ArticlePDF Available

Do More Expensive Wines Taste Better? Evidence from a Large Sample of Blind Tastings

Authors:

Abstract

Individuals who are unaware of the price do not derive more enjoyment from more expensive wine. In a sample of more than 6,000 blind tastings, we find that the correlation between price and overall rating is small and negative, suggesting that individuals on average enjoy more expensive wines slightly less. For individuals with wine training, however, we find indications of a positive relationship between price and enjoyment. Our results are robust to the inclusion of individual fixed effects, and are not driven by outliers: when omitting the top and bottom deciles of the price distribution, our qualitative results are strengthened, and the statistical significance is improved further. Our results indicate that both the prices of wines and wine recommendations by experts may be poor guides for non-expert wine consumers.
Do More Expensive Wines Taste Better?
Evidence from a Large Sample of Blind Tastings
Robin Goldsteiny, Johan Almenbergz, Anna Dreberx,
John W. Emerson{
, Alexis Herschkowitschy, and Jacob Katzy
SSE/EFI Working Paper Series in Economics and Finance No. 700
April 24, 2008
Abstract
Individuals who are unaware of the price do not derive more enjoyment
from more expensive wine. In a sample of more than 6,000 blind tastings,
we …nd that the correlation between price and overall rating is small and
negative, suggesting that individuals on average enjoy more expensive
wines slightly less. For individuals with wine training, however, we …nd
indications of a positive relationship between price and enjoyment. Our
results are robust to the inclusion of individual …xed e¤ects, and are not
driven by outliers: when omitting the top and bottom deciles of the price
distribution, our qualitative results are strengthened, and the statistical
signi…cance is improved further. Our results indicate that both the prices
of wines and wine recommendations by experts may be p oor guides for
non-exp ert wine consumers.
Keywords: Wine, price/quality relation, expertise.
JEL codes: L15, L66, M30, Q13.
We thank Jacopo Anselmi, Zoe Chance, Shane Frederick, Richard Friberg, Barry
Goldstein, Erik Grönqvist, Daniel Horwitz, Roy Ip, Magnus Johannesson, Thomas
Pfei¤er, and Sue Stubbs for helpful comments and suggestions. Johan Almenberg
thanks the Ragnar and Torsten Söderberg Foundations for …nancial support, and Jo-
han Almenberg and Anna Dreber thank the Jan Wallander and Tom Hedelius Foun-
dation for …nancial support. The Program for Evolutionary Dynamics is sponsored by
J. Epstein.
yFearless Critic Media, 2011B Bouldin Avenue, Austin, TX 78704. Corresponding
author (Goldstein). Email: robin@fearlesscritic.com.
zDepartment of Economics, Stockholm School of Economics, and the Research
Institute of Industrial Economics, Stockholm. Corresponding author. Email: jo-
han.almenberg@hhs.se.
xDepartment of Economics, Stockholm School of Economics, and Program for Evo-
lutionary Dynamics, Harvard University.
{Department of Statistics, Yale University.
1
1 Introduction
When symbolic content is an important part of consumption, the enjoyment of
a good might become decoupled from its innate qualities. The symbolic content
of a price tag has been emphasized in marketing research (e.g., Cialdini, 1998).
At the same time, when goods with similar characteristics di¤er in price, a
reasonable prior is that the more expensive good will, on average, be of a higher
quality. People have been shown to expect a positive correlation between price
and quality (e.g., Rao and Monroe, 1989). Consistent with this expectation,
a meta-analysis reports positive correlations between price and quality ratings
for most, but not all, of 1,200 product markets, but also …nds that the range of
these correlations is very large (Tellis and Wernerfelt, 1987).
For some goods, there is much heterogeneity in consumer tastes, making
it harder to infer quality from revealed preferences. Nonetheless, a reasonable
prior is that consumers on average will derive more enjoyment from the good
with the higher price. Previous research suggests that wine might be a good
where consumer tastes are highly heterogeneous (Amerine and Roessler, 1976;
Lecocq and Visser, 2006). While individuals may frequently disagree over which
wine they prefer, the above hypothesis suggests a positive correlation between
the enjoyment of a wine and its price.
Many factors, such as peer consumption and marketing actions, can in‡uence
how a good is experienced. Price may in itself be such a factor. Recent research
has shown that individuals appreciate the same wine more when they think that
it is more expensive (Brochet, 2001; Plassmann et al., 2008). In other words,
the price of a good a¤ects the experienced utility derived from that good. Thus,
to test the conjecture mentioned above, we need to examine the enjoyment of
wine when individuals are unaware of the price.
In this paper, we use a large sample of more than 6,000 US blind tast-
ings, compiled by food and wine critic Robin Goldstein. Blind tastings o¤er
the opportunity to isolate the experience of the wine itself from psychological
confounds related to its price, presentation or published expert ratings.
We investigate the relationship between price and sub jective appreciation
of wines, when the price is unknown to the tasters. Subjective appreciation is
measured by overall ratings assigned to wines by individual participants.
