Concordance or conflict? A 40-year longitudinal study of a South African wine club versus expert ratings and price

Wine selection is a complex decision-making process influenced by a multitude of factors, including sensory preferences, expert evaluations and price. As global wine markets have grown increasingly diverse, consumers have come to rely more on external cues, such as expert ratings, to guide their choices. These ratings, frequently awarded during tasting competitions and specialist publications, are widely regarded as indicators of quality. However, the extent to which they align with general consumer preferences remains debated.

Previous research, such as the study by Schiefer and Fischer (2008), highlights the gap between wine expert ratings and consumer preferences, emphasizing that expert evaluations may not reliably predict the sensory enjoyment of the average consumer. Moreover, consumer preferences tend to be heterogeneous, further complicating the development of universally applicable quality indicators. Schiefer and Fischer (2008)’s findings suggest that while experienced and knowledgeable wine drinkers might align more closely with expert opinions, the broader consumer base may benefit more from alternative approaches, such as sensory descriptors tailored to their tastes.

Building on this foundation, this study examines the relationship between aggregated consumer ratings and expert ratings in a unique and novel panel data set spanning close to four decades. In addition to these ratings, the analysis incorporates factors such as wine price, type (red or white) and the year of the tasting to explore the determinants of wine preference and evaluate the alignment between consumer perceptions and expert evaluations. This contextual approach provides a comprehensive understanding of the dynamics influencing wine preference and quality perception, contributing to ongoing discussions about the practical value of expert ratings in guiding consumer decisions.

The relationship between wine ratings and price premiums has long been a focal point in wine economics. In particular, the understanding of how expert evaluations influence market dynamics is a key question. In this context, Ali et al. (2008) provide compelling evidence of the strong influence of Robert Parker’s ratings on Bordeaux en primeur prices. Their analysis shows that even small increases in expert scores can lead to substantial price premiums (see also Babin and Bushardt, 2019). These findings underscore the powerful role of expert evaluations in shaping consumer perceptions and willingness-to-pay, often inflating prices beyond what intrinsic quality alone would warrant. That said, the degree to which other rating systems (such as consumer-driven reviews or alternative guides) affect price formation remains comparatively underexplored.

Expanding the analysis to the dynamic interaction between ratings and prices, Kaimann et al. (2023) investigate the interplay of quality signals in the wine market using a data set of 13,911 observations over two decades. They demonstrate that expert ratings significantly influence prices, with a one-point increase in rating associated with an 8% price rise. Lagged ratings also exert considerable influence on current scores, indicating path dependence in expert assessments. The study further reveals that expert ratings disproportionately impact high-quality wines, emphasizing the global relevance of these findings across diverse regions and varieties. This aligns with earlier research highlighting the amplifying effects of positive reviews on price premiums for top-tier wines (Villas-Boas et al., 2021; Oczkowski and Pawsey, 2019; Ashton, 2016).

Goldstein et al. (2008) present a pivotal contribution by analysing over 6,000 blind tastings conducted across a diverse sample of participants in the USA. Their study challenges the commonly held assumption that more expensive wines inherently deliver greater enjoyment, revealing a nuanced relationship contingent upon the taster’s level of expertise. For non-expert consumers, there exists a small but statistically significant negative correlation between price and subjective enjoyment, while expert tasters exhibit a non-negative relationship, suggesting that expertise aligns perceptions more closely with premium wine attributes. This finding underscores the limitations of price as a signal of quality for non-experts and highlights the psychological influence of extrinsic cues such as pricing and branding on perceived quality (Plassmann et al., 2008).

Extending this inquiry, Chocarro and Cortinas (2013) examine how expert opinions shape consumer evaluations of wine, emphasizing the moderating effects of information complexity, expert consensus and consumer knowledge. Their experimental design demonstrates that expert ratings significantly influence evaluations, particularly among consumers with low wine knowledge. This group relies on these “weak-tie” sources to reduce uncertainty when evaluating experience goods. By contrast, high-knowledge consumers exhibit greater independence in their evaluations, likely due to their confidence in their own judgement. Interestingly, neither expert consensus nor information complexity significantly affected consumer responses, suggesting a preference for straightforward and accessible ratings. The study highlights the segmentation within the wine market, where differing levels of consumer knowledge necessitate tailored marketing strategies.

