Guinness. As a beer, its rich, dark color is instantly recognizable. As a company, Guinness transformed its product into an international success story…by using statistical analysis.
Grab a pint and take a look:
Prior to the late nineteenth century, brewing was no exact science. If it looked good, smelled good, and had the right consistency, it was fine to drink. Brewing was not a large industrial enterprise, rather public houses were often tied to one brewer’s product and the beers were fairly local.
As late as the 1860s there were only a handful of registered brewing companies and the five or six on the London Stock Exchange rarely got attention (Payne, 1967). Yet there was such a transformation in the late 1800s that, by 1905, Guinness had become the 10th largest British company.
The economic environment for brewers in the late nineteenth century certainly helped spur Guinness’ success as an industrial brewer: there was a steadily rising demand for beer in Britain, with the period from 1874 – 1914 experiencing peak consumption (Dingle, 1972). The price of a pint of beer remained constant during this period, at a steady 2 and 1/2 d. (old pence), cheap enough for the working classes to imbibe on a regular basis. At the same time, real wages were rising. The increase in purchasing power allowed beer consumption (almost all of which was done in public houses) to rise as well. For Guinness, this meant, that while their prices remained constant, their output more than doubled between 1887 and 1914 (Ziliak, 2008). Furthermore, drastic improvements to transportation infrastructure and mechanization of processes allowed for a potential transition from small micro-breweries to larger industrial-scale brewing.
However, a big problem remained for brewers: scientific processes were unrefined. The quality of the beer could not be guaranteed and its shelf life was hard to gauge. Even if breweries could ship their beer to more consumers in more places, there was a chance that, upon arrival at its destination, the beer would be undrinkable.
To solve this problem, Guinness hired experimental brewers to focus on the science of brewing. The company’s “look-touch-and-sniff approach” since their founding in 1759 was no longer good enough (Ziliak, 2008). For these new positions, Guinness required a background in science from Oxford and Cambridge. In addition to focusing on a better approach for large-scale industrial brewing, Guinness was also focused on improving the economic bottom line. Could a more precise scientific approach improve product quality and increase profits? Could costs of production be reduced?
The company was working to set itself apart from – and rise above – the competition. Enter the ‘Student’.
The ‘Student’ at Guinness
William Sealy Gosset was hired in 1899, fresh from studying math and chemistry at Oxford. A small, bespectacled man who often did his mathematical work on scraps of paper, Gosset did not like to tabulate results but preferred to start at the beginning for each calculation. When he was first hired by Guinness, he worked as an apprentice brewer but quickly moved up the ranks and by the end of his career he was named Head Brewer of Guinness. He worked at Guinness until he died suddenly in 1937.
Gosset is perhaps most famous under a different name. While performing experiments at Guinness, Gosset published several papers anonymously as Student. No one knows for sure why he kept his name hidden, but one theory is that it was at the request of Guinness to prevent competition from knowing what the company was up to. Student‘s t-test, which established the first tool for determining statistical significance for even very small samples, is perhaps Gosset’s most commonly recognized discovery. When the t-statistic is calculated for a sample, the result can be compared to the theoretically expected value for that sample size, as found in Student‘s t-table. With this discovery and many others Gosset “introduced the quantitative side of scientific brewing, and with it, a storehouse of statistical and experimental theory and tools” (Ziliak, 2008).
When he joined Guinness, he began to work with a team run by T.B. Case, Guinness’ first experimental brewer hired in 1893. His team had, among other things, been trying to determine what quantity of soft resins found in hops were the best for prolonging the life of a beer.
But how could the team determine the best ratio when samples were limited? Samples had to be small for economic reasons: farmers were unwilling to allot a sizable amount of space to new varieties of barley that might not be used. Furthermore, making a new beer took time, so it wasn’t as if they could churn out new product samples every hour of the day.
Gosset was convinced there had to be a way to make a broad inference from a small sample, whether it concerns hops, barley yield or other aspects of the brewing process. He was able to gain ground with his observations quickly. One of his first reports determined a positive correlation between the square root of the sample size and the level of statistical significance. Would this hold true for a small sample size, so necessary for their brewing experiments? Sure enough, when studying malt extract, Gosset discovered a way to calculate a t-statistic with sample size of two. He expanded this theory to a sample size of four, ten, and so on, achieving a method for quality control using small sample sizes using his t-test.
Experimental Brewing and the Bottom Line
In just a few short years, Gosset was able to extend his theories and experiments to extract correlations between hops input and the life of the beer, just as Case had wanted. And, he was able to determine superior hops based on the quantity of soft resins found within. He later did the same for barley. As Ziliak (2008) points out, “Gosset biometrically proved the commercial value of three varieties [of barley] that would eventually be grown on well over ‘five million acres’.”
These early findings stemming from work that began with two samples of malt extract at the turn of the century gave Guinness a scientific basis from which to make economic decisions about their business. The new information linking resins and hops to the lifespan of the beer enabled the company to be more confident in shipping the beer internationally. The information on the best type of barley encouraged them to invest significantly in just a few varieties to ensure quality would remain high. Through his work, Gosset proved that experiments with small samples could improve decision-making – and bottom lines – on a grand scale.
As Student, Gosset developed an important tool for statisticians everywhere, but his focus was on the applications of the tools in the brewing industry. After his death, Gosset was described by a colleague as maintaining “an intimate connexion [sic] between his statistical research and the practical problems on which he was engaged” (McMullen, 1939). Gosset remained a brewer at heart, using his findings to determine which choices were best economically for the business. A very successful venture indeed.
So. The next time you find yourself at a pub with a pint of Guinness, be sure raise a toast to Gosset. Slainte!
Dingle, A. E. “Drink and Working-Class Living Standards in Britain, 1870 – 1914.” The Economic History Review, New Series, Vol. 25, No. 4, November 1972. pp 608 – 622.
McMullen, L. “‘Student’ as a Man.” Biometrika, Vol. 30, No. 3, January 1939. pp. 205-210.
Payne, P.L. “The Emergence of the Large-Scale Company in Great Britain, 1870 – 1914.” The Economic History Review, New Series, Vol. 20, No. 3, December 1967. pp 519 – 542.
Student. “The Probable Error of a Mean.” Biometrika, Vol. 6, No. 1, March 1908. pp. 1 – 25.
Ziliak, Stephen T. “Guinnessometrics: The Economic Foundation of ‘Student’s’ t.” Journal of Economic Perspectives, Vol. 22, No.4, Fall 2008. pp. 199 – 216.
Image 1: Lee Saba
Image 2: via Wikipedia