Our main …nding is that, on average, individuals who are unaware of the
price do not derive more enjoyment from more expensive wine. In fact, they
enjoy more expensive wines slightly less.
We use an ordered probit estimator, as well as a linear estimator (OLS).
In both cases, we use robust standard errors. The ordered probit estimator is
particularly well suited to an ordinal dependent variable, but we …nd that OLS
also performs well, and yields estimates that are easier to interpret. In any case,
the two models generate highly consistent results. The dependent variable is the
overall rating assigned by an individual to a wine. The key independent variable
is the price of the wine, expressed as the natural logarithm of the average retail
2
price per 750 mL in US dollars.1
In our baseline model, we regress the overall rating on the price of the wine,
using both estimators. If individuals found that more expensive wine tasted
better, the coe¢ cient on price would be positive. Our baseline model allows us
to reject this hypothesis: the coe¢ cient on price is negative, and statistically
signi…cant, regardless of which of the two estimators we use.
Next, we extend the model by taking into account that about 12% of the
participants in the blind tastings had some form of wine training, such as a som-
melier course. A number of studies have reported positive correlations between
price and subjective appreciation of a wine for wine experts (e.g., Oczkowski,
1994; Landon and Smith, 1997; Benjamin and Podolny, 1999; Schamel and
Anderson, 2003; Lecocq and Visser, 2006).
In the extended model, we allow for the possibility that individuals with wine
training (hereafter: “experts”) experience wines di¤erently from non-experts.
We include a dummy variable for being an expert, as well as an interaction
term for price and the expert dummy. In a linear regression, this allows both
the intercept and the slope coe¢ cient to di¤er between the two groups.
Previous research suggests that non-experts may not be particularly sensitive
to some of the re…nements that are held in high esteem by wine a…cionados.
Weil (2001, 2005) uses the following experimental setup: two bottles of wine are
poured into four containers. Tasters are then given three of the containers and
asked to distinguish which one di¤ers from the other two. A random guess has
1/3 chance of being correct. In Weil (2001), the two wines are identical apart
from year, but one wine is from a “good” vintage, and the other from a “bad”
vintage.2The tasters get it right 41% of the time – only marginally better
than randomized guessing.3In Weil (2005) the wines are a reserve bottling and
a regular bottling, from the same producer and year. The fraction of correct
answers is merely 40%.
Moreover, Weil …nds that even when tasters can distinguish between the
vintages, they are about as likely to prefer the good one as the bad one. And
among those that can distinguish the reserve bottling from the regular bottling,
only half prefer the reserve. In both cases, the wines di¤er in price by an order
of magnitude.
Our data also indicates that experts, unlike non-experts, on average assign
as high or higher ratings to more expensive wines. The coe¢ cient on the
expert*price interaction term is positive and highly statistically signi…cant. The
price coe¢ cient for non-experts is negative, and about the same size as in the
baseline model. The net coe¢ cient on price for experts is the sum of these two
coe¢ cients. It is positive and marginally statistically signi…cant.
1If we didn’t do this, we would b e expecting a one dollar increase to have the same e¤ect at
the $5 p rice level as at the $50 price level. We get similar qualitative results using the dollar
prices, but the statistical signi…cance of the co cients is n ot as good (but still signi…cant).
2Weil uses pairs for which the fam ous wine critic Robert Parker has rated one of the b ottles
“average to “appalling and the other bottle “excellent” to “the …nest.
3All the sign i…cant di¤eren ce is driven by the testers’ ability to distinguish b etween the
goo d and bad v intages from Bordeaux Pomerol.
3
The linear estimator o¤ers an interpretation of these e¤ects. In terms of
a 100 point scale (such as that used by Wine Spectator), the extended model
predicts that for a wine that costs ten times more than another wine, non-
experts will on average assign an overall rating that is about four points lower,
whereas experts will assign an overall rating that is about seven points higher.
The magnitude of these e¤ects is moderate, but non-negligible given that
wine prices cover a large range. In this sample alone, prices range from $1.65
to $150. In wine markets in general, the range is even larger.
We test the robustness of our results by adding individual …xed e¤ects to
our model. This does not a¤ect the qualitative results, and the coe¢ cients
themselves change only slightly, regardless of whether we use ordered probit or
OLS.
To make sure that our results are not driven by wines at the extreme ends of
the price range, we also estimate the extended model using a reduced sample,
omitting observations in the top and bottom deciles of the price distribution.
We use both ordered probit and OLS, with and without individual …xed e¤ects.
We …nd the same qualitative results with the reduced sample. In fact, the e¤ects
are larger and the statistical signi…cance improves even further. In other words,
our …ndings are even more pronounced when looking only at mid-range price
levels.
Our paper is organized as follows. In section 2, we describe our data. In
section 3, we present our econometric model and report the regression estimates.
We also perform a robustness check. We conclude in section 4, where we discuss
some implications of our results and suggest directions for future research.
