I. Introduction
Although it is widely known that wine has been produced for thousands of years, it is less well known among ordinary consumers that wine is not always made from grapes. Although it is not classified as wine, mead is created through a similar process from fermenting honey and water (sometimes with various fruits and spices) and has been produced in Europe, Africa, and Asia as far back in time as wine (Toussaint-Samat, Reference Toussaint-Samat2009). Wherever grapes were scarce, making wines from other fruits blossomed (Schneider, Reference Schneider2007); in modern times, “fruit wine” has come to mean any fermented alcoholic beverage made from any plant matter that can be fermented. This definition usually excludes beer, mead, cider, and perry, which historically have constituted separate categories.
The market size for fruit wines is not easy to determine, because most fruit wines are sweet; therefore, the dessert wine category is often used as a proxy. Although the sweet wine market has been estimated to be just 2 percent of the global wine market (Rivard, Reference Rivard2009), nongrape wines have represented the fastest-growing sector of the U.S. wine industry since demand for fruit wines grew by more than 89 percent in 2014 (Wine America, 2012).Footnote 1
Given the small but growing nature of the fruit wine industry, knowledge of consumer preferences can be critical for properly targeting segments of the population who are more likely to purchase fruit wines at a given price. When selling fruit wines, it is often not sufficient to increase consumer awareness by providing information regarding salient characteristics of the wines that may affect choices. Wines’ sensory characteristics can be key elements in customers’ decisions to make repeat purchases, so studying the combined effect of information and sensory evaluations may provide unique insights regarding the importance of both types of cues in marketing research (Combris, Bazoche, Giraud-Hraud, and Issanchou, Reference Combris, Bazoche, Giraud-Hraud and Issanchou2009).
For wines in particular, sensory evaluation is an important element guiding the consumer- driven wine-making process (Lesschaeve, Reference Lesschaeve2007) and has often been explored in relation to wine market prices (see Goldstein et al., Reference Goldstein, Almenberg, Dreber, Emerson, Herschkowitsch and Katz2008; Granato, Katayama, de Castro, Reference Granato, Katayama and de Castro2011; and Lecocq and Visser, Reference Lecocq and Visser2006, among others). Marketers pay special attention to the determination of pricing strategies by employing a variety of methods to uncover consumers’ willingness to pay (WTP). Experimental economic methods, including auctions (Lusk and Shogren, Reference Lusk and Shogren2007), have often been conducted to elicit consumers’ valuations for wines or wine-specific attributes. For example, Lecocq, Magnac, Pichery, and Visser (Reference Lecocq, Magnac, Pichery and Visser2005) use a Vickrey auction to investigate the impact of information on bids for four red wines (two Bordeaux and two Burgundy wines) and find a higher WTP when participants are provided with information about the wines’ characteristics and expert opinions after first blind-tasting the wines. However, when such information is provided before the blind-tasting, the tasting does not exert a significant effect on WTP. More recently, Vecchio (Reference Vecchio2013) uses a fifth-price Vickrey auction to determine price premiums for sustainable wines (Protected Geographical Indication wines from Sicily), and Bazoche, Deola, and Soler (Reference Bazoche, Deola and Soler2008) use the Becker-DeGroot-Marshak mechanism to study the impact of information about pesticide use in farming on WTP for four Bordeaux wines.
Wineries rely on a variety of marketing tools, including packaging, bottle shapes, designer labels, and advertising (Boudreaux and Palmer, Reference Boudreaux and Palmer2007; de Mello and Pires, Reference de Mello and Pires2009), to convince consumers that their wines are of higher quality and therefore worth a higher price. However, little work has been done on the signaling effect of bottle size formats on the implied quality of wines. Bottle size format is one of the most accessible and easy-to-process cues to which consumers are exposed and has rightly been studied elsewhere, albeit for many goods other than wine. For example, Mathur and Qiu (Reference Mathur and Qiu2012) and Yan, Sengupta, and Wyer (Reference Yan, Sengupta and Wyer2014) provide evidence that consumers associate package size with the quality of the goods contained therein (the products used in these studies are potato chips, shampoo, laundry detergent, orange juice, vitamin pills, and pharmaceutical drugs). Their results suggest that marketers can manipulate consumers’ perception of superior product quality through smaller package sizes, even in the absence of unit price information. In addition, Shreay, Chouinard, and McCluskey (Reference Shreay, Chouinard and McCluskey2016) conclude that different package sizes of the same product may reflect product differentiation, which can explain quantity surcharges.
