Beerometrics: Econometrics and the Science of Beer

UPDATE: I got my Ziliak's confused.  Corrected below.  My Apologies to the right one. 

We all know beer saved the world, but now we know that it also contributed significantly to modern econometrics (HT: Freakonomics).


Stephen Ziliak from Roosevelt University [not James Ziliak (onetime professor of economics at the University of Oregon)] has a great paper out about the contributions of William Sealy Gosset, a brewer at Guiness, who write under the pseudonym of Student and made major contributions to statistics. 

Here is an extended and fascinating quote (footnotes omitted):

A general solution to the problem of random error in small sample analysis was given in 1908, by Student. “Student” is the pen name of William Sealy Gosset (1876–1937), an Oxford-trained chemist and experimental scientist who worked his entire adult life as a brewer and business man for the Guinness Brewery, Dublin (1899 to 1937) and Park Royal (1935 to 1937). Student was experimenting on three of the chief inputs to Guinness stout – barley, malt, and hops – when he made the discovery leading to what scientists now call Student’s t-distribution, table, and test of significance.

But Student’s contribution to experimental science and the theory of errors extends far beyond Student’s t – however permanent and fundamental t is. Between 1904 and 1937, Student innovated – more than two decades before R.A. Fisher – a useful collection of experimental concepts, methods, and attitudes, which were used for doing routine work at cooperating farms and at the Guinness brewery.

As Head Experimental Brewer, a position he held from 1907 to 1935, Student’s main charge was to experimentally brew, and to gradually improve, a consistent barrel of Guinness stout, input by input, from barley breeding to malt extract, at efficient economies of scale. Pounding out more than 100 million gallons of stout in annual sales, the problem Student faced at Guinness was economically motivated and non-trivially large. While endeavoring to control product and reduce costs at the large brewery Student was consistently faced with a small number of observations on new barley to try, at n = 2, 4, or – if he was lucky – 7. In the process, he – though self-trained in statistics – managed to solve a general problem in the classical theory of errors which had eluded statisticians from Laplace to Pearson.

Less well-known is Student’s contribution to experimental design, systematically ignored by Fisher. Student found a method for maximizing the power to detect big economic differences (low Type II error) when the quantitative difference is really there to be detected. Student opposed Fisher’s randomized field experiments on grounds that, as Student proved as early as 1911, decisively so in 1923, and again in 1938, balanced designs are more precise, powerful, and efficient compared to random.

Brewers and economists alike have not noticed as much as they might that Student’s exacting theory of errors, both random and real, marked a significant advance over ambiguous reports of plant life and fermentation asserted by Priestley and Lavoisier down to Pasteur, Fisher, and Johannsen, working at the Carlsberg Laboratory in Denmark.

The experimental concepts which Student used at the brewery to revolutionize science and brewing are outlined here, basically in order of their development by Student in his job as apprentice brewer (1899–1906), Head Experimental Brewer (1907–1935), and finally Head Brewer of Guinness (1935–1937): (1) net pecuniary advantage and the purpose of the experiment; (2) profitable odds versus a fixed rule for the level of statistical significance; (3) small samples of repeated and independent experiments; (4) random error versus “real” error; and (5) the power and efficiency of “balanced” over “randomized” field experiments in economics. The balance of this article illustrates these concepts with experiments designed and/ or evaluated by Student at Guinness’s brewery.

Here is another blog post on the same paper from Wonkblog.