2 Data
The data set contains 6,175 observations from 17 blind tastings organized by
Robin Goldstein. The blind tastings took place in the US between April 2007
and February 2008. In total, 506 participants tasted wine ‡ights composed from
523 di¤erent wines. The wines were presented in a double-blind manner, so that
neither the person serving the wine nor the person tasting the wine knew the
identity and/or price of the wine. Each taster assigned an overall rating to every
wine tasted, prior to discussing the wines with the rest of the group. The rating
was the response to the question “Overall, how do you …nd the wine?”, and
the available answers were “Bad”, “Okay”, “Good”, and “Great”. In the data,
these alternatives are coded from 1 to 4, with 1 corresponding to “Bad”and 4
corresponding to “Great”.4The price per bottle ranged from $1.65 to $150. The
prices are average retail prices and were obtained from www.wine-searcher.com.
The wines represent a broad variety of types (e.g. red, white, rosé, sparkling),
country origins, and grapes.
4Tasters ticked one of four boxes. In about 3% of the sample, tasters ticked in between
two boxes, su ggesting a rating somewhere in b etween the two resp onses. For simplicity, we
dropped these observations from the regression. Including them makes no di¤erence to ou r
qualititative results, an d the changes to the estimates are negligible.
4
The participants were unpaid volunteers, ranging from 21 to 88 years of age.
Selection bias is a concern with any voluntary subject pool, and we have no
reason to think that this is an exception. It is quite likely that the sample
contains an over-representation of highly educated individuals, and an overrep-
resentation of individuals working in the food and wine industries. Nonetheless,
the size of the sample and the general diversity of the tasters lead us to hope
that inference will not be too restricted. For a more extensive description of the
blind tastings, please see Chapters 8 and 9 and Appendix 1 in Goldstein (2008).
3 Regression Analysis
Throughout the regression analysis, we use both an ordered probit estimator
and a linear estimator (OLS). In both cases, we consistently use robust standard
errors. The dependent variable is the overall rating, measured on a scale from
1 to 4, with 4 being the highest rating. The main independent variable is the
price variable, expressed as the natural logarithm of the average retail price per
750 mL in US dollars.
In Model 1, we regress the overall rating assigned to wine i, by individual
j, on the price of the wine. About 12% of participants had some wine training,
such as a sommelier course. In Model 2, we allow for the possibility that these
"experts" rate wines in a di¤erent manner. We include a dummy variable for
being an expert, as well as an interaction term for price and the expert dummy.
In a linear regression, this allows both the intercept and the slope coe¢ cient to
di¤er for experts and non-experts. In terms of the linear speci…cation, we can
write these two models as
yi=0+1ln(Pi) + i(1)
and
yij =0+1ln(Pi) + 2EX P ERTj+3ln(Pi)E XP E RTj+i(2)
where Piis the price of wine i, and E XP ER Tjis a dummy variable indicating if
taster jhas wine training. If individuals found that more expensive wine tasted
better, the correlation between overall rating and price would be positive. In our
sample, this is not the case: for both the ordered probit estimates and the OLS
estimates, the coe¢ cient on price is negative. In Model 1, the OLS coe¢ cient is
about -0.04, implying that a 100% increase in ln(price) is associated with a 0.04
reduction in the overall rating. The negative e¤ect for more expensive wines is
statistically signi…cant.
Unlike the non-experts, experts assign as high, or even higher, ratings to
more expensive wines. Model 2 shows that the correlation between price and
overall rating is positive – or, at any rate, non-negative - for experts. The
price coe¢ cient for non-experts is still negative, of about the same size as be-
fore, and with greatly improved statistical signi…cance. The coe¢ cient on the
ln(price)*expert interaction term is highly statistically signi…cant (ordered pro-
bit p-value: 0.017; OLS p-value: 0.015). For experts, the net coe¢ cient on price
5
is the sum of the two, i.e, about 0.11 for the ordered probit and 0.09 for OLS.
This net coe¢ cient is marginally signi…cantly di¤erent from zero (ordered probit
p-value: 0.099; OLS p-value: 0.095). A full set of estimates is shown in Table 1.
Table 1. Dependent variable: overall rating.
(1) (2) (3) (4)
ln(Price) -0.047 -0.061 -0.038 -0.048
(0.039)** (0.013)** (0.038)** (0.012)**
ln(Price)*Expert 0.171 0.138
(0.017)** (0.015)**
Expert -0.558 -0.448
(0.001)*** (0.001)***
Constant 2.297 2.337
(0.000)*** (0.000)***
Observations 5986 5972 5986 5972
R2/pseudo-R 20.000 0.002 0.001 0.005
Robust p-values in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%
Ordered probit
OLS
In sum, we …nd a non-negative relationship between price and overall rating
for experts. Due to the marginal signi…cance of the price co cient for experts,
it remains an open question whether this coe¢ cient is positive, but our results
indicate that this is in fact the case.
