Relevant hypotheses

Based on the existing research reviewed in the background section, we formulate some relevant hypotheses and present them below.

Information provision often affects consumer uncertainty and ambiguity and both shift and rotate consumer demand curves (Johnson and Myatt Reference Johnson and Myatt2006; Liaukonyte et al. Reference Liaukonyte, Streletskaya and Kaiser2015); at the same time, the effect of sensory information provision after tasting is unclear as consumers should have full private information on the sensory match. We can test the impact of third-party information by comparing consumer bids and their distribution in the control and treatment groups.

H1. Provision of peer and expert ratings changes the distribution of consumer bids for the same product even after tasting.

While the above hypothesis focuses on whether information provision can affect aggregate consumer demand for wine even after tasting, from a managerial standpoint, it is important to see the impact of owns rating on consumer WTP, alongside peer and expert information.

H2. Consumers exhibit higher WTP for wine they rated highly after tasting or selected for tasting.

H3. Positive expert and peer information increases consumer WTP for wine, while negative peer and expert information reduces consumer WTP for wine.

Previous research suggests that consumers are prone to confirmation bias and tend to value expert opinions on wine more than peer.

H4. Impact of peer and expert information on consumer demand differs.

H5. Peer and expert information congruent with consumer own evaluation of wine have a differential impact on consumer WTP.

We use two experimental studies with wine tasting and demand elicitation via an incentive-compatible mechanism to examine the hypothesis outlined above. Both studies were IRB-approved as exempt and took place in an experimental economics lab at a large Pacific Northwest University. Recruitment was conducted through email solicitations to the existing pool of the experimental economics lab with details about the study. All participants were screened on having had previous wine consumption and winery visit experience, age over 21 years (to comply with state regulations), and pregnancy status (IRB excluded pregnant participants due to the potentially increased risk associated with alcohol consumption). On average, between 12 and 14 years of wine drinking experience was reported. Over half of our sample in each study reported drinking wine at least once a week, while an additional 30–37% of the sample drank wine more occasionally, on special occasions. Participation in each study was associated with a $35 endowment, that participants could keep in full if no wine was purchased. We discuss the design, sample, and estimation approaches of each study in detail in the next sections.

Methods and experimental design

This study had two treatment groups (1: providing the average peer rating for each wine after the tasting, and 2: providing expert ratings sourced from the Wine Spectator) and one control group, where participants were only reminded of their own rating for the wine during the WTP elicitation stage.

Each session in the experiment consisted of a wine tasting stage and a computerized WTP elicitation stage for the all wines in the experiment. Tasting stage setup was similar in all groups, and only the additional information provided during the auction on wine ratings differed across the treatment and control groups.

During the tasting stage, all participants tasted and rated the same four wines sequentially, with the order of wines randomized for each tasting session. Each session included up to nine participants at a time. To rate the wines, participants used a five-star system common in many consumer rating applications, where each star level was described using a corresponding Wine Spectator scale. The Wine Spectator’s 100-point scale is divided into six categories: 100 to 95 points range represents a “classic: a great wine,” 94 to 90 points are awarded to wine that is “outstanding: a wine of superior character and style,” 89 to 85 point signify a wine that is “very good: a wine with special qualities,” 84 to 80 describe a wine that is “good: a solid, well-made wine,” while 79 to 75 points are assigned to “mediocre [wine]: a drinkable wine that may have minor flaws,” while anything below 74 points is “not recommended” by the Wine Spectator. We use these category descriptions for our five-star rating, with one star described as the mediocre wine category and five stars described as the top category wine. Figure 1(A) presents the rating description seen by participants for each wine.

Figure 1. Computer interface for study 1.

The bottles used in the first experiment were masked, with labels and identifying brand features covered universally and similarly across all bottles, to limit brand and packaging design impacts on tasting and demand elicitation. This masking also prevented respondents from seeing information on vintage, region, and any descriptions of the wine usually present on the wine bottle. All four wines were Pinot Noirs from different wineries and vintages, chosen to represent a broad range of Wine Spectator expert rankings. Finding and sourcing even mediocre wine is quite difficult, as only 49 wines among the 2016–2021 vintages were rated at or below 79 points, out of total of 43,860 wines rated for the same period (Wine Spectator 2021). This suggests that out of all reviewed wines, only under around 0.11% were rated as mediocre or not recommended. Only 3 wines out of these 49 were rated below 75 (not recommended). As a comparison point, 1,545 wines, or around 3.52%, were rated as Classic, the highest category, in the same time period. In essence, most wines reviewed in the Wine Spectator are rated 80 points and above. For our experiment, we were able to source wines rated between two and five stars, one wine per each of these star categories.