From an economic perspective, as long as the quality of the good remains unchanged, when examining marginal increases in the quantity of a good, the price a consumer is WTP for the marginal increase should be progressively lower due to diminishing marginal utility. If, on the other hand, (perceived) quality differs because of changes in the quantity of the good, then diminishing marginal utility is not relevant to explain price premium differences. Perception of quality with respect to quantity is summarized in the popular proverb “Good things come in small packages,” which is to say that the size of an item does not always indicate its quality; smaller things often are of a better quality than larger ones.
The notion that prices may convey quality signals is well known in the behavioral marketing literature (Bagwell and Riordan, Reference Bagwell and Riordan1991; Rao and Monroe, Reference Rao and Monroe1989; Shiv, Carmon, and Ariely, Reference Shiv, Carmon and Ariely2005), but the relation of bottle size formats and prices is not well established. In a related paper, Outreville (Reference Outreville2011) studies prices posted on company websites for several champagne producers (e.g., Dom Pérignon, Moët & Chandon, Deutz) in a variety of bottle formats (from demi [375 mL] and jennie [500 mL] to nabuchodonosor [15 L ]). He finds that smaller champagne bottles are sold for a higher normalized price relative to the 750-mL format. Similar results are obtained for Bordeaux and Burgundy wines. On the other hand, Brunke, Thiemann, Mueller, and Albrechts (Reference Brunke, Thiemann, Mueller and Albrechts2009) find no impact of bottle formats smaller than 750 mL on prices of wines offered for sale at auction markets in Germany. A recent contribution to this broad literature regarding consumer-packaged goods comes from Yan et al. (Reference Yan, Sengupta and Wyer2014), who introduce the importance of the effect of package size on a product's unit price rather than on its total price. They investigate how variations in package sizes of food and nonfood items (but not wines) can affect both the unit price and the total price and find that smaller packages are judged to be better , as they are associated with higher unit prices (but lower overall prices).
Thomas (Reference Thomas2000) notes the importance of packaging: it embodies aesthetics and emotions and represents the last chance to communicate with and influence consumers. Wine makers and marketers are aware of the power of packaging on consumers’ purchasing behavior and therefore attach great importance to the product information that appears on the label. Behavioral marketing literature has established that product information may shape hedonic and sensory judgments. For example, Allison and Uhl's (Reference Allison and Uhl1964) influential work concludes that consumers show a preference for their favorite beers in labeled conditions but not in blind tests. Johansson, Haglund, Berglund, Lea, and Risvik (Reference Johansson, Haglund, Berglund, Lea and Risvik1999) reach a similar conclusion by performing a blind and a nonblind preference test for tomatoes in a selected trained panel. Finally, in a wine-tasting experiment, the presentation of positive and negative information about the wine affects not only consumers’ overall assessment of the wine after the tasting but also the experience itself (Siegrist and Cousin, Reference Siegrist and Cousin2009).
To test whether the bottle size format signals quality changes, we use a controlled laboratory experiment in which we simultaneously auction a pomegranate and a grape sweet wine. We vary on a between-subjects basis the size of the bottle, from 500 mL to 750 mL, but keep the content constant across bottle size formats. For both wines, we find evidence consistent with diminishing marginal utility; for the pomegranate wine, our estimates imply a premium for the smaller bottle size, which is consistent with changes in perceived quality due to changes in its perceived scarcity.