How large are these price e¤ects? The coe¢ cients are of a moderate mag-
nitude, but non-negligible, given that wine prices cover a large range. In this
sample alone, prices range from $1.65 to $150. In wine markets in general, the
range is even larger. Suppose we have two wines, A and B, and Wine A costs
ten times more then Wine B in dollar terms. In terms of a 100 point scale (such
as that used by Wine Spectator), the OLS estimation of Model 2 predicts that
non-experts will assign an overall rating that is four points lower for wine A,
whereas experts will assign an overall rating that is seven points higher.5
In addition, the expert dummy is negative, quite sizable, and statistically
signi…cant (OLS expert dummy coe¢ cient: -0.448; p-value: 0.001). In other
words, the OLS estimation of Model 2 consists of two linear relationships, one
with a higher intercept but a negative slope and one with a lower intercept but
a positive slope. The point where the two lines cross each other is the price level
at which experts and non-experts are expected to assign the same rating. If we
take the model literally, this point occurs at the price of $25.70, i.e., ln(price)
= approx. 3.25. At this price, the model predicts that both groups will assign
a rating of about 2.2. Below this price, the model predicts that experts will
assign lower ratings to a wine than non-experts, and vice versa.
5If the dollar price increases by a factor of 10, ln(price) increases by about 2.3. Hence th e
predicted e¤ect on the overall rating of tenfold increase in the dollar price is 2.3 times the
ln(price) co cient for non-experts an d experts, respectively, adjusted to a 100 point scale.
6
We also test a third model, including individual …xed e¤ects. In terms of
the linear speci…cation, Model 3 can be written as
yij =0+j+1ln(Pi) + 2ln(Pi)EX P ERTj+i(3)
where jis a dummy for each individual taster. Including individual …xed
ects has very little e¤ect on the qualitative results and the minor di¤erences
only serve to reinforce our earlier conclusions, as both the negative e¤ect for
non-experts and the positive e¤ect for experts become slightly stronger. These
results are presented in Table 2. For each of the four regressions in Table 2, a
Wald test rejects that the …xed e¤ects are jointly equal to zero, by a wide margin
(p-value <0.001), suggesting that this is a suitable addition to the model.
Table 2. Individual fixed effects.
(1) (2) (3) (4)
ln(Price) -0.070 -0.089 -0.050 -0.064
(0.007)*** (0.001)*** (0.009)*** (0.002)***
ln(Price)*Expert 0.209 0.151
(0.011)** (0.013)**
Individual fixed effects Yes Yes Yes Yes
Constant 2.487 2.183
(0.000)*** (0.000)***
Observations 5986 5972 5986 5972
R2/pseudo-R 20.080 0.081 0.181 0.182
Robust p-values in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%
Ordered probit
OLS
To make sure that our results are not driven by wines at the extreme ends of
the price distribution, we also run our regressions on a reduced sample, omitting
the top and bottom deciles of the price distribution. Given the broad range of
prices in the sample, this is an appropriate precaution. The remaining wines
range in price from $6 to $15.
Using the reduced sample, we estimate Model 2 with and without individual
xed e¤ects. Doing so produces consistent and even more pronounced estimates.
The coe¢ cient on price is still negative, and in each case larger than when using
the full sample. The statistical signi…cance of the coe¢ cients improves further,
7
and the R-squared is higher. These estimates are presented in Table 3.
Table 3. Reduced sample1, with and without individual fixed effects.
(1) (2) (3) (4)
ln(Price) -0.225 -0.173 -0.182 -0.122
(0.001)*** (0.019)** (0.001)*** (0.025)**
ln(Price)*Expert 0.523 0.515 0.421 0.364
(0.002)*** (0.006)*** (0.002)*** (0.009)***
Expert -1.301 -1.044
(0.000)*** (0.000)***
Individual fixed effects No Yes No Yes
Constant 2.622 1.910
(0.000)*** (0.000)***
Observations 4817 4817 4817 4817
R2/pseudo-R 20.003 0.094 0.007 0.206
Robust p-values in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%
1) The reduced sample excludes the top and bottom deciles of the price distribution.
Ordered probit
OLS
In sum, we use the reduced sample to check the robustness of our results
with regard to mid-range price levels. Based on the above, we conclude that
our results are not only robust but in fact even more pronounced when omitting
observations at the extremes of the price distribution.
4 Conclusion
The pleasure we get from consuming wine depends both on its intrinsic qualities
such as taste and smell and external attributes such as price and presentation.
One may argue that the former in‡uences our subjective appreciation through
a bottom-up process, where the sensory apparatus plays a key role, and that
the latter works through a top-down process, where beliefs and expectations
about quality are important determinants.6In this paper we have explored the
bottom-up e¤ects, by looking at how participants in blind tastings rate wines.
We …nd that, unless they are experts, individuals who are unaware of the price
enjoy more expensive wines slightly less.
There is a large relevant literature related to marketing. Lee et al. (2006)
look at how knowledge of a beer’s ingredients (normal beer with added vinegar)
can a¤ect sub jective appreciation. They show that the timing of the information
plays a substantial role. One group of tasters is told about the vinegar, tastes the
beer, and assigns ratings. A second group is told about the vinegar after tasting
6This, in turn, m ight depend on ulterior motives such as status concern s. Wine as a status
signal, and the prospect that expensive wine could function as a positional goo d, is discussed
in Goldstein (2008), chapter 5.