Building on this foundation, Oczkowski and Doucouliagos (2015) provide a comprehensive meta-regression analysis of 180 hedonic wine price models spanning two decades and multiple countries. They identify a moderate partial correlation of +0.30 between price and quality ratings, emphasizing the dual role of intrinsic attributes, such as sensory quality, and extrinsic factors, such as producer reputation, in shaping wine prices. Reputation often dominates as a price determinant, yet sensory quality retains a statistically significant impact, particularly when fine-grained quality measures like 100-point scales are used. Their findings demonstrate the value of methodological rigour in uncovering systematic heterogeneity across studies, further broadening the understanding of price–quality relationships in diverse contexts. Where Goldstein et al. (2008) reveal a disconnect between price and enjoyment for non-experts, Oczkowski and Doucouliagos (2015) and Oczkowski (2018) emphasize the broader interplay of sensory quality and reputation, extending the scope of inquiry by analysing how methodological design influences results.

Their analysis is supported by Núñez et al. (2024), who perform a meta-regression of 223 articles estimating hedonic functions and document selection and publication biases. They find robust price elasticities around 5%–6%; quality ratings have a larger impact on price elasticity for red wines; the type of data source matters, but the econometric techniques used do not. The remaining attributes mostly have small influences. Interestingly, the most cited studies are those that show a smaller effect.

Turning to the cognitive dimensions of wine evaluation, Masset and Raub (2023) examine the influence of extrinsic cues − such as price, expert ratings and bottle presentation − on wine quality perceptions and willingness-to-pay. Their experimental study, building on the seminal work of Hodgson (2008), highlights how tasters often fail to identify identical wines served under different conditions, favouring those with higher-status cues. Interestingly, experienced tasters adapt their ratings more strongly to extrinsic information, driven by confidence in their expertise and a desire to align with normative expectations. This work underscores how cognitive biases influence wine evaluation, reiterating findings from earlier studies showing that price cues affect neural processing of taste (Plassmann et al., 2008) and that non-experts derive less enjoyment from expensive wines when blind to price information.

There is also ample South African evidence that extrinsic information shapes perceived quality and purchase cues over time.

Conceptually, Priilaid (2006) collates blind and sighted tastings of South African reds (1993–2001) and show that sighted quality ratings are dominated by extrinsic cues (incl. price and region of origin). Their meta-model showed an adjusted of $R2=0.31$⁠, where price alone explains roughly 84% of variance in sighted assessments. As for the combination of price and origin, this explained about 95% of the variance. The results also showed that intrinsic merit contributes only marginally once the extrinsic cues were controlled for. This divergence between sighted and blind scores implies that extrinsic information can mask underlying quality signals. But what about other cues like awards?

Herbst and Von Arnim (2009) find that although South African consumers recognize awards as a quality cue, they are secondary to variety, vintage, producer reputation, origin and price. Their study showed that more knowledgeable consumers are relatively sceptical and support independent oversight of competitions. They do however note in their study that decision-making is a complex set of interactions, and wine awards do nevertheless play a role in supporting consumer decisions in certain circumstances and for certain customer segments. Generational segmentation further nuances the cue effects.

In a cohort-focused study of South African Generation Y (Gen Y) consumers, Lategan et al. (2017) use best–worst scaling (BWS) on 13 attributes and report that “tasted previously” (⁠$BWS=2.76$⁠) and “someone recommended it” (⁠$BWS=2.32$⁠) are dominant cues. For Gen Y, personal experience and word of mouth in wine selection is most important. Extrinsic heuristics also feature: “brand” (⁠$BWS=1.14$⁠), “in-store promotion” (⁠$BWS=0.79$⁠) and “medal/award” (⁠$BWS=0.10$⁠). These results are consistent with the use of simplified quality signals among less knowledgeable consumers ^[1]. In a broader study that also used BWS across 14 attributes, and featured a national panel (⁠$N=646$⁠), Pentz and Forrester (2020) likewise find “previous tasting” and “peer recommendations” to be most influential, while “in-store promotions” are least valued by older consumers and “low alcohol” by younger consumers. Interestingly “award” featured dead in the middle with a $BWS=0.14$⁠, but had a standard deviation of $σ=1.87$⁠. This latter result signals some respondents value awards, while others actively down-weight them. It is likely that awards act as a secondary cue, and in this study, the individuals use it to support primary signals such as familiarity (prior tasting) and social proof (word of mouth).