In addition to bottle size effects, we also explore in a within-subjects design the effect of prior expectations (visual treatment), blind tasting (sensory treatment), and information (information treatment) on WTP for the two wines. We find that tasting has a negative effect on WTP with respect to expectations from the visual first stage but that information about the wine offsets the negative effect of the taste (sensory) treatment. In the next section, we present our experimental design; we present the statistical and econometric analysis in Section 3; and we offer our conclusions in the final section.
II. Experimental Design
Our experiment combines characteristics of a within- and a between-subjects design (Charness, Gneezy, and Kuhn, Reference Charness, Gneezy and Kuhn2012). In June and July 2015, a marketing research company recruits adults (ages 18 years and older) from the general population in Athens, Greece, to participate in a market research session of approximately one hour at the university campus. Consumers’ various consumption habits are screened; selected participants have no allergies associated with food/drink consumption, drink at least one glass of wine twice per week, and buy a bottle of wine at least once per month. In all, 160 participants complete our experiment over 17 sessions (all but 2 sessions are conducted with 10 subjects per session). In addition, participants are asked not to smoke, eat, or drink anything except water for at least 1.5 hours before their session to avoid any influence on their taste perception from prior cigarette/food/beverage consumption. Sessions are scheduled Monday through Saturday during morning and afternoon hours to accommodate participants’ schedules. The experiment is fully computerized using zTree (Fischbacher, Reference Fischbacher2007).
Upon arrival, participants are given a consent form to sign and randomly seated at private booths. Printed instructions are given to all participants, and the moderator also reads them aloud. Participants are specifically instructed to raise their hands to privately ask any questions; the moderator then shares her answer with the group. Participants each receive a show-up fee of 10 euros and another 10 euros upon successful completion of the experiment. Participants can earn or lose money during the experiment (described below), so average total payouts are 25.7 euros (S.D. = 3.90, min = 8, max = 40). After instructions are read aloud, participants answer a series of computerized control questions to enhance their comprehension of the instructions. They are free to refer to their printed instructions or to ask questions of the moderator and generally show a good understanding of the process, correctly answering an average of 10.5 out of 12 questions.
The experiment consists of three stages (experimental instructions are reproduced in English in online Appendix C). Stage 3 involves two risk-preference elicitation tasks; these data will be analyzed elsewhere and are not further discussed in this paper. In Stage 1, participants complete a typical real-effort task of counting and reporting the number of zeros shown in a 4 × 4 matrix. This task is repeated 10 times (the elements of the matrix are random and change with each repetition but are the same for all participants at a given repetition), with participants earning 0.5 euros every time they correctly solve the task within 30 seconds. The task's purpose is to mitigate house money effects by making participants earn part of their endowments (e.g., Corgnet, Hernán-González, Kujal, and Porter, Reference Corgnet, Hernán-González, Kujal; and Porter2014; Jacquemet, Joule, Luchini, and Shogren, Reference Jacquemet, Joule, Luchini and Shogren2009). The task is purposefully made easy (as evident by the fact that the money earned averages 4.83 euros, with a standard deviation of 0.31 euros, and that 73.8 percent and 20.6 percent of participants earn exactly 5 euros and 4.5 euros, respectively), allowing participants to start Stage 2 of the experiment with approximately equal endowments.
