 28(1), 915. Mtimet, N., & Albisu, L. M. (2006). Spanish wine consumer behavior: A choice experiment approach. Agribusiness, 22(3), 343362. National Institute on Alcohol Abuse and Alcoholism. (1995). No. 28 PH 335. Olsen, J., Thach, L., & Hemphill, L. (2012). The impact of environmental protection and hedonistic values on organic wine purchases in the US. International Journal of Wine Business Research, 24(1), 4767. Onozaka, Y., & Thilmany McFadden, D. (2011). Does local labeling complement or compete with other sustainable labels? A conjoint analysis of direct and joint values for fresh produce claims. American Journal of Agricultural Economics, 93(3), 693706. Panconesi, A. (2008). Alcohol and migraine: Trigger factor, consumption, mechanisms. A review. Journal of Headache and Pain, 9(1), 1927. Papazian, R. (1996). Sultes: Safe for most, dangerous for some. FDA Consumer Magazine, 30(10). Piggott, N. E., & Marsh, T. L. (2004). Does food safety information impact U.S. meat demand? American Journal of Agricultural Economics, 86(1), 154174. Rabe-Hesketh, S., & Skrondal, A. (2008). Multilevel and longitudinal modeling using stata (2nd ed.). Stata Press. Robin, S. (2010). What are the causes of a wine headache? Livestrong. Last accessed July 21, 2012 from <http://www.livestrong.com>. Roe, B., Levy, A., & Derby, B. (1999). The impact of health claims on consumer search and product evaluation outcomes: Results from FDA experimental data. Journal of Public Policy & Marketing, 18(1), 89105.

89

Scarpa, R., Notaro, S., Louviere, J., & Raffaelli, R. (2011). Exploring scale effects of best/worst rank ordered choice data to estimate benets of tourism in alpine grazing commons. American Journal of Agricultural Economics, 93(3), 813828. Wine Spectator. (2012). Wine spectators 100-point scale. Last accessed June 3, 2012 from <http://www.winespectator.com/display/show/id/scoring-scale>. Street, D. J., Burgess, L., & Louviere, J. J. (2005). Quick and easy choice sets: Constructing optimal and nearly optimal stated choice experiments. International Journal of Research in Marketing, 22(4), 459470. Vally, H., & Thompson, P. J. (2001). Role of sulte additives in wine induced asthma: Single dose and cumulative dose studies. Thorax, 56, 763769. Wine Institute. (2011a). US/California wine production. Last accessed June 2, 2012 from <http://www.wineinstitute.org/resources/statistics/article83>. Wine Institute. (2011b). Wine consumption in the U.S. Last accessed June 2, 2012 from <http://www.wineinstitute.org/resources/statistics/article86>. Wine Institute. (2012). 2011 California wine shipments within U.S. reach all-time high: U.S. is worlds largest wine market. Last accessed June 2, 2012 from <http:// www.wineinstitute.org/resources/pressroom/03222012>. Yang, W., Guner, S., Rock, C., Anugu, A., Sims, C., & Gu, L. (2012). Prospecting antioxidant capacities and health-enhancing phytonutrient contents of southern highbush blueberry wine compared to grape wines and fruit liquors. Sustainable Agriculture Research, 1(1).

Você também pode gostar