The data presented in the next section enable us to examine the “gurus vs geeks” questions that we later empirically test: (1) contrasting expert and community signals (Bazen et al., 2024; Ali et al., 2008; Friberg and Grönqvist, 2012; Oczkowski and Pawsey, 2019), and to (2) quantify the rating–price linkage and its dynamics (Lecocq and Visser, 2006; Oczkowski and Doucouliagos, 2015; Kaimann et al., 2023; Núñez et al., 2024).

A data set was obtained from a wine club that has been in existence since 1987, with the name/producer, cultivar or blend, price, average score and ranking (within each tasting) recorded in a database for most of the wines ever tasted ^[2].

The wine club consists of 12 couples and strives to hold 12 blind tastings per year (i.e. one per month), although holidays and other factors usually result in around 10 tastings annually. Hosting duties rotate equally among the couples. Each couple has carte blanche regarding the topic, venue and presentation of their tasting. The majority of tastings are hosted at members’ homes, although some have taken place in hired venues, restaurants or the wine farms themselves.

The club’s membership has remained remarkably stable over its $±$37-year existence, with nine of the original couples still active. The composition of the members is diverse. One is a Cape Wine Master who produces under their own label and five others are winemakers (one of whom is retired). These six members are regarded as expert tasters within the group. Almost all of the other male members are avid wine lovers with varying degrees of tasting ability. Most of the ladies also enjoy wine and, like the non-winemaker male members, have differing levels of tasting expertise. The average age of the club members is $±$62, meaning the group was in their late twenties when the club was formed.

The club uses a 20-point scoring system, nominally with 3 points for colour, 7 for nose and 10 for taste. However, this breakdown is not “officially” applied, and only the aggregate score (out of 20) is recorded for each wine. Average attendance at each tasting is around nine male and six female members, resulting in $±$15 scores contributing to each wine’s average. The format of nearly all tastings includes a welcome glass (of any kind), followed by the formal tasting (typically 6–8 wines), and then dinner.

Given the club’s demographic composition, it is important to note that gendered consumption differences may influence how tasters use information and form preferences − even in a blind tasting. Evidence reports systematic differences in sensory preference and behaviour between females and males ^[3]. These differences tend to dissipate with experience (Bruwer et al., 2011). Our data set does not include gender-coded scores, so we cannot test these mechanisms directly. We now note this as a limitation and an avenue for future research.

This section draws on two data sources:

1. an informal wine tasting club; and

2. the Platter’s South African Wine Guide.

The wine club, based in Stellenbosch − the epicentre of South Africa’s wine industry and a hub for wine tourism — has held regular tastings since its inception in August 1987. Over the period from 1987 to April 2024, members hosted 352 monthly tastings, evaluating a total of 2,460 wines.

For each tasting, records were kept of the host, date, theme and detailed information on each wine, including its origin, cultivar, price and the group’s average score (on a 20-point scale). While the data set is largely complete, in some instances price or score data were missing. The number of tastings conducted per year is illustrated in Figure 1 ^[4].

Some tastings included categories such as beer, fortified wines, sparkling wines, rosé and non-South African wines. In several cases, price information was also missing. We excluded all six of these categories from the database. Furthermore, we removed all wines for which the Club rating was not recorded. This step was necessary to ensure alignment with our second data source, the Platter’s South African Wine Guide.

Platter’s Guide is a unique annual South African publication that records every wine that is commercially available in the country each year. Wines are tasted during the course of the year in question by a panel of expert wine tasters, and are then rated on a fivestar scale. The tasters are sighted, i.e. tastings are not blind. Each wine in our final database therefore had two ratings, namely that of the Club and the Platter rating. The data set spans close to four decades, offering a rich panel component that allows for the investigation of trends and patterns over space and in time. This meant that the full sample of wines included in the analysis amounted to 693.