After wine tasting and rating were completed for all four wines, we used computerized Becker-deGroot-Marschak (BDM) (Becker et al. Reference Becker, DeGroot and Marschak1964) mechanism auctions for demand elicitation. The BDM was selected as the elicitation method because subjects do not bid against each other but rather submit a sealed bid for wine. Once all bids are submitted in a session, we randomly chose a market price for one randomly selected wine bottle (from a distribution around the retail price of the wine). If a participant’s bid is equal to or exceeded the randomly drawn price, they receive the wine bottle at the randomly drawn price, and not at the value of their bid; otherwise, no transaction results from the auction. This method allows us to reliably incentivize participants to bid their true maximum WTP for all auctioned items.

At the start of the auction stage, participants were informed that all wines in the experiment retailed for between $15 and $35 to bound their price expectations, in line with other work on wine WTP elicitation (Streletskaya et al. Reference Streletskaya, Liaukonyte and Kaiser2019). The information provided during the auction stage differed based on control or treatment assignment, with the computer screens provided in Fig. 1B–D. In the control group, the bidding screen reminded participants of their own rating for the wine submitted during the tasting stage, as shown in Fig. 1 B. In the peer treatment group, participants also saw the average rating, down to half-star approximation, of other people in the room, as well as their own rating, as shown in Fig. 1C. In the expert treatment group, participants saw the Wine Spectator rating for the wine, in addition to their own wine rating, Fig. 1D.

Once all participants submitted their bids to all four wines, we revealed the randomly drawn binding auction and the randomly drawn clearing price for that auction. Participants who submitted a bid higher than or equal to the clearing price receive the wine bottle for the binding auction at the clearing price and keep the rest of their participation payment. Participants with bids lower than the clearing price keep $35 of their participation payment and do not receive a wine bottle. This approach to market clearing with multiple product auctions is generally considered incentive-compatible under monotonicity assumptions (Azrieli et al. Reference Azrieli, Chambers and Healy2018). All participants took part in a practice round of bidding for a one-dollar bill to illustrate the incentives to reveal one’s maximum WTP.

An extensive sociodemographic survey was administered to all participants after the auction bids were submitted. At the end of the experiment, the market was cleared and the subjects were given their compensation.

Sample and estimation approach

One-hundred seventy-three wine drinkers over 21 years old participated in this experiment. Detailed summary statistics for the overall sample, and treatment and control groups, are presented in Table 1. Participants in the overall sample varied in age between 21 and 65 years old, with average age of around 33 years old for the full sample. Some difference is present for participant age in the control group compared to both treatment groups, and the average participant age in both treatments is not significantly different from each other. While overall the descriptive statistics look similar between all groups, the nature of random assignment with a finite sample yields some variation across the groups.

Table 1. Descriptive statistics: Means and (Standard Deviations) of demographic variables by treatment

Pairwise mean comparison of bid values and own ratings with Bonferroni correction across participant groups suggests no difference at the 5% significance level. On average, the study participants had significant experience with wine, drinking wine for over 10 years, but that was not reflected in the scores of the objective wine knowledge test. On average, respondents scored below 50%, which is in line with general consumer wine knowledge (Aqueveque Reference Aqueveque2015). Our participants regularly reported trying new wines, which is common for a highly differentiated product market. Finally, it is interesting to note that on average, own ratings were significantly lower than the expert ratings for all wines, with consumers seemingly harsher than experts in wine quality evaluation.

We use a series of standard Tobit model specifications, with clustered standard errors by participant, to estimate consumer WTP for wine in the presence of treatment, own, peer, and expert ratings of wine, depending on specification:

$$\left\{ {\matrix{ {WTP_{ijt}^* = {\alpha _{it}} + {\beta _{it}}OwnRatin{g_{ij}} + {\beta _{expert}}ExpertRatin{g_j} + {\beta _{peer}}PeerRating + } \cr { + {\beta _t}Treatment + \sum\limits_i {{\gamma _l}} {X_{li}} + {\varepsilon _{ijt}}{\rm{ }}} \cr {WT{P_{ijt}} = \max \left( {WTP_{ijt}^*,\;0} \right)} \cr } } \right.$$