In Stage 2, groups of five participants bid in a series of second-price Vickrey auctions (Vickrey, Reference Vickrey1961). Matching in groups is random and remains the same throughout the session. Participants are unaware of which other participants in the session compose their group. The group size is decided with three things in mind: i) avoid disengaging off-margin bidders from the auction procedure (Shogren, Margolis, Koo, and List, Reference Shogren, Margolis, Koo; and List2001) by forming groups that are “too large”; ii) given that price feedback in repeated second-price auctions is discouraged (Corrigan, Drichoutis, Lusk, Nayga, and Rousu, Reference Corrigan, Drichoutis, Lusk, Nayga and Rousu2012), avoid “too small” groups that will, by design, easily reveal the bidding behavior of other participants; and iii) increase the number of independent observations (if we count the auction group as the unit of an independent observation). In addition, recent theoretical (Banerji and Gupta, Reference Banerji and Gupta2014) and empirical studies (Rosato and Tymula, Reference Rosato and Tymula2016) have shown that if participants have reference-dependent preferences, then the equilibrium bid is lower when the number of bidders is larger. Thus, by keeping the number of bidders constant, we eliminate a possible confound.
The mechanics of the auction are explained in the instructions. To ensure that participants fully understand the procedure, three hypothetical training rounds for two nonfocal food products are conducted: a pack of biscuits and a chocolate bar (pictures B2a and B2b in online Appendix B show the products as presented to the participants). Bids are entered simultaneously for the two goods, closely mimicking the real auction rounds that follow.
Right after the training rounds, participants are served 40 mL of each of the two wines in International Standards Organisation (ISO) wine-tasting glasses (see picture B1 in online Appendix B). Wines are only identified by a three-digit number at this stage. Participants are reminded to taste the wines only when the moderator instructs them to do so. Research assistants in the lab ensure that everyone follows this rule; we did not observe anyone breaking it. The real auction rounds that follow consist of three within-subjects treatments: visual, sensory, and information (see Figure 1). Participants progress from the visual to the sensory and then to the information treatment. Each within-subjects treatment involves three repetitions of a second-price auction for the two wines for a total of nine rounds. Participants are told that only one round and one product will be randomly selected for each auction group at the end of the session using a bingo cage. Just before each three-round auction repetition, participants complete hedonic evaluations for both wines (on a scale from 1 to 9) as well as paired comparisons.Footnote 2
Within-subjects Treatments
In the first within-subjects treatment (visual), participants are asked to observe the two wines and evaluate them based on their expectations of liking each one (the wines’ presentation order is randomized between participants). Participants then bid in three repeated auction rounds their WTP to buy a bottle of each wine. In the second within-subjects treatment (sensory), participants are asked to first taste and then evaluate each wine. The wines’ tasting order is randomized for each participant according to private instructions on their screens, and they are instructed to rinse their mouths with water before tasting the second wine. Participants then bid again in three repeated auction rounds. In the third within-subjects treatment (information), participants are given information about each wine, shown pictures of the wines on their computer screens (see pictures B3a, B3b, B3c, and B3d in online Appendix B), and have the option of examining real bottles of wines in the lab if they wish to do so. The information given in this treatment is taken from the bottle labels, including details about the grape and pomegranate varieties, the fruits’ growth locations, bottling locations, and serving advice.Footnote 3
Furthermore, to examine bottle size effects, we vary on a between-subjects basis the bottle size of the wines at two levels: a standard 750-mL bottle and a 500-mL bottle. Table 1 shows the number of participants in each cell of this 2 × 2 experimental design. They know from the beginning of the auction the size of the bottle for each wine they are bidding on. This information is also provided during the bidding stage via descriptive text on their computer screens. Both wines are bottled and labeled on demand by their producers for the purposes of this experiment.