The paired wine Club and Platter’s ratings form the basis for the examination of the relationship between expert evaluations, pricing and consumer preferences. The analysis begins with a breakdown of observations per year, further categorized by wine type, offering a view of the data set’s time series characteristics. The distribution of wine types and vintages is then explored to provide insights into the diversity and representativeness of the data set. Finally, we examine the ratings variable, including its range, distribution and trends over time, which are integral to understanding the dynamics of expert evaluations in the wine market.

The variables recorded and used in the analysis are described in Table 1.

The choice of wines that were tasted is quite even with 363 red and 330 white wines included in the data. The data also show that the preferences of the Club change over the course of the year. White wines were generally preferred during the summer months (January to March, but in their case stretching into April) and red wines during most of autumn (fall) and winter (in this case May to August), with some caveats during spring, when white wines were preferred during September to November, but red during December (Figure 2) ^[5].

Another temporal component to wine consumer preferences that is of interest, is the vintage of the wines tasted. Most reds, having different ageing potential dependent on the time in barrel fermentation and time in bottle. Figure 3 shows the estimated mean cellaring (or yearbook − vintage) with confidence intervals for red wines over time ^[6]. The figure illustrates the estimated marginal mean cellaring time per year over a 37-year period. There appears to be a slight increase in the cellaring duration for red wines before tasting. Although this is not the focus in this paper, the trend could reflect the Club’s evolving preferences as they gain access to, and develop a taste for older vintages, or it might indicate a broader change within the wine market.

Besides the type of wine being tasted within this novel data set, a unique aspect that was captured within the notes are the ratings and prices paid for the various wines. Ratings themselves are interesting and much research has been conducted in this space. Two primary methods are commonly used in wine tasting evaluations due to their simplicity: (1) score average ^[7], and (2) the rank average ^[8] (Ashenfelter and Quandt, 1999). Figure 4 shows the distributional comparison between the five-point Platters’s Guide ratings and the 20-point Club rating.

It is relevant to note that some of the earliest scoring systems for wine used the 20-point scale. One notable example is the Davis scorecard ^[9]. This system emphasized deducting points for defects rather than rewarding positive qualities. The system only allocated two points for general quality. Although this method of scoring was effective in identifying faults, the Davis score card was less suited to expressing nuanced distinctions between high-quality wines (Lehrer, 2009)^[10].

Research indicates that the ranking derived from the score average typically demonstrates greater accuracy compared to rankings based on the rank average or alternative methods like the Shapley ranking (Cao and Stokes, 2017). This suggests that using numerical scores directly may provide a more reliable reflection of wine quality compared to rank-based approaches because rank-based approaches can be influenced by ordinal scale limitations and ties.

In recent decades, the 100-point scoring system, introduced by Robert Parker, has become the standard for many wine critics and publications. Unlike the 20-point scales, Parker’s system starts at 50 and awards additional points for various attributes, such as colour, aroma, flavour and overall impression. Its popularity, particularly among American consumers, has been attributed to its resemblance to school grading systems. This has made it more intuitive and accessible (McCoy, 2005). The 100-point scale’s granularity allows for finer distinctions between wines, often influencing pricing and market demand significantly. However, critics of the system argue that it can oversimplify the complexity of wine, reducing subjective experiences to a single numerical value (Lehrer, 2009).

To assess the concordance between the Club’s and Platter’s wine ratings, a Cohen’s weighted kappa (⁠$κw$⁠) was calculated. Weighted kappa adjusts for the severity of disagreements between categories (5 for Platter ratings and 20 for Club ratings, normalized to a common scale). It does this by assigning weights to different levels of disagreement. The weight is calculated as:

where n is the number of categories, $|i−j|$ is the absolute difference between the ratings assigned by the Club and Platter systems and $m≥1$ determines the weight’s sensitivity to differences. For this analysis, $m=2$ was used to calculate the quadratic weighted kappa (QWK), which gives more weight to larger disagreements ^[11]. The weighted kappa statistic is then computed as:

where $pij$ is the observed proportion of times the Club assigned rating i and Platter assigned rating j, and $pi$ and $qj$ are the marginal probabilities for the Club and Platter ratings, respectively.