8
the beer, but before the ratings are assigned. On average, individuals in the …rst
group assign signi…cantly lower ratings, suggesting that informing participants
about the vinegar in‡uences the experience in itself. Using fMRI, McClure et
al. (2004) …nd that having the subject’s favourite brand’s name on a drink
makes it taste better than if it is unlabeled. In another fMRI study, Plassmann
et al. (2008) test whether marketing actions such as changes in the price can
in‡uence the experienced pleasantness of a product such as wine. Testers are
given di¤erent wines that they are told di¤er in price. In reality, some of the
wines are the same but simply presented with di¤erent prices. Prices are found
to correlate positively with experienced pleasantness, measured through both
subjective reports and fMRI scans.
Marketing provides one channel through which consumers can be in‡uenced
to buy certain wines. But it is not the only one: wine critics/experts may
also play a role in a¤ecting wine prices and shaping consumer preferences. For
example, Hadj Ali et al. (2007) …nd a positive e¤ect of wine critic Robert
Parker’s ratings on the price of Bordeaux wine.
There is, however, some research expressing skepticism towards wine ratings
and their use for the average wine drinker. According to Quandt (2007), many
wine ratings do not actually convey any information, nor is there substantial
agreement in ratings by experts. Consistent with this view, Weil (2007) inves-
tigates whether wine descriptions of experts actually convey any information to
wine consumers. This is tested by having testers match wine descriptions to
wines. In a similar setup to Weil (2001, 2005), tasters are asked to distinguish
the odd one out of three di¤erent glasses of wine. Only about 50% of the par-
ticipants can distinguish the odd one out, and of those that manage to do it,
only about half can correctly match a wine critic’s description of the wine with
the wine itself which is no better than a random guess.
Our results indicate another reason for why the average wine drinker may
not bene…t from expert wine ratings: he or she simply doesn’t like the same
types of wines as experts. This is consistent with Weil (2001, 2005), who …nds
that even among the subset of tasters who can distinguish between good and
bad vintages, or reserve or regular bottlings, they are as likely to prefer the
“better”one as the “worse”one.
These …ndings raise an interesting question: is the di¤erence between the
ratings of experts and non-experts due to an acquired taste? Or is it due to
an innate ability, which is correlated with self-selection into wine training?7
Investigating this further would be a fruitful avenue for future research.
In sum, in a large sample of blind tastings, we …nd that the correlation
between price and overall rating is small and negative. Unless they are experts,
individuals on average enjoy more expensive wines slightly less. Our results
suggest that both price tags and expert recommendations may be poor guides
for non-expert wine consumers who care about the intrinsic qualities of the wine.
7For a furth er discussion, see Chapter 4 of Goldstein (2008).
9
REFERENCES
Amerine, M A and E B Roessler (1976) Wines: Their Sensory Evaluation. San
Francisco: W.H. Freeman and Company.
Benjamin, B A and J M Podolny (1999) Status, Quality, and Social Order in
the California Wine Industry. Administrative Science Quarterly 44:563-589.
Brochet, F (2001) Chemical Object Representation In The Field of Conscious-
ness. Working paper, General Oenology Laboratory, France.
Cialdini, R B (1998) In‡uence: The Psychology of Persuasion. New York:
Collins.
Goldstein R (2008) The Wine Trials. Austin: Fearless Critic Media.
Hadj Ali, H, S Lecocq and M Visser (2007) The Impact of Gurus: Parker Grades
and en Primeur Wine Prices. AAEA Working Paper No.1.
Landon, S and C E Smith (1997) The Use of Quality and Reputation Indicators
by Consumers: The Case of Bordeaux Wine. Journal of Consumer Policy 20:
289-323.
Lecocq, S and M Visser (2006) What Determines Wine Prices: Objective vs.
Sensory Characteristics. Journal of Wine Economics 1(1): 42-56.
Lee, L, S Frederick and D Ariely (2006) Try It, You’ll Like It- The In‡uence of
Expectation, Consumption, and Revelation on Preferences for Beer. Psycholog-
ical Science 17(12): 1054-1058.
McClure, S M, J Li, D Tomlin, K S Cypert, L M Montague, P R Mon-
tague (2005) Neural Correlates of Behavioral Preference for Culturally Familiar
Drinks. Neuron 44: 379-387
Oczkowski, E (1994) A Hedonic Price Function for Australian Premium Wine.
Australian Journal of Agricultural Economics 38: 93-110.
Plassmann, H, J O’Doherty, B Shiv and A Rangel (2008) Marketing Actions Can
Modulate Neural Representations of Experienced Pleasantness. Proceedings of
the National Academy of Sciences 105(3): 1050-1054.
Quandt, R E (2007) On Wine Bullshit: Some New Software. Journal of Wine
Economics 2(2): 129-135.
