Statistical awareness of research workers in British anaesthesia

A survey was made of simple statistical errors in the precirculated booklets of abstracts to the Anaesthetic Research Society. In the five booklets June 1988-November 1989, only four of 19 figures included clearly labelled error bars when necessary; in four figures there was no indication of variability. In a more detailed survey of the 115 abstracts presented in 1990, errors were placed in categories: the presentation of method or choice of statistical test; variability; probability. There was a total of 115 errors in 61 abstracts, which was 65% of the 94 abstracts presenting numerical information. The most common errors were: failure to identify tests of inferential statistics (n = 29 abstracts); failure to present data to allow interpretation of P values (n = 21); misuse of SEM (n = 13). Confidence limits were given in seven abstracts. Type II (beta) error was not searched for formally, but no abstract that reported a negative result included the power of the study. With this level of elementary errors, there is considerable room for improvement in the use of straightforward guidelines to the presentation of statistical information. The problem is not confined to British anaesthesia: in more than 30% (25 of 71) of the figures contained in abstracts of the International Anesthesia Research Society in 1990, plots of variables against time showed only mean values.