#OTD 2012 Genuchi Taguchi d (b 1 Jan 1924) 🇯🇵 founder of the Taguchi method for quality product improvement - reduction of process variation through robust design of experiments. His methods revolutionized manufacturing quality control practices & mindsets. 1/4
2/ He collaborated throughout the 1950s with other notable statisticians such as CR Rao Fisher Shewart & was sponsored by Tukey at Princeton
3/ his methods differed from the conventional specifications based on tolerances alone but developed the concept of quality loss (rather than just quality)
4/ His method involves 3 steps: System design (operational specifications parts, materials, assembly); Parameter design ( deciding the most cost effective methods to control variation); Tolerance design (identification of key noise factors)
• • •
Missing some Tweet in this thread? You can try to
force a refresh
In June 1905 Karl Pearson introduced the terms ‘kurtosis’, ‘leptokurtic’, ‘platykurtic’ & ‘mesokurtic’ to describe shapes of skewed frequency distributions. The paper was otherwise a huffy (& lengthy) rejoinder to some German critics of his 1899 paper on skew variation 1/4
2/ As well as highly entertaining criticisms of his foes, the paper also presents derivations & distribution data for a variety of biological data, including human skulls, crab ‘foreheads’, shell lengths, & organ weights.
3/ In 1927 WS Gosset ("Student") provided a humourous aide memoire: 'platykurtic' = platypus w short tails vs 'leptokurtic' = 2 kangaroos ‘lepping’ (I don't make this stuff up you know)
#OTD 1866 Charles Davenport b (d 18 Feb 1944)🇺🇸@amstat Fellow 1921. Director Cold Spring Harbor Labs. His 1899 text ‘Statistical Methods with Special Reference to Biological Variation’ actively promoted statistical methods pioneered by Galton & Pearson for biological research 1/7
2/ Reviews were cool (which incensed him): there were a lot of mistakes in the first edition, the amount of hand calculation was formidable, & he seemed to underestimate the difficulties for most researchers in their application.
3/ If you think Fisher, Galton, & Pearson were vile eugenicists they were nothing compared to Davenport. He founded the Eugenics Record Office ERO in 1910, & launched proposals to transform the human race (actually only part of it) by selective breeding of the picked few,
Louis-Dominique-Jules Gavarret (28 Jan 1809-30 Aug 1890) 🇫🇷was the first (1840) to apply a forerunner of confidence intervals to clinical data, & described 5 principles for rigorous clinical trial conduct. Unfortunately his book 'Principes de Statistique Médicale' 1/7 🧵
was almost forgotten by the end of the 19th c. He learned how ‘le calcul des probabilités’ could be applied during a debate in 1835 at the Académie des Sciences, where Navier (of Navier-Stokes equation fame) described how it could be applied to therapeutic research 2/7
& destroyed his debate opponent). Using Louis’ blood-letting data & the maths of “large numbers" learned when a student of Poisson, Gavarret calculated the range of possible errors of means (‘limites d’oscillation’) & argued against taking the point estimate at face value. 3/7
Laplace may have been the first to suggest application of ‘probability calculus’ to medicine, but so-called ‘numeric methods’ to compare efficacy of interventions were already in use. Pierre Charles-Alexandre Louis (1787–1872)🇫🇷 assessed early vs delayed blood-letting 1/4
on mortality rates in pneumonia, demonstrating (1828) that early blood-letting killed patients at a higher rate than bleeding later (of course blood letting was never a good idea, even though it was "scientifically justified" & backed up with vociferous opinions) 2/4
Louis argued that medical progress could not be made if cases were assessed individually. He promoted the importance of ‘population’ comparisons, and averaging cases based on similarities in age, sex, diet, disease severity, & onset.