Between-subjects Treatments
III. Results
The essence of experimentation is the ability to establish exogeneity of the treatment by comparing groups of participants who have the same expectation regarding the distribution of all covariates that are potentially relevant for the treatment outcome. Typically, before presenting any results, experimenters perform balance tests to scrutinize for randomization success. With our data, we fail to reject the null of no difference for the between-subjects treatments at α = 5% significance level for age (Kruskal-Wallis χ 2 = 0.071, p-value = 0.995), body mass index constructed from self-reported weight and height (Kruskal-Wallis χ 2 = 0.096, p-value = 0.992), gender (Pearson's χ 2 = 1.117, p-value = 0.773), education level (Fisher's exact p-value = 0.614), household size (Pearson's χ 2 = 10.810, p-value = 0.545), smoking status (Fisher's exact p-value = 0.718), consumption frequency of alcoholic drinks in general (Pearson's χ 2 = 2.438, p-value = 0.883) and of wine in particular (Pearson's χ 2 = 3.174, p-value = 0.366), number of alcoholic drinks consumed in a typical day (Fisher's exact p-value = 0.883), and income levels (Fisher's exact p-value = 0.069).Footnote 4
A. Hedonic Evaluation of Wines
We first start by analyzing the hedonic score data for wine evaluations. Figure 2 shows histograms of the evaluations for each wine, pooled over all treatments. It is obvious that the grape wine is evaluated better than the pomegranate wine; although the figure pools the data over all treatments, a similar picture emerges if one examines score differences on a treatment-by-treatment basis (see figures A1, A2, and A3 in online Appendix A). We can use a sign test to calculate the differences between the hedonic scores for the two wines and test the null hypothesis that the median of the differences is zero. The null is highly rejected (p-value <0.001) over the one-sided alternative that the median of the difference: hedonic score grape wine − hedonic score pomegranate wine >0. The null is highly rejected as well (p-value = 0.004) if we perform the test separately for each within-subjects treatment.
Hedonic Evaluation Scores by Wine
We can further examine the factors that affect hedonic score evaluations by means of random effects-ordered logit regressions, as shown in Table 2 (Table A1 in online Appendix A shows results with additional controls added in the model specification). Table 2 depicts results separately for each type of wine as well as a pooled model. The pooled model interacts the treatment dummies with the wine dummy (grape wine) to examine any differential treatment effects across products.
Random Effects-ordered Logit Models of Hedonic Evaluations
Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. Base categories are: (Pomegranate wine: 750 mL, Grape wine: 750 mL), Visual treatment.
We can see in Table 2 that the information treatment boosts hedonic scores for both wines with respect to the visual treatment. In addition, tasting improves hedonic evaluations of the grape wine when compared with evaluations based on expectations only. However, tasting does little in that regard for the pomegranate wine. Second, the pooled model shows a positive and statistically significant effect for the grape wine dummy, which indicates an overall preference for the grape wine. The interaction terms for the within-subjects treatments indicate that the sensory and information treatment effects are greater for the grape wine than for the pomegranate wine.
Paired comparison data are depicted in Figure 3. The question we seek to answer is whether the two types of wine are perceived as dissimilar at each of the within-subjects treatments. Besides the visual dissimilarity, which is more pronounced in the visual treatment (i.e., subjects are shown to prefer the grape wine more than the pomegranate wine), we can test whether the observed percentages are significantly different from expected percentages using a χ 2 goodness of fit test. For the visual treatment, when we test whether the observed percentages (49.38 percent prefer the grape wine, and 30 percent prefer the pomegranate wine) do not reflect an equal split between the two wines (i.e., 50 percent prefer the grape wine, and 50 percent prefer the pomegranate wine), we reject the null (χ 2 = 12.13, p-value = 0.007). Although the sensory treatment shows a shift toward the pomegranate wine, we fail to reject the null (χ 2 = 5.52, p-value = 0.137) of an equal split between the two wines. The shift toward preferring the pomegranate wine is further reinforced in the information treatment, and the observed difference in favor of the pomegranate wine is statistically different than an equal split of subjects between the two wines (χ 2 = 6.25, p-value = 0.044). Furthermore, a Pearson χ 2 test of whether choice in the paired comparison is not related to the within-subjects treatments rejects the null (χ 2 = 19.48, p-value = 0.003), which indicates that the within-subjects treatments exert a statistically significant effect on paired comparisons.