Here, subscript i refers to a participant, j to particular wine, and t refers to experiment assignment group. The constant is denoted as ${\alpha _{it}}$ , PeerRating and ExpertRating provide the average in wine peer rating or the expert wine rating, respectively, and are equal to 0 when no peer or expert information was available to participants. $OwnRating$ includes participant i’s own rating for each wine j. $Treatment$ refers to the treatment (1 and 2) and control (0) groups. ${\gamma _l}$ represents the marginal effect of sociodemographic or bottle attribute l on WTP, while X_li is the level of attribute l for individual i, and finally, the error term is ${\varepsilon _{ijt}}\sim N\left( {0,1} \right).\;$

Postestimation tests and the significance of parameter estimates, in some cases, are used to test the hypotheses of interest. In addition, we use a set of Kolmogorov–Smirnov (KS) distribution tests to evaluate the equality of consumer bid distributions for the control and treatments group.

Estimation results

H1. Provision of peer and expert ratings changes the distribution of consumer bids for the same product even after tasting.

We use our full sample to examine H1. The results of KS tests suggest that the distributions of consumer WTP for wine are not statistically different between the control group and participants receiving peer or expert rating information (Combine KS D: 0.0299, p-value 0.999). No differences in the WTP distribution between control and treatment groups hold for each individual wine bottle evaluated (bottle 1: Combined KS D 0.0853, p-value 0.950; bottle 2: Combined KS D 0.1184, p-value 0.675; bottle 3: Combined KS D 0.0563, p-value 1.000; bottle 4: Combined KS D 0.0979, p-value 0.869). Our Tobit regression results, with own rating (specification 2), and with demographic controls for gender, age, education, and race (specification 3), and without (specification 1), are similar and suggest no difference in consumer WTP for wine in presence or absence of peer and expert valuation information. Overall, we use data from all 173 participants in all of our specifications, which keeps the number of clusters the same across specifications, as all consumers have data on all controls used in the specification. The detailed estimation results are presented in Table 2.

Table 2. H1, Tobit estimation results, impacts of expert and peer information on consumer WTP

Standard errors are in parentheses.

*p < 0.1, **p < 0.05, ***p < 0.01.

H2. Consumers exhibit higher WTP for wine they rated highly after tasting or select for tasting.

Table 2 presents a highly significant and positive estimate for the impact of consumer own rating on WTP for wine. Specifically, a one-star increase in own rating is associated with a $2.86 increase in consumer WTP for the wine, when controlling for respondent demographic characteristics.

We continue to analyze our data by limiting the data in consideration to respondents who receive either peer or expert valuation to focus on the impact of peer and expert information. This leaves 119 participants in our sample. The results from a set of Tobit estimations for this sample are provided in Table 3.

Table 3. Tobit estimation results, expert and peer impacts on consumer WTP

Standard errors are in parentheses.

*p < 0.1, **p < 0.05, ***p < 0.01.

Again, consistently, own rating of wine is estimated to play a significant role in consumer WTP formation. An additional star in own rating is associated with between $2.74 and $3.33 increase in consumer WTP for the bottle.

H3. Positive expert and peer information increases consumer WTP for wine, while negative peer and expert information reduces consumer WTP for wine.

Controlling for own rating for the wine, we estimate that positive peer and expert ratings significantly increase consumer WTP for the wine (specifications 1–3, model 2). A one-star higher peer evaluation for the wine increases consumer WTP by between $1.79. and $1.96, when controlling for own wine rating. A one-star higher expert rating is similarly associated with a $1.15 to $1.16 increase in consumer WTP for the wine.

H4. Impact of peer and expert information on consumer demand differs.

The difference between the impacts of peer and expert rating information on consumer WTP is not significant at the 5% level. The p-values of the F-test for the equality of coefficients are 0.157, 0.0717, and 0.0677 for specifications 1–3 (Table 3), respectively. We conclude that both peer and expert valuation impacts consumer WTP for wine.

Another way to look at the impact of outside information in presence of tasting is to focus on the impact of the difference between peer or expert rating with own rating of the participant. We take this approach in specifications 4–5 (Table 3). The results are similar and consistent with results from the base specifications 1–3 and reflect the fact that while higher peer and expert evaluations increase consumer WTP, when own rating for the wine was lower compared to expert or peer evaluation, consumers exhibit lower WTP for wine. The important takeaway here is that for both peer and expert information, F-tests (p-values 0.0, 0.0) suggest that the overall impact of positive peer and expert evaluation overweighs the impact of the disagreement between own and external ratings, leading to, on average, higher WTP. This is in line with results from specifications 1–3. Similar to alternative specifications, we do not observe any difference between peer and expert effects (F-test p-value 0.7335). Specifications 3 and 5 both include a dummy variable control to mark the wines consumers rated at two stars or below, to control for potential strong dislike. The results are consistent between all specifications.

