Behind the Global Wine Score
While wine e-commerce global market represents US$ 10 bn, wine lovers and investors don’t have many indicators to rely on to judge about the wine quality. The main indicators are wine ratings as well as tasting comments coming from various experts (critics) or countless people (collaborative websites like @cellartracker).
We The Global Wine Score, bring simplicity and efficiency in this crowd by giving a single score aggregated from many experts’ ratings. As you may imagine, calculating this score from all those different sources is not done without strong mathematics behind it.
So, what are the principles driving the Global Wine Score?
Diversity of critics
Critics have different tastes, different palates, different indulgences and different rating scales. Even when using the same scale, two experts giving a same value could express different assessments of the quality of the wine. Experts simply do not rate wines in the same way at all!
Previous articles on this blog already showed differences between the critics (here and here). To aggregate all these ratings, we need to take into account the critics’ behavior when rating the wines.
Diversity of vineyards
Wine lovers are everywhere and so are vineyards. From the very recognized “Old World” countries, such as France, Spain or Italy, to younger ones, nevertheless promising and delightful (USA, Chile, New Zealand, and the list goes on), the wine diversity is booming, as well as the number of wines! All these regions have different soils, grape varieties, different ways to grow vineyards and to produce wines.
To assess wines from a given area, it is needed to sense all the slight nuances contained in a glass of wine. Unfortunately, life being too short and a day lasting only 24 hours, it is impossible for a critic to taste and rate absolutely all the wines all over the world by being a specialist for all the regions. Usually, one region has its own specialists’ pool in assessing wines.
How could someone compare ALL the wines rated by a BUNCH of different critics summarized in one score?
You guessed it. That’s exactly what we are achieving with The Global Wine Score
In order to do that our team of mathematicians, data scientists and above all, wine lovers ( www.scorelab.io ), developed a complex proprietary algorithm to solve these issues.
Normalization process
The aim is to normalize this heterogeneous data coming from the different critics. For a given level of quality, we assess the rating each critics would give.
Let’s consider 2 experts with different rating habits :
Critic A is strict and give very few good ratings, while Critic B is more generous. In order to summarize their different points of view, we analyse the Cumulative Distribution Function (CDF) of their ratings.
For non-mathematician readers, let’s be clear : the CDF of an expert represents how many wines, in %, he has rated below a given level. For example, the strict expert would rate 80% of the wines below 89, the nicer one below 94.
The figure below illustrates this principle, where Critic A is stricter than Critic B, and we’d want to know how Critic A would rate “if he was as nice as B”.
Step 1: The meta-journalist
With this normalization method, we can then perform a projection from any experts onto another. We’d optimally find a “referent” expert, but since the Referent, Robert Parker, retired, such an ideal does not exist.
Our algorithm overcomes this problem by constructing a fictional “referent” journalist — the meta-journalist — who has scored all wines based on the real journalists ratings.
Step 2: From raw experts data to normalized ratings
Then, all the ratings from the wine experts are converted on the meta-journalist scale (100-point scale, the most common wine scale used globally) by following the normalization process explained earlier.
Step 3: Average
Once all ratings are put on the same scale, each wine gets its Global Wine Score from an average of all the ratings normalized without weighing any journalist.
There is a minimum of 3 ratings needed to generate a Global Wine Score. We remove those with 1 or 2 ratings to avoid high variancy on ratings, and to give a trusty information, allowing safe comparisons between 2 items through GWS.
Step 4: Taste the wine!
After all this process, the last but not least step is to taste the wine, guess the score and see on www.globalwinescore.com how close your guess is to the Global Wine Score, have fun!