Beer and Statistics

Beer and Statistics are partially confounded.   Why?

Reason 1

Every student of statistics learns about the t-distribution. W.S. Gossett (1876-1937) discovered that distribution as part of his 38-year career at Guinness, at which his last position was Head Brewer. Gossett published his statistical papers using a pseudonym "Student" to conform with company policy on intellectual property--though what Student invented is not what students of introductory statistics typically learn. 

Gossett, trained in chemistry and mathematics, asked and answered many useful questions about the economics of brewing, guided by data analysis and experimentation.  He sought help from the principal statisticians of his time, including Karl Pearson and R.A. Fisher.  Nevertheless, he knew enough to keep his wits about him and keep his focus on problems of prediction and brewing economics without being ensnared in Pearson-Fisher philosophical controversies.

Stephen Ziliak provides an insightful summary of important aspects of Gossett's work in "Guinessometrics: The Economic Foundation of 'Student's' t" (2008), published in the Journal of Economic Perspectives and available here.

Reason 2

George Box sponsored a Beer and Statistics seminar for 30 years at the University of Wisconsin, starting in 1961.   Monday nights during the academic year, George invited scientists or engineers to present interesting problems that faculty and students could then discuss, led by George.   Refreshments included beer, to help make the sessions less formal than the typical university seminar.  George was a masterful consultant--he liked to help scientists and engineers make progress using statistical methods and his command of relevant statistical theory and methods moved the conversation along briskly. I learned a lot about statistical consulting by attending those sessions.

Here's part of the advertising copy for the seminar from 1980:

George Box’s Monday Night Beer and Statistics Symposium, an informal gathering, serves two purposes.  First, it allows investigators an opportunity to discuss and seek advice about the statistical aspects of their problems.  Second, it gives advanced graduate students in statistics further experience in statistical consulting; the method is learning by example. (Thanks to Dr. Conrad Fung for digging this information out and sharing with me.)

In 1980 and 1981, I had the job to find guest presenters for the Beer and Statistics seminar.  Gossett would have been a perfect guest--he had scientific and commercial problems that were challenging and he appreciated the power of proper analysis and experimentation that George could describe. 

Gossett's statistical approach to brewing, coupled with more modern methods of inference and data display could benefit the new generation of brewers in North America.  I'll report on any progress I make in building that bridge.