H5. Peer and expert information congruent with consumer own evaluation of wine have a differential impact on consumer WTP.

To investigate the impact of the scale of difference between peer and expert rating and own rating, we estimate a final set of Tobit models that include squared difference between the own and external ratings, in addition to the difference in ratings non-squared. As peer ratings are on average lower than expert ratings, this affects the scale of the square difference measure, and we separately estimate the model for the peer and expert treatment groups. The results are presented in Table 4. A significant coefficient on the squared difference suggests that as the difference increases, its impact increases or decreases. In other words, as external and own ratings become less congruent, their impact changes. We do not find strong support for this hypothesis. The results are consistent between different specifications, with no strong support for the difference in effect between incongruent ratings and similar ratings.

Table 4. The impact of rating congruency between own and external ratings

Standard errors are in parentheses.

*p < 0.1, **p < 0.05, ***p < 0.01.

Methods and experimental design

The main difference between study 1 and study 2 lies in the ability of participants to select some wine for tasting and forgo others in study 2. Additionally, we do not provide explicit rating information and rely on participants’ own observations of peer tasting popularity. The aim of this design change is to modify the experimental design closer to common setup of tasting experiences when explicit peer and expert ratings are not available, and when consumers choose the wines to taste. Specifically, while explicit ratings from peers and experts are available for some wines, especially when the rating is high and used as a promotional strategy, and consumers are able to look them up for particular wines online, in most tasting rooms consumers are more likely to rely on other popularity cues, such as the number of other visitors examining or tasting a particular wine (Streletskaya Reference Streletskaya2022). Six wines of six different varieties were presented for tasting and bidding in each experimental session. The experiment started with participants arriving in the lab, reading and signing experimental instructions and the consent forms, and being endowed with $35.

Participants observed a list of six wines available for tasting and were instructed to select the wine they would like to taste by marking them on their screens. Each wine cost them $1 to taste, to mimic the tasting fee common in wineries. Compared to study 1, we do not mask the bottle and labels of the wines used in this experiment, further moving the environment closer to that of an actual tasting room. However, the selection of the wines remains the same and fixed across the treatment groups, allowing for experimental identification of treatment effect. Participants could not discuss or observe tasting choices of other participants before submitting their own choices. Once all participants submitted their choices, the experimenter sequentially went through all the six wines in the tasting session, pouring one wine at a time to all participants who selected the wine. Thus, during the actual tasting participants observed the general popularity of each wine among their peers, with up to six participants in each experimental session. The tasting itself was organized in the manner described in study 1.

Once the tasting concluded, consumers went through a series of computerized BDM mechanism auctions for six wines, with the procedure outlined in detail in study 1. While participants differed on the amount of sensory descriptions and information available about the wines, in this experiment we focus solely on the impact of observing wine popularity on consumer WTP on average.

An extensive sociodemographic and wine consumption survey was administered to all participants after their completion of the auctions, and the subjects were given their compensation and wine, according to the auction clearing results, at the end of the experiment.

Data were analyzed using the Tobit estimation approach detailed in study 1.

Estimation results

On-hundred eighty-nine participants who were over 21 years of age and had previous wine tasting experience took part in the study. In general, the demographic characteristics of the sample (see Table 5) are comparable to those of study 1 sample. Participants, on average, had almost 14 years of experience drinking wine and tried new wines consistently.

Table 5. Descriptive statistics of the sample, study 2

Participants tasted approximately half of the wines available for tasting (48%), and the average WTP for these wines did not differ significantly from study 1, averaging at $10.12. On average, each experimental session included five participants, implying each participant on average observed the choices of approximately four peers. In line with the general tasting level, most participants observed on average half of their peers tasting each wine, with sizable standard deviation. Two sessions during experiment only had one participant. We drop these sessions in the analysis, as the single participant by definition did not observe any peer social cues.

The following general hypotheses, formulated earlier in the paper, with originarling preserved, are examined in study 2.

H2. Consumers exhibit higher WTP for wine they rated highly after tasting or selected to taste.

