The BA Festival of Science: Happy Birthday to the t-test
Posted by Katrina Charles, BA Media Fellow
Right, now that all the fuss of the LHC is over [far from it - Ed.], we can get back to interesting science – statistics.
One hundred years ago this year, “Student” (William Sealy Gosset) published his new statistical hypothesis test for looking at whether differences in small data sets were due to chance, or were actual differences. To honour this landmark centenary, the President of the BA Mathematical Sciences section, Stephen Senn, gave what was a very amusing (I’m talking laugh out loud funny) presentation about the history of the t-test. Of course there was some maths in there too, but after a long day it was easy to glaze over for those bits.
The t-test is so widely used today that film studies and Jewish studies are the only academic disciplines in which Senn has yet to find examples of its usage, and he is continuing to search those disciplines. He has found examples, from searches in JSTOR, in a number of papers on law, music and religion as well as the usual sciences.
But it didn’t happen overnight. When the test was originally published it took 16 years before anyone noticed that the data set that Gosset used was wrong. Gosset had to publish under a pen name because his employer, the brewer Guinness, who had sent him on sabbatical to study with Karl Pearson at University College London, didn’t want the brewing industry to know that they were using this statistical method. They also didn't want him to use their secret brewing data, so instead he used data on the effects of the two optical isomers of the drug hyoscine on the amount of sleep that inmates of an insane asylum were getting, published by Cushny & Peebles in 1905.
The data in the paper showed no difference between two isomers, but in Gosset’s paper he used his new test to show a significant difference. Gosset’s work was included in Ronald A. Fisher’s Statistical Methods for Research Workers, first published in 1925. A mistake in Gosset's work was subsequently spotted - he had erroneously compared a control to an isomer, so Fisher removed the names but kept the numbers, as an exemplar of the t-test, in later editions. And the rest, as they say, is history...