Rao, A R and K B Monroe (1989) The E¤ect of Price, Brand Name, and
Store Name on Buyers’Percetions of Product Quality: An Integrative Review.
Journal of Marketing Research 36: 351-37.
10
Schamel, G and K Anderson (2003) Wine Quality and Varietal, Regional and
Winery Reputations: Hedonic Prices for Australia and New Zealand. The Eco-
nomic Record 79(246): 357-369.
Tellis, G J and B Wernerfelt (1987) Competitive Price and Quality Under Asym-
metric Information. Marketing Science 6(3): 240-253.
Weil, R L (2001) Parker v. Prial: The Death of the Vintage Chart. Chance
14(4): 27-31.
Weil, R L (2005) Analysis of Reserve and Regular Bottlings: Why Pay for a
Di¤erence Only the Critics Claim to Notice? Chance 18(3): 9-15.
Weil, R L (2007) Debunking Critics’Wine Words: Can Economists Distinguish
the Smell of Asphalt from the Taste of Cherries? Journal of Wine Economics
2(2): 136-144.
11
... In the treatment group, we utilize Ultralytics YOLOv8, an advanced deep-learning and computer-vision tool, to anonymize players while maintaining the integrity of their movement. We measure performance through the ''blind tasting'' method, where evaluators rate performance without recognizing the players' identities, similar to wine tastings where the wines' identities are not disclosed (Goldstein et al., 2008). This experimental design allows us to measure the causal effect of superstar status on performance evaluations. ...
... Although the effect size of approximately 0.1 is modest, nuanced differences can drive considerable disparities in exposure and wages between superstars and non-superstars. The results support previous findings from blind experiments, where consumers show a similar bias for pricey and labeled wines (Goldstein et al., 2008). Our results indicate that superstars do not necessarily perform and contribute more than non-superstars (Jedelhauser et al., 2023); rather, they are perceived to be better. ...
... Price estimation judgments might however, also be taken to reflect the judgment of a wine's perceived quality. 8 Now while in many product categories price (as a proxy for product quality) and hedonic evaluation tend to be correlated, wine is one of the (possibly) few product categories where the research shows either no correlation (e.g., Almenberg & Dreber, 2011;Ashton, 2014;Harrar et al., 2013;Sample, 2011;Werner et al., 2021), or else even a slightly negative correlation between price and liking, at least amongst regular (that is, non-expert) consumers when tasting wineblind (see Goldstein, Almenberg, Dreber, Emerson, Herschkowitsch, & Katz, 2008). Robert Goldstein, a prominent North American food and wine critic, and his colleagues (see Goldstein et al., 2008) conducted a meta-analysis 5 Note also the participants were not asked how much they liked the wines in this study. ...
... 8 Now while in many product categories price (as a proxy for product quality) and hedonic evaluation tend to be correlated, wine is one of the (possibly) few product categories where the research shows either no correlation (e.g., Almenberg & Dreber, 2011;Ashton, 2014;Harrar et al., 2013;Sample, 2011;Werner et al., 2021), or else even a slightly negative correlation between price and liking, at least amongst regular (that is, non-expert) consumers when tasting wineblind (see Goldstein, Almenberg, Dreber, Emerson, Herschkowitsch, & Katz, 2008). Robert Goldstein, a prominent North American food and wine critic, and his colleagues (see Goldstein et al., 2008) conducted a meta-analysis 5 Note also the participants were not asked how much they liked the wines in this study. 6 In this case, the researchers conclude by noting that: "Our finding that laypeople can be tricked to find budget wines more pleasant by deceptive higher pricing could be considered a two-edged sword as this could both be used to enhance consumers experience as well as wine-sellers [sic] profit. ...
... the role of different levels of experience in taste divergence (see Goldstein et al., 2008 for wine; McAuley and Leskovec, 2013 for beer). Our analysis suggests a new mechanism that can also account for part of this discrepancy-at least for red wine-namely, that the two crowds seem to differ in the scope of their evaluation. ...
Article
Full-text available
Crowdsourcing platforms—such as Vivino—that aggregate the opinions of large numbers of amateur wine reviewers represent a new source of information on the wine market. We assess the validity of aggregated Vivino ratings based on two criteria: correlation with professional critics’ ratings and sensitivity to weather conditions affecting the quality of grapes. We construct a large, novel dataset consisting of Vivino ratings for a portfolio of red wines from Bordeaux, review scores from professional critics, and weather data from a local weather station. Vivino ratings correlate substantially with those of professional critics, but these correlations are smaller than those among professional critics. This difference can be partly attributed to differences in scope: Whereas amateurs focus on immediate pleasure, professionals gauge the wine’s potential once it has matured. Moreover, both crowdsourced and professional ratings respond to weather conditions in line with what viticulture literature has identified as ideal, but also hint to detrimental effects of global warming on wine quality. In sum, our results demonstrate that crowdsourced ratings are a valid source of information and can generate valuable insights for both consumers and producers.