The results indicated a very low level of agreement, with $κw=0.005$ (95% CI: [0.003, 0.006]), based on 693 observations. The unweighted kappa, which does not consider the weight of disagreements, was 0, further underscoring the minimal concordance between the two rating systems. This finding aligns with the work of Hodgson (2008), who examined the reliability of wine judges in a major US competition and reported substantial inconsistency in scoring. His analysis revealed that individual judges often exhibited significant variability in their evaluations, with some applying narrow scoring ranges and others showing marked discrepancies even when assessing the same wines. Such variation, coupled with the low inter-judge agreement, raised concerns regarding the reliability of competition outcomes as indicators of intrinsic wine quality.

Our findings contribute by showing that both consumer−expert and expert−expert concordance in wine evaluation can be quite limited. Furthermore, studies that compare consumer and expert preferences directly (e.g. Goldstein et al., 2008) frequently observe substantial divergence, particularly for non-iconic wines or those in lower price tiers. Hodgson (2008)’s seminal findings in the US competition context are echoed in research from Australia (Gawel and Godden, 2008) and Europe (Ashton, 2011).

In this section, we leverage our unique longitudinal data set to examine how these different evaluations correlate with price premiums. Using a Bayesian framework, we model the relationship between CPI-deflated wine prices and ratings using a Gamma likelihood with a log link. Our motivation for using a Bayesian framework lies in the fact that this type of analysis offers several advantages over traditional frequentist approaches: non-Gaussian specification suited to strictly positive, right-skewed prices, direct estimation of full posterior uncertainty (with 95% credible/ highest posterior density [HPD] intervals) for parameters and derived quantities and the ability to make direct probabilistic statements about parameters and predictions ^[12]. The method also implements hierarchical (partial) pooling via random intercepts for yearbook and wine farm. This unique Bayesian structure shrinks noisy subgroup estimates towards the grand mean, stabilises inference in sparse cells and improves predictive performance, while retaining between group heterogeneity. Estimation is implemented in brms /Stan with standard diagnostics and posterior-predictive checks reported in  Appendix 2 (Bürkner, 2017; Kruschke, 2021).

All wine prices were deflated to reflect 2024 prices using the consumer price inflation (CPI) index. To facilitate interpretation, all club scores are normalized to the same scale as Platter, with the minimum being cut off at 10. We rescale the Club’s 20-point scores to Platter’s five-point scale to enable direct comparability in coefficient magnitudes and graphical presentation later on in the paper ^[13]:

Building on Ali et al. (2008)’s findings, this analysis estimates the price premium associated with Platter’s ratings and evaluates whether the Club’s ratings also concur with the premium prices. By integrating expert and consumer perspectives, the study seeks to determine the extent to which these ratings converge or diverge in their impact on wine pricing. The following hypotheses are tested:

To address the research questions and test the hypotheses, we use a Bayesian regression model with the following specification, where the independent variable deflated prices (y) is modelled by:

In this model, $β0$ represents the global intercept term, capturing the baseline level of wine prices across all years. The vector of covariates $Xij$ includes main effects and interaction terms between Platter’s and club ratings and wine type. The associated coefficient vector $β$ reflects how ratings and their interactions with wine type relate to wine prices.

There might also be a temporal effect where the whole market for wine shifts due to unobservable factors. In contrast, simultaneity may also arise via (i) anticipatory pricing by producers expecting strong expert scores, (ii) extrinsic cue effects, where higher posted previous prices shape expert perception and (iii) omitted factors such as brand reputation and distribution that affect both ratings and price.

While we cannot measure the aforementioned factors directly, we model them as random deviations from the average price level using the yearbook edition. Likewise, unobserved heterogeneity between wine farms is accounted for through a second random intercept. These latent effects are captured through random deviations $ηj$ for yearbook and $ζk$ for wine farm. Both random intercepts are assumed to follow normal distributions, $ηj∼N(0,ση2)$ and $ζk∼N(0,σζ2)$⁠.

The dependent variable, deflated wine price $yijk$⁠, is assumed to follow a Gamma distribution with mean $μijk$ and shape parameter $α$⁠. This distribution accommodates the positively skewed and non-negative nature of wine prices. The log link ensures that the conditional mean $μijk$ remains positive.

