Earlier this month I attended the 5th European Design-of-Experiment User Meeting in Cambridge, England, which, considering the topic being statistical design of experiments, turned out not to be as dull as one might think. All the credit for the pizzazz goes to our colleagues across the Atlantic—PRISMTC; in particular Paul Nelson and Andrew Macpherson. They conjured up an in-conference experiment that developed an optimal blend of three local beers (all made by Milton Brewery and sold in bulk by Polypins (36 pints) from £56, Firkins (2 polypins) from £84), a pale ale called Cyclops (30-80%), a bitter under the label Justinian (20-70%) and a dark mild named Medusa (0-50%). Prior testing by these two boffins of stats and zymurgy (that is, the study of yeasty concoctions) led them to constrain the ranges of the three brews to the ranges shown in parentheses.
Paul and Andrew laid out a clever design that, via balanced incomplete blocking, restricted any one taster to only 4 blends, while testing enough combinations often enough to provide adequate power for discerning just the right formula. The fun bit was them asking us conference-goers to provide the necessary data prior to an atmospheric dinner at Magdalene (for some reason pronounced in English as “maudlin”) College.
This limitation on beer was one departure from a similar mixture experiment on beers that I ran* with my two sons and son-in-law as the tasters (little chance them going along with such a sensible restriction). The other wrinkle was them requiring all of us to taste a strip of paper that ferreted out about a third of the tasters being “super tasters”—those who immediately recoiled from the bitter taste (many thought it just tasted like paper).**
It turned out that the bitterest blend, in contrast to the mildest of the beer mixtures, was not greatly liked. I think this must be an acquired taste! You can see this on the triangular, 3D response surface graph of the predicted response—the lowest corner being the B:Bitter. Surprisingly, mixing in some A:Ale makes a relatively tasty brew—these two beers synergize, that is, provide much better results than either one alone. But the tastiest blend of all is the peak at the C:Mild corner, with 30% of Cyclops, 20% of Justinian and 50% of the Medusa, some blends on a ridge through the middle of the triangular mixture space look promising.
Three cheers for three beers and hats off to the brilliance of Paul and Andrew of PRISMTC for pulling off this fun, clever and informative taste test. See their full, illustrative report here.
*See Mixture Design Brews Up New Beer Cocktail—Black & Blue Moon
**Check out this BBC report and short video on testing for super tasters