Misleading statistics: errors in textbooks, software and manuals.

Statistics is an essential tool in medicine in general and in medical research in particular. This is shown by the large number of articles on statistical topics carried by medical journals, by the demand for consultancy from medical statisticians and by the number of textbooks in medical statistics, especially introductory texts, now on the market. Yet, despite this importance and wealth of advice, the statistical component of much medical research is very badly done." Statistics is a difficult subject. Few people, including statisticians, find it easy at first acquaintance. Modern medical education usually contains only a short introduction to statistics, and very little about scientific thinking and research methods in general. Thus the medical researcher has at best acquired only a limited knowledge of statistical methods from formal education. There are few postgraduate courses available and many clinicians do not have the time to attend them anyway. Researchers are reduced to learning the subject from books or to relying on the statistical software to do their thinking for them. Readers of the UE are likely to be asked for thenstatistical help and advice by other medical researchers, or asked to recommend suitable books and microcomputer software. Unfortunately, many texts in statistics aimed at doctors are bad books, filled with errors, and conveying no understanding on which to build. Statistical software, too, may hold traps for the unwary and the associated manuals may be highly misleading.