Paired Comparisons of Wines by Within-subjects Treatment
In a conditional analysis context, the factors that affect choice in the paired comparison test can be examined by means of a multinomial logit model. Results are shown in Table 3 (Table A2 in online Appendix A shows results with additional controls added in the model specification). Results confirm the information shown in Figure 3; the sensory and information treatments exert a statistically significant effect, and the negative sign indicates it is less likely that someone will choose the grape wine over the pomegranate wine as compared to the visual treatment. That is, the negative coefficients for the within-subjects treatments highlight the shift toward the pomegranate wine, as illustrated in Figure 3. In addition, when we test for a difference between the sensory and the information treatment coefficients, we fail to reject the null for the outcomes “Chooses the grape wine” and “Likes both wines equally well.”
Multinomial Logit Model for Pairwise Comparisons
Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. “Chooses pomegranate wine” is the base outcome. Base categories are: (Pom: 750 mL, Grape: 750 mL), Visual treatment.
B. Descriptive Analysis of Bids
Table 4 shows descriptive statistics of bids per treatment for each type of wine. For the pomegranate wine, we make two observations. First, across all between-subjects treatments, bids decrease in the sensory treatment (after tasting), as compared to the visual treatment, but they increase above the level of bids in the visual treatment when information about the wine is provided. Second, the pomegranate wine is valued more in a 500-mL bottle format, competitive with a 750-mL bottle of grape wine. For the grape wine, the within-subjects treatments show a linear relationship of bids; that is, bids increase in the sensory treatment and further increase in the information treatment. With respect to bottle size effects, the grape wine is valued more when it is in a 750-mL bottle, competitive with a 500-mL bottle of pomegranate wine. Table 4 is complemented with appropriate statistical tests for treatment effects (i.e., the Kruskal-Wallis test (Kruskal and Wallis, Reference Kruskal and Wallis1952), the K-sample median test (Mood, Reference Mood1954), and the Friedman test (Friedman, Reference Friedman1937, Reference Friedman1939)) that reveal statistically significant treatment effects. We quantify these effects in a regression context momentarily.
Descriptive Statistics of Bids per Wine Product and Treatment
The Kruskal-Wallis test evaluates the equality of the between-subjects treatment separately for each of the within-subjects treatments and wine type (column-wise). Similarly, the K-sample test performs a nonparametric evaluation of the equality of medians. The Friedman test compares the within-subjects treatment (visual, sensory, information) for each of the between-subjects treatments and wine types (row-wise).
The bottle size effects are graphically depicted in Figure 4, which shows average bids over rounds and over the within-subjects treatments per bottle size and wine type. It is obvious that there is an inverse relation between bottle size and valuations for pomegranate wine, which is more pronounced in the information treatment. For the grape wine, there is a positive relationship between bottle size and valuations, which remains stable over the within-subjects treatments.
Average Bids per Round and Treatment
C. Econometric Analysis
To check whether the results obtained hold in the context of conditional analysis and to quantify treatment effects, we estimate random effects-regression models where the grouping structure of the data consists of three levels of nested groups (i.e., three random effects): the auction group, j, the individual, i, and the auction round, t. The model specification we estimate is as follows:
$$Bid_{jit}\; = x_{jit}b + \; u_{j} + \; v_{ji} + \varepsilon_{jit}$$
where j = 1 … J indexes the auction groups, i = i … N indexes individuals in an auction group, t = 1 … T indexes auction rounds (in our case J = 32, N = 5 and T = 9), and x is a vector of independent variables. The random effects u
j
, v
ji
, and ε
jit
are i.i.d.
$N(0,\; \sigma _u^2 )$
,
$N(0,\; \sigma _v^2 )$
and
$N(0,\; \sigma _\varepsilon ^2 )$
, respectively and independently of each other.
