How many deaths are caused by smoking? Bad science at its worst.
The official answer to this question is given in a document prepared for the Commonwealth Department of Community Services and Health: "The quantification of drug caused [sic] morbidity and mortality in Australia", 1988, by Holman, Armstrong, Arias et al. Its authors calculated the number of deaths due to smoking by comparing death-rates for a variety of diseases and other conditions in smokers and non-smokers. To do so they drew heavily on the work of E.C.Hammond. He had followed the progress of a large group of men and women over a period of years. He used 68,116 volunteers from the American Cancer Society to recruit what was to become over a million trial subjects. The dangers inherent in such a gargantuan trial soon became apparent when J.Berkson (Proceedings of the Staff Meetings of the Mayo Clinic, 30, 319) showed that their enormous sample was far from representative of the American population. Hammond was later to admit that his volunteers had to some unknown extent been selecting, against his instructions, sick smokers, so that his aim of comparing the histories of healthy smokers and healthy non-smokers must necessarily have been thwarted. (Journal of the American Medical Association, 166, 1159) One of the largest medical trials ever conducted had become one of the worst ever conducted. Indeed, so clearly did it show how not to conduct a scientific trial that Wallis and Roberts (1957) used it as an example (Example 287) in their book "Statistics: A New Approach".
Holman et al make no reference to Berkson's demonstration or Hammond's admission. Nor do they refer to Wallis and Roberts' use of Hammond as literally a textbook example of bad science.
Installed 5 December 2000.