This formulation captures both the main effects of ratings on wine prices and their interaction with wine type, while also accounting for unobserved year- and wine farm-specific effects^[14]. Estimation of the model was done using R and the brms package from Bürkner (2017).

In this section, we analyse the results of the Bayesian model according to Kruschke (2021)BARG framework to elucidate the relationship between wine prices and the independent variables: Platter’s ratings, Club ratings. All model diagnostics are reported on in  Appendix 2.

Table 2 and Figure 5 provide the estimates for the Bayesian model. The model provides evidence supporting the association between wine ratings and deflated prices, as well as insights into the variation in this relationship by wine type. All model estimates were back-transformed from the log scale to the original response scale for interpretability. Results are reported alongside 95% HPD intervals, which reflect highest posterior density credible intervals based on posterior draws.

The intercept (⁠$β0$⁠) (38.1; 95% CI: [23.3, 61.1]) represents the baseline deflated price for red wines when Platter’s and Club ratings are at their reference levels. Both Platter’s and Club ratings were positively associated with deflated prices, supporting the hypotheses that higher ratings from experts and consumers are associated with increased prices. A one-unit increase in Platter’s ratings was associated with a 35% increase in price (1.35; 95% CI: [1.22, 1.48]). This strong and significant association underscores the prominent role of expert evaluations in shaping wine prices. Similarly, a one-unit increase in club ratings was associated with a 21% increase in deflated price (1.21; 95% CI: [1.09, 1.32]), highlighting the relevance of consumer preferences in association with wine prices, though with a smaller effect size compared to expert ratings.

For Platter’s ratings, the interaction term (1.03; 95% CI: [0.91, 1.16]) indicated a negligible attenuation of the price impact for white wines. However, the credible interval includes 1, suggesting no significant difference in the effect of expert ratings between red and white wines. Similarly, the interaction between club ratings and wine type (0.95; 95% CI: [0.83, 1.08]) suggested a modest reduction in the price impact for white wines. While the effect was not statistically meaningful, as the credible interval includes 1, this finding suggests consistency in the evaluation of wine types, both for consumers and experts alike ^[15].

The estimated marginal means (EMMs) reveal that both consumer (Wine Club) and expert (Platter’s) ratings are significantly associated with wine prices, with expert ratings exerting a stronger effect. For red wines rated by the club, predicted prices increase steadily with higher ratings, ranging from 112.0 ([83.9, 143.0]) at a score of 1.0 to 237.0 ([186, 293.0]) at a score of 5.0. Similarly, for Platter’s ratings, red wine prices rise from 89.8 ([64.9, 118.0]) at a score of 1.0 to 296.0 ([242.0, 348.0]) at a score of 5.0. White wines exhibit comparable trends, with prices increasing with higher ratings but consistently remaining below those of red wines for equivalent scores. For example, white wine prices increase from 86.2 ([61.2, 110.0]) to 150.0 ([113.0, 192.0]) across the range of club ratings, while prices for Platter’s ratings increase from 59.5 ([42.8, 77.0]) to 218.9 ([179.0, 259.0]) over the same score range.

Finally, estimated random effects indicate that both producer- and time-specific factors shape wine prices. Variation across wine farms corresponds to a multiplicative factor of 1.31 $×$ ([1.23, 1.39]), meaning that a wine with a baseline predicted price (based on rating) of R150 could range from approximately 115 to 195 simply due to differences between producers. Taken together, these results highlight that both farm-level heterogeneity and temporal fluctuations contribute substantially to observed pricing patterns.

In summary, the results provide robust evidence supporting H1 and H2: higher Platter’s and Club ratings are significantly associated with higher price premiums. Specifically, Platter’s ratings exert a stronger influence on price premiums compared to club ratings. This is consistent with prior findings emphasizing the market power of expert evaluations (Ali et al., 2008; Lecocq and Visser, 2006; Ashton, 2016). A one-unit increase in Platter’s ratings corresponds to an approximate 35% increase in price, whereas the equivalent increase for club ratings is 21%.