In addition, about 5.5 percent of all bids for the pomegranate wine is exactly zero. This result calls for the use of a censored regression model for the pomegranate wine to address possible censoring from the left (Tobit model). The Tobit model slightly complicates the analysis, because the researcher could be interested in four marginal effects: i) marginal effects on the latent variable,
$\displaystyle{{\partial E{\rm [}Bid{^\ast} {\rm \vert}x]} \over {\partial x}}$
(these are the raw coefficient estimates); ii) on the observed variable,
$\displaystyle{{\partial E[Bid \vert x]} \over {\partial x}}$
; iii) on positive bids,
$\displaystyle{{\partial E[Bid \vert Bid \gt 0,\; x]} \over {\partial x}}$
; and iv) on the probability of being uncensored,
$\displaystyle{{\partial Pr[Bid \gt 0 \vert x]} \over {\partial x}}$
. For the grape wine, because zero bids are only a small fraction of total observations (0.7 percent), we estimate a random effects linear regression model.
Results are exhibited in Table 5 (Table A3 in online Appendix A shows results with additional demographic and attitudinal variables added in the model specification). The last column shows results from the linear regression model, while all the other columns show marginal effects from the Tobit model identified by appropriate headings.
Random Effects Tobit Model (Pomegranate Wine) and Random Effects Linear Regression Model (Grape Wine)
Standard errors in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01
We first start with the results from the within-subjects treatments. For the grape wine, bids increase statistically significantly after tasting (sensory treatment) and increase again after information is revealed (information treatment). A Wald test of whether the coefficients for the sensory and information treatments are statistically different fails to reject the null (χ 2 = 1.51, p-value = 0.218), which indicates that the two treatments have a similar effect on bids as compared to the visual (baseline) treatment. In terms of magnitude, the effects are rather small: 0.22 euros and 0.35 euros for the sensory and information treatments, respectively.
For the pomegranate wine, the effect in the sensory treatment is statistically significant and negative but rather small in magnitude. For example, for those who bid positively, tasting the pomegranate wine reduces their bids by 0.21 euros. In addition, the likelihood of a participant's bidding a positive amount is reduced after tasting by 1.4 percent. The information treatment does not exert a statistically or economically significant effect with respect to the visual treatment. This result illustrates a U-shaped effect of the within-subjects treatments on bidding behavior, wherein bids first decrease after tasting (sensory treatment) but then resume the level of the visual treatment after information is revealed in the information treatment.
For the between-subjects treatments (designed to examine bottle size effects), we find a negative effect for the grape wine for the smaller bottle size format. For example, when compared to a treatment wherein grape wine is presented in a 750-mL bottle, the 500-mL bottle is statistically significant at the 10 percent level for just one of the treatments but when we control for additional factors (see Table A3 in online Appendix A) both treatment effects are statistically significant. In any case, the size of the effects is substantial. In terms of average predicted WTP, we find that a 500-mL bottle of grape wine is valued at 4.52 euros (pooled over treatments 1 and 4; see also Table 6), while a 750-mL bottle is valued at 6.49 euros (pooled over treatments 2 and 3), which implies a premium of 43.6 percent for an increase in volume of 50 percent (see also Table 6 for average WTP predictions per treatment).
Average Predicted WTP per Bottle Size and per 100 mL
Note: PW stands for pomegranate wine; GP stands for grape wine. Numbers in bold highlight WTP for the 500 mL bottle size. Predictions are based on the marginal effects of the observed bids i.e., second column of results in Table 5.
The bottle size effects are reversed for the pomegranate wine. Table 5 shows that with respect to the treatment wherein both wines are presented in 750-mL bottles, no other treatment differs statistically significantly. This result implies, however, that a 500-mL bottle is valued at least as much as a 750-mL bottle, which, given the volume difference, implies a premium for the bottle size of 500 mL on a per-liter basis. In terms of average predicted valuations, conditional on bidding positively, a 500-mL bottle of pomegranate wine is valued at 5.59 euros, whereas a 750-mL bottle is valued at 5.25 euros, which implies a 6.08 percent discount for an increase of 50 percent in volume. In other words, the producer is much better off selling 500-mL bottles of pomegranate wine.