... While various studies have examined consumer opinions on quality and willingness to pay for organic wines (e.g. Goldstein et al., 2008;Hardt, 2012;Janssen & Zander, 2014;Mann et al., 2012;Stöckl et al., 2024), and the relationship between wine prices and ratings, the Austrian fine wine market lacks a comprehensive comparison of quality and prices across different viticultural practices. ...
... On the other hand, relying on one expert makes the algorithm susceptible to biases and noise of the expert. Despite their extensive training, even experts are susceptible to noise in their decision process (Kahneman et al., 2021;Hasan et al., 2023;Koriat, 2012;Goldstein et al., 2008;Litvinova et al., 2022;Kurvers et al., 2023) and might even make inconsistent decisions on the same image (Hasan et al., 2023Litvinova et al., 2022). Hence, there seems to be a need to not just rely on one expert but take multiple readings to reduce the noise in the final decision. ...
Article
Full-text available
A crucial bottleneck in medical artificial intelligence (AI) is high-quality labeled medical datasets. In this paper, we test a large variety of wisdom of the crowd algorithms to label medical images that were initially classified by individuals recruited through an app-based platform. Individuals classified skin lesions from the International Skin Lesion Challenge 2018 into 7 different categories. There was a large dispersion in the geographical location, experience, training, and performance of the recruited individuals. We tested several wisdom of the crowd algorithms of varying complexity from a simple unweighted average to more complex Bayesian models that account for individual patterns of errors. Using a switchboard analysis, we observe that the best-performing algorithms rely on selecting top performers, weighting decisions by training accuracy, and take into account the task environment. These algorithms far exceed expert performance. We conclude by discussing the implications of these approaches for the development of medical AI.
... Das prominenteste Beispiel für den MPE im Marketing dürfte wohl das Phänomen der preisinduzierten Qualitätswahrnehmung sein (Shiv, Carmon & Ariely, 2005a;Goldstein et al., 2008). Demzufolge kann selbst dann, wenn der für ein Produkt gezahlte Preis in keiner Beziehung mit dessen objektiver Qualität steht, die unbewusste Vorstellung der Konsument*innen über das Verhältnis zwischen Preis und Qualität die subjektive Qualitätserfahrung mit dem Produkt beeinflussen (Shiv et al., 2005a). ...
Article
Die Moralisierung der Märkte hat einen spürbaren Einfluss sowohl auf die marktorientierte Unternehmensführung als auch auf das Markenmanagement. Der in diesem Zusammenhang unlängst diskutierte Ansatz des "Markenaktivismus" kann dabei als ein Instrument der moralischen Differenzierung verstanden werden. Sein Einsatz birgt aber nicht nur Chancen, sondern auch Risiken, da er eine klare Haltung zu kontroversen Themen erfordert. Das Markenmanagement sollte daher sorgfältig die Vor-und Nachteile abwägen, bevor es diesen Ansatz instrumentalisiert.
... Some authors (Shiv et al., 2005) suggest that pricing decisions might also affect objective quality: when the consumption is based on credence attributes, a higher (lower) price may lead to higher (lower) experience utility, so that consumers may gain more pleasure from consuming an expensive wine relative to a cheaper one when other quality dimensions are identical. This hypothesis is confirmed by experiments showing that when price information is omitted in blind tests, consumers tend to assign a slightly lower rate to expensive wines compared to cheaper wines (Goldstein et al., 2008). ...
Chapter
Since whisky can be described as a blend of chemicals, and since each chemical ingredient has its own aroma, it seems reasonable to think that the flavor of whisky is determined by its chemical constitution. This Democritean perspective has been studied by flavor chemists and is the main focus of this chapter. Chapter 9 expands this view by considering the possible role of microscopic structures.
Article
Full-text available
Is it possible to exploit cognitive biases so that a non-professional taster prefers one wine to several other absolutely identical wines? To address this question, three complementary experiments were carried out. Each time, five wines were tasted blind in a tasting laboratory by 24 to 34 tasters. Converging evidence from the experiments shows that participants were not capable of identifying that some of the wines they were tasting were absolutely identical. Moreover, the results show that by providing information about the wines’ ratings, prices, or reputation, tasters’ expectations can be modified, and, as a result, their evaluations of the wines can be altered. Specifically, we show that it is possible to modify the ranking between different wines and to get tasters to prefer a wine over other absolutely identical wines. Finally, a surprising finding was that experienced tasters express stronger opinions and adapt their evaluations more strongly after being given manipulative information on the wines they taste.
Article
Full-text available
We present an analysis of equilibrium in markets with asymmetrically informed consumers. Some consumers know both price and quality of all sellers, whereas others know neither but may search among sellers. The equilibrium correlation between price and quality generally increases with the level of information in the market and can be negative when this level is sufficiently small. A meta-analysis of the available empirical studies strongly supports the model's predictions.