In contrast, H3, positing differential effects across wine types, finds no significant empirical support. The interaction terms between ratings and wine type exhibit overlapping credible intervals, suggesting that the effect of ratings on price does not meaningfully vary between red and white wines. Nevertheless, a modest difference is observed in the price trajectories: for the club ratings, the price premium for white wines grows more slowly than for reds, while for Platter’s ratings, the effect size remains proportional across wine types, albeit from a lower base for white wines.

These findings reinforce the substantial role of both expert and consumer evaluations in wine pricing mechanisms. In particular, the dominance of expert ratings aligns with prior meta-analyses highlighting the premium attached to authoritative reviews in hedonic pricing models (Oczkowski and Doucouliagos, 2015; Villas-Boas et al., 2021; Bazen et al., 2024). Additionally, the consistency of the rating−price relationship across wine types suggests that market participants do not systematically differentiate between reds and whites in their responsiveness to quality signals. The latter contributing to the broader literature on price formation in differentiated product markets (Masset and Raub, 2023; Schiefer and Fischer, 2008).

From a practical perspective, these results suggest that wineries and marketers should continue to emphasize expert endorsements in their positioning strategies, given their stronger price impacts relative to consumer-driven ratings. Additionally, the stable rating effects across varietals imply that the concept of a premium wine can be applied across both red and white wine categories without significant adjustment for type. It should be noted, however, that our analysis is confined to retail list prices. In practice, price premia attached to ratings are likely to vary substantially across channels (e.g. restaurants, online discounting or “mystery wine” offerings), and the magnitude of a point difference at the very top of the scale may be attenuated dependent on the settings.

While most studies cited in this paper are based in the USA or Europe, it is worth reflecting on how these results relate to the South African market context. The consistent price premiums observed, despite the low concordance between expert and consumer ratings, may be partly explained by structural and cultural features of the domestic wine industry. South Africa’s premium wine market remains relatively concentrated, with a limited number of producers capturing most of the attention from critics, distributors and consumers ^[16]. This fosters conditions where expert scores can exert outsized influence, especially among higher-income or export-oriented segments.

This reliance of consumers in South African on experts may place heightened value on their authority due to the country’s historical reliance on a small number of gatekeepers in wine journalism and retail (Priilaid, 2006). Experimental and neuroeconomic work also shows that consumers respond strongly to extrinsic cues such as medals and expert scores, often over and above intrinsic quality (Priilaid and van Rensburg, 2012; Plassmann et al., 2008; Masset and Raub, 2023).

Our findings indicate that expert and consumer ratings function as complementary (partially orthogonal) quality signals. Even with low agreement between Platter’s Guide and the Club, both signals are positively associated with price (with the expert effect stronger in magnitude). This has two implications. Firstly, it suggests audience segmentation: more expert‐attentive segments (premium/collectible/export) are likely to respond primarily to critic scores, whereas socially influenced segments (retail/online) may respond more to community ratings (Babin and Bushardt, 2019; Kaimann et al., 2023; Bazen et al., 2024). Secondly, it implies cue complementarities: displaying both consumer and expert ratings can raise perceived quality. Expert endorsements serve as a signal of authority, while community ratings as social proof (Friberg and Grönqvist, 2012; Villas-Boas et al., 2021).

From a producer and intermediary perspective, the aim would be to align information cues with the characteristics of the distribution channel and marketer. In premium and direct‐to‐consumer context (where search costs for expert knowledge are lower and reputational capital is salient), expert evaluations should be displayed more prominently. In mass retail and online environments, where social information is comparatively more diagnostic, a combination of expert scores with community (consumer) ratings may generate complementary signalling effects (authority and social proof). In markets where brand reputation is already strong (Oczkowski, 2018), ratings can be used to calibrate amongst peers and price points rather than determine overall positioning.

This study provides new empirical evidence on the relationship between wine ratings and price formation. The paper leverages a rare longitudinal data set spanning nearly four decades from a private Stellenbosch-based wine tasting club. The analysis integrates expert evaluations from Platter’s South African Wine Guide with aggregated consumer ratings, offering a dual perspective on quality assessment and the formation of price premiums in the wine market.

