Support and Illumination
- PDF / 434,423 Bytes
- 6 Pages / 504 x 719.759 pts Page_size
- 59 Downloads / 172 Views
0092-86 15/97 Copyright 0 1997 Drug Information Association Inc.
SUPPORT AND ILLUMINATION DAVIDJ. FINNEY Consultant Biomeuician, Edinburgh, United Kingdom
Those who teach statistics often fail to transmit understanding of vital parts of their commonly used technical vocabulary, especially that relating to the nature and objectives of various concepts. Consequences for choice of methods and for logical interpretation of analyses can be serious. %day, many biologists and other scientists use statistical procedures despite being themselves almost without training in the discipline; some may even deliberately have avoided exposure to the study of what they mistakenly perceived as requiring great mathematical talent. Luck of understanding of principles leaves many a biologist prey to those who suggest rhai artracrively packaged commercial software will meet all his needs! Key Words: Statistics; Software; Terminology
SUPPORT
needed correction to the fallacy: “Post hoc, ergo propter hoc,” the uncritical assumption that calculated numerical differences, however small, between mean measurements for differently treated sets of animals have been CAUSED BY the treatment. The tests brought rigor into the thinking of those who failed to realize that a difference found in a particular experiment may have arisen from the natural variability between individuals that is always present in measurements on living material. In a very simple experiment, suppose two sets of subjects to be selected at random from a population: assign the first set to treatment A, the second to treatment B, and after an agreed period measure each subject. The measurements may then receive any standard statistical analysis. A statistical significance test merely assesses the plausibility of a null hypothesis that, in respect of this measurement, members of the parent population are not affected differentially by whether their treatment was A or B. Only by extraordinary coincidence would mean measurements calculated from the two sets be exactly equal; the test merely inquires into the probability that a difference at least
MOST STATISTICIANS KNOW the saying: “Too many people use statistics as a drunken man uses a lamppost, for support but not for illumination.” To some extent, this state of affairs can be a consequence of misuses of language. This paper illustrates some hazards. Its purpose is to improve scientific understanding, rather than directly to aid new drug applications. Excessive dependence upon significance testing, misuse of undoubtedly clever software, and uncritical adoption of practices designed to please journal editors or regulatory authorities whose statistical understanding may suffer from the same deficiencies is especially referred to.
STATISTICAL SIGNIFICANCE When first introduced, the concept of a formal test of statistical significance was a Presented at the DIA Workshop “Statistical Methodology in Clinical Research and Development,” April 1517. 1996, Copenhagen. Reprint address: David J. Finney. Consultant Biometrician, 13 Oswal
Data Loading...
