Mystery Cup: The Mystery Cup Family Tree
From the "too much time on our hands" department, a
data-grubber's view of the Mystery Cup Challenge results...
--By Jim Schulman--
While looking at fellow alt.coffee people's ratings in the great looking graphs Doug published with the
Mystery Cup results; I asked myself three questions I couldn't answer by simple eyeballing.
First and silliest: who won; who got closest to Barry and Doug's cupping scores? Yes, I know it isn't a competition, but I can't help myself.
Second: who scored them the same way I did; who shares my taste? More generally, this is asking whether the participants fall into distinct taste groups, or whether it's all just random noise and sounding off, like the people who bash wine tasters always claim.
Third: Which coffees and which taste aspects separate the sheep from the goats? Which of the coffees and tasting aspects generate the most controversy, and thus create the distinct taste groups? We're all coffee lovers here, so if one person's godshot is indeed another's person taste nightmare, it would be fun to find out why.
Is the suspense killing you? Check out the cuppers family
tree to see who your sibs and cousins are. Congratulations, Ben Alpers and Steve, you're taste brothers to Doug and Barry respectively (Doug and Barry are 1st taste cousins).
So what does this family tree mean? Basically, it shows who are most similar in their cupping scores. How did I arrive at it, and is it any good? Here are the visuals: look and make up your own mind.
Start with this
'map of each tasters
scores. It shows the tasters in rows, and the all the coffees' taste aspects as columns. Fat cyan shows a high score, fat magenta a low score, white and pastels are intermediate.
As a picture, this lacks contrast, and it's hard to detect anything. One of the reasons is that scorers are using different scales. Bernie Digman's, at the narrow extreme, run from 6 to 8.5; while Mr. Kim's, at the wide extreme, run from 1 to 8.5. The wide scorers tend to drown out the narrow ones, so contrast is lost overall. The easiest way to fix this is to scale the scores so that everyone's range and midpoint are the same. This gets us to
the scaled map.
Now there's contrast; but the map looks like a random scatter of blotches. It's still impossible to tell who's like whom. So let's play a game ... Rearrange the rows and columns to put all the magentas in one place, and cyans in another ... getting tired? This is a game for computers to play. Mine came up with this arrangement:
arranged map; it's OK, but not great. Trouble is, so far nobody's worked out a program that'll always find the best arrangement, but if your computer can do better, let me know.
How does the family tree fit in? The family tree puts people into the order found on the arranged map; the kinship lines mirror the order in which the rearrangements were made. As you can see, there are four magenta clusters along the main diagonal of the arranged map - these show the boundaries of the "tribes" marked on the family tree. A similar family tree,
the taste tree exists for the coffee tastes. It too has four tribes corresponding to the clusters.
So now you know who you're like. Want to know why? Have a look at the tasters