Despite the Club’s composition of knowledgeable and experienced tasters, agreement with expert ratings (Platter's) was notably low. The near-zero weighted kappa statistic indicates minimal concordance between consumer and expert assessments. This finding strengthens the known subjectivity and variability inherent in wine evaluation (Hodgson, 2008; Schiefer and Fischer, 2008; Goldstein et al., 2008). The low agreement suggests that, even among informed consumers, alternative factors shape wine preferences in ways that diverge from expert norms and intrinsic quality metrics. Despite low agreement between the expert and consumer ratings, low $κ≈0$⁠, both independently predict wine prices. This highlights a key insight: Platter’s and the Club’s ratings capture distinct, partially orthogonal dimensions of perceived quality.

A Bayesian price model confirms that higher ratings from both Platter’s (H1) and the Club (H2) are associated with significant price premiums. The stronger association for expert ratings aligns with research on the signalling value of professional evaluations (Ali et al., 2008; Lecocq and Visser, 2006; Ashton, 2016; Villas-Boas et al., 2021) and extends earlier South African findings on non-linear rating−price relationships (Priilaid and van Rensburg, 2012). H3, proposing differential effects by wine type, finds limited support: although red wines command higher baseline prices, the marginal effects of ratings are largely consistent across red and white wines.

Taken together, these findings contribute to the wine business literature by illustrating how expert and consumer evaluations coexist and influence market outcomes, even when they diverge observationally. Practically, the results suggest that wineries and marketers could benefit from presenting both expert and consumer ratings. This would be particularly true in heterogeneous markets where salient signals drive consumer preferences.

Figure A1 provides an overview of how many of each wines type were tasted each year. It highlights the differing patterns in tasting frequency for red, rosé, sparkling and white wines across five year intervals.

Figure A2 displays traceplots for all estimated parameters across the four MCMC chains. The chains exhibit good mixing and appear stationary after the warmup phase, with no signs of divergence or poor exploration of the posterior space. Visual inspection confirms convergence for all parameters. One minor deviation was observed in Chain 3 for a single coefficient, which shows a slight shift in density mass relative to the others, though this deviation is negligible and does not materially affect posterior summaries.

Posterior density plots for all model parameters are shown in Figure A3. The densities appear smooth and unimodal, with no indication of multimodality or erratic behaviour, suggesting well-behaved posterior distributions. Convergence diagnostics further support these visual impressions. One commonly used diagnostic is the Gelman–Rubin statistic (⁠$R^$⁠), which assesses whether the independent Markov chains have converged to the same target distribution. Values of $R^$ close to 1.00 indicate good convergence; all parameters had $R^$ ≤ 1.01, with most exactly equal to 1.00.

Another key diagnostic is the effective sample size (ESS), which measures the number of independent samples equivalent to the correlated MCMC draws. It is reported for both the bulk of the posterior distribution (Bulk ESS) and its tails (Tail ESS). Larger ESS values imply more reliable estimation. In this model, all Bulk ESS values for regression coefficients exceeded 900 − for example, 1,336 for Platter Score and 1,284 for Wine Club − indicating high precision in posterior summaries. Tail ESS values were similarly robust, all exceeding 1,200. The standard deviation of the year-level random intercept, which draws on fewer observations, still achieved acceptable values (Bulk ESS = 310; Tail ESS = 727).

Together, these diagnostics confirm that the model chains mixed well, converged successfully, and generated a sufficient number of effective samples to support robust posterior inference.

To evaluate the overall fit of the model, we conducted a posterior predictive check, a standard diagnostic procedure in Bayesian analysis. This method involves simulating new data from the posterior distribution of the model and comparing it to the observed data. The idea is straightforward: if the model is a good fit, the simulated (predicted) data should resemble the actual data.

Figure A4 displays the posterior predictive distribution (or PP Check) overlaid on the observed distribution of the deflated wine prices. Visually we can see the model captures the pronounced right-skewed shape of the empirical distribution, which is characteristic of price data and consistent with our choice of a Gamma likelihood combined with a log-link function. Importantly, the predictive distribution aligns well with the observed data across the entire range, including the long right tail, where price outliers often reside.

This result provides strong evidence that the model is appropriately specified for the underlying data generating process. The close correspondence between observed and predicted distributions suggests that key structural features of the data − such as skewness, scale and spread − are accurately captured by the model. As such, the posterior predictive check supports the adequacy of the model for inference purposes.