These effects are better illustrated in Table 6, which shows the average predicted WTP per bottle size and the implied WTP per 100 mL of wine. Consistent with the discussion above, a larger bottle implies a lower WTP per 100 mL for both wines. However, the difference in terms of WTP per 100 mL between bottle sizes is much larger for the pomegranate wine than for the grape wine; a 500-mL bottle of pomegranate wine is valued equally to or more than a 750-mL bottle of pomegranate wine (compare, for example, predicted WTP for 500 mL versus 750 mL of pomegranate wine: 4.32 euros versus 4.39 euros and 5.95 euros versus 5.02 euros). This result can be explained in terms of perceived quality and diminishing marginal utility. Diminishing marginal utility causes a negative effect on WTP per 100 mL for quantity increases for both wines. The negative effect is reinforced for pomegranate wine due to a perceived quality effect; that is, perceived quality increases WTP for the 500-mL bottle as compared to the 750-mL bottle. Thus, both effects reinforce each other in widening the gap in valuations between bottle sizes for pomegranate wine on a per–100 mL basis.
The effect sizes in Table 5 show that the premium to be gained by a smaller bottle size of pomegranate wine is higher if the pomegranate wine is paired with a bigger bottle of grape wine. This is likely due to a comparison effect that makes the smaller bottle size look even better for the pomegranate wine when it is compared to a larger bottle of grape wine.
IV. Conclusions
The objective of this paper is twofold: i) to investigate the impact of taste and information on WTP for two wines, and ii) to test whether bottle size formats for wines signal quality changes to consumers. Our findings advance knowledge on two separate fronts in the behavioral marketing literature and provide further insights relevant for wine marketing as well as for the broader market of consumer-packaged goods.
First, we add to the behavioral marketing literature by investigating whether packaging elements influence consumers’ decision making by signaling price and quality changes. Results from the second price auction indicate that participants in our experiment are willing to pay more for an increased quantity of grape wine at a rate consistent with diminishing marginal utility. On the other hand, valuations for the pomegranate wine are driven mostly by perceived quality changes due to a smaller bottle format. When perceived quality changes for the smaller bottle size are combined with diminishing marginal utility for the larger bottle size, the gap widens between the large and small bottle sizes in terms of valuations per 100 mL (see particularly Table 6), so that a 500-mL bottle of pomegranate wine is valued at least as much as a 750-mL bottle. Managerially, our study establishes the relationship between wine bottle sizes and WTP, and it offers an alternative approach for marketers to influence consumers’ perceptions of product quality. However, our results are not relevant to explanations of valuations for quantities larger than 750 mL, such as magnum (1.5 L), imperial (6 L), or other large bottle sizes, for which consumers have good reason to expect quality increases along with quantity increases (see Outreville, Reference Outreville2011).
Second, this paper contributes to prior research about the role of information in consumer behavior by examining whether prior expectations, tasting, and product information shape hedonic and sensory judgments and alter consumers’ WTP for wines. We find that tasting has a negative effect on WTP with respect to expectations from the visual treatment (prior expectations) but that when information about wines is provided, WTP is brought back to the level of the visual treatment. This finding should be of particular interest to the wine industry, as it highlights the importance of information provided on the wine label in determining consumers’ WTP. More specifically, results indicate that even if wine drinkers are not fond of a wine's taste, the presence of detailed information on the bottle for that specific wine could strongly affect positively their WTP. Therefore, label information is an important element in the toolkit that marketers can use to influence consumers’ perceptions of a product's quality and their resulting consumption choices.
Having presented these conclusions, we want to emphasize that the roles of perceived product quality, expectations, and tasting should be evaluated on a wine-by-wine basis; it would be useful to examine the generalizability of our findings across a wider range of wine types, levels of wine sweetness, and different bottle size formats.
Supplementary Material (Appendices)
For supplementary material accompanying this paper visit https://doi.org/10.1017/jwe.2017.3.