Article
The inspiration for the exegesis that follows comes directly from Harry G. Frankfurt, distinguished emeritus professor of philosophy at Princeton University, who recently published an enormously successful essay on the subject of bullshit (Frankfurt, 2005). In the author's view, “One of the most salient features of our culture is that there is so much bullshit. Everyone knows this. Each of us contributes his share. But we tend to take the situation for granted. Most people are rather confident of their ability to recognize bullshit and to avoid being taken in by it. So the phenomenon has not aroused much deliberate concern. We have no clear understanding of what bullshit is, why there is so much of it, or what functions it serves.” (Frankfurt, 2005, p. 1) A fine philosophical insight into bullshit is provided by Bernie Laplante, played by Dustin Hoffman in the movie “Hero.”
Article
I report my tests of the hypothesis that wine consumers cannot match critics' descriptions of wines with the wines themselves. My results suggest that testers' ability to match the descriptions with the wines is no better than random. I report on more than two hundred observations of wine drinkers who engaged in the following experiment. The drinker faces 3 glasses of wine, two of which contain identical wines and the third contains a different wine. I record whether the drinker can distinguish wines—whether he can tell the singleton from the doubleton and, if the drinker can distinguish, which wine he prefers. I present the testers with descriptions of the two wines written by the same wine critic/reviewer. I find that 51 percent of the testers who can distinguish the wines correctly match the description of the wine with the wine itself. The percentage matching does not significantly differ from the expected-if-random half. I have recorded the sex of the testers and I can find that men can distinguish the wines better than random, but women cannot. The differences are so small, even though significant, however, that the Exact F test detects no significant difference between the ability of men and women in these experiments. The results span tests of wines from Bordeaux, Burgundy, the Rhône, Spain, Germany, and Australia; the tests use only still wines, all less than ten years old. (JEL Classification: C93, D12) I dedicate this paper to Adrienne Lehrer, whose 1983 book, Wine and Conversation , with a different experimental design, anticipated these results.
Article
The authors integrate previous research that has investigated experimentally the influence of price, brand name, and/or store name on buyers' evaluations of prod¬uct quality. The meta-analysis suggests that, for consumer products, the relation¬ships between price and perceived quality and between brand name and perceived quality are positive and statistically significant. However, the positive effect of store name on perceived quality is small and not statistically significant. Further, the type of experimental design and the strength of the price manipulation are shown to significantly influence the observed effect of price on perceived quality.
Article
The hedonic technique is applied to wines. In the price equation we include objective characteristics appearing on the label, as well as sensory characteristics and a grade assigned by expert tasters. We have three almost identically structured data sets (two on Bordeaux wines, and one on Burgundy wines). The results are used to make comparisons between two of the most important wine regions in France, and comparisons over time (the two Bordeaux data sets are sampled at different points in time). (JEL Classification: D49.) Another puzzle is the lack of correlation between price and pleasure. Perhaps it is not so surprising that a first-rate example of a little known wine can seem much more memorable than something more famous selling at ten times the price; part of the thrill is the excitement of discovery and the feeling of having beaten the system. What is more extraordinary is the wild price variation at the very top end. Demand bubbles up mysteriously, apparently fuelled by fashion and rumour as much as by intrinsic quality .—Jancis Robinson, Confessions of a Wine Lover , Penguin Books, 1997.
Article
I report my tests of the hypothesis that wine consumers cannot distinguish the difference between regular and reserve bottlings of otherwise-similar wines. My results suggest that only about one pur chaser in five should even consider buying the reserve bottling of a wine, rather than the reg-ular. But since I cannot tell you how to distinguish that one person out of five, we can all aspire to drinking the reserve. I have more than 800 observations of wine drinkers' who engaged in the following experiment. The drinker faces 3 glasses of wine, two of which contain identical wines [either the regular or the reserve] and the third contains a different wine [the other one]. I record whether the drinker can distinguish wines—whether he can tell the singleton from the doubleton and, if the drinker can distinguish, which wine he prefers. I find that just over 40 percent of the drink ers distinguish correctly, whereas one-third could if the process were random. Of those 40 percent who can distinguish, 52 percent prefer the more expensive, reserve wine, whereas half would if the process were random. For this data set, 52 percent does not differ significantly from the expected-if-random half. I have recorded the sex of the testers and I can find that men can distinguish the wines better than random, but women cannot. The differences are so small, even though significant, however, that the Exact F test detects no significant difference between the ability of men and women in these tests. The results span tests of wines from Bordeaux, Bur-gundy, the Rhone, Spain, New Zealand, Italy, California [both red and white], Oregon, and Aus-tralia; the tests include still and sparkling wines.
Article
This paper examines how a producer's status in the market influences its choices about product quality, and the outcomes that result. The authors compare economic models of reputation that emphasize the role of past quality as a source of information about current quality with sociological models of status that emphasize the role of affiliations. They test hypotheses about the complementary effects of status and reputation in an analysis of more than 10,000 affiliation decisions made by 595 wineries over a 10-year period. Results show that actors occupying high-status positions obtain greater benefit from subsequent high-status affiliations than do actors occupying low-status positions. As such, these actors are more willing and able to pay for subsequent high-status affiliations and to use them to advance their position in the larger status ordering. [ABSTRACT FROM AUTHOR]