In study 2, we do not directly collect information on wine ratings, as not all wines are tasted and no rating information is provided to participants or revealed by the participant to the experimenter. However, we allow consumers to opt into a costly voluntary tasting. The WTP for tasting indicates a relatively higher preference for the chosen wine, and it should be correlated with higher WTP. Note that this is an endogenous relationship, compared to study 1, where participants were assigned to taste all wines. The Tobit estimation results for different specifications are provided in Table 6. The results are consistent and on average selecting a wine for tasting is associated with an approximately $4 higher consumer WTP. Specifications 1–4 differ on the number of controls included; as some of the gender and income data are not reported by five respondents, their bids were omitted from the estimation.

Table 6. Tobit estimation results, tasting choice, and WTP for wine

Standard errors are in parentheses.

*p < 0.1, **p < 0.05, ***p < 0.01.

H3. Positive expert and peer information increases consumer WTP for wine, while negative peer and expert information reduces consumer WTP for wine.

Study 2 allows us to evaluate the impact of latent peer preference cues on consumer own WTP valuation for the wines available in the study. Compared to study 1, we are able to see the impact of peer information for both tasted wines and wines that have not been tasted. Previous research suggests that in absence of direct experience with a product, consumers might rely more on external cues (Chung and Saini Reference Chung and Saini2022), which would increase the importance of peer influences on wines that have not been tasted. On the other hand, consumers prefer the wines that they select for tasting and might not care about peer preferences for options they do not personally consider attractive. We test the potential for such behavior with a series of Tobit model estimations, with results presented in Table 7. We use a continuous interaction of the level of tasting by peers observed for each wine in a session by the dummy variable indicating whether that wine has been tasted by the participant or not. As the decision of others to taste a wine is independent from your own decision to taste it by design, it removes the endogeneity problem and allows for causal identification of the impact of peer popularity on consumer WTP.

Table 7. Tobit estimation results, impact of tasting popularity, measured as percentage of peers tasting the wine, on consumer WTP

Standard errors are in parentheses.

*p < 0.1, **p < 0.05, ***p < 0.01.

We use two approaches to measure peer popularity. One measures the proportion (in percentage terms) of people in each experimental session, not including the participant themselves, who have tasted a given wine. Alternatively, we use the total number of people in the session, not including the respondent themselves, tasting that wine. The second measure ignores differences in the number of people present in the experimental session. Both measures vary for each person for each of the six bottles used in the experiment. Results are presented in Tables 7 and 8.

Table 8. Tobit estimation results, impact of tasting popularity, measured as number of peers tasting the wine, on consumer WTP

Standard errors are in parentheses.

*p < 0.1, **p < 0.05, ***p < 0.01.

The estimation results are consistent between the two approaches to measuring peer popularity and between specifications with various levels of controls. The difference in scale of the coefficient has to do with the differences in the interpretation of Tobit coefficients. In Table 8, seeing one more person tasting a wine bottle leads to a 59-cent reduction in WTP for wine consumers who did not taste and an almost $1 increase in WTP for wine consumers who tasted themselves. In Table 7, an increase in the observed wine popularity of 1%, so having 1% more people tasting it leads to an approximately 2-cent reduction in consumer WTP for not tasted wine and an approximately 4-cent increase for tasted wine. As most of our tasting sessions only provided four peer-choice observations for each wine, a change in one person can be estimated as a change of around 25%. With that adjustment results in Table 7 are comparable with Table 8, with 25% change in popularity during tasting leading to a 30-cent and a 1-dollar change for non-tasted and tasted wines, respectively.

While the results for wines that have not been tasted are less significant (most of them only significant at the 10% level), the asymmetric nature of the peer popularity for preferred and less preferred wines, as measured by consumer decision to taste, is very interesting. This finding provides indirect support to our hypothesis on the impacts of congruency between external and internal evaluations.

H5. Peer and expert information congruent with consumer own evaluation of wine have a differential impact on consumer WTP.

For ease of exposition, we provide the estimates of the impact of wine popularity for tasted and non-tasted wines in Fig. 2, for both popularity measures. Overall, the pattern is very consistent, with peer popularity increasing consumer WTP for wines they have tasted, but somewhat reducing it for wines they did not pick for tasting. In other words, when peer popularity is congruent with their own preferences for wine, consumers increase their WTP; however, incongruent popularity leads to a decrease in WTP. Moreover, the impact of congruent peer rating has a larger impact than the comparative incongruent peer rating.

Figure 2. Coefficient estimate plot, Tobit estimation, wine popularity, with confidence intervals.