Roosevelt University’s Stephen Ziliak has uncovered beer’s very important contribution to the world of statistics research. As he details in this month’s Journal of Wine Economics, Guinness brewmaster W.S. Gossett published in 1904 one of the first papers on statistical significance. It explored how many trial brews Guinness needed to ensure it produced pints consumers would swallow:
Guinness malt was produced in Gosset’s time primarily from Irish and English barley stock — Old Irish, Prentice, Plumage Archer, and Spratt Archer were effective varieties. Malt extract was measured by “degrees saccharine” per barrel. An extract in the neighborhood of 133 saccharine gave the targeted level of alcohol for Guinness’s beer. A much higher degree of saccharine would affect the stability and life of the beer, but it also increases alcohol content — which in turn increases the excise tax which Guinness owes to the British government, which — sad to say — ups the price of Dad’s pint.
If, on the other hand, the alcohol content comes in too low, if the degree of saccharine is insufﬁcient, customers would riot, or switch to Beamish and Beck’s. In Gosset’s view, .5 degrees saccharine was a difference or error in malt extract which Guinness and its customers could swallow. “It might be maintained,” he said, “that malt extract “should be [estimated] within .5 of the true result with a probability of 10 to 1”. Using the mean differences of saccharine values of extracts, between the Main and Experimental breweries, Gosset calculated the odds of observing the stipulated accuracy for small and then large numbers of extracts. [He] concluded, “In order to get the accuracy we require [that is, 10 to 1odds with .5 accuracy], we must, therefore, take the mean of [at least] four determinations.”
There you have it — statistical significance, in a pint.