Statistical methods in anesthesia articles: an evaluation of two American journals during two six-month periods

Simple criteria were used to evaluate the statistical analyses in 243 articles from two American anesthesia journals published in the latter six months of 1981 and 1983. Eighty-two percent of the articles reported the use of control measures and 37% reported randomization of treatment, where they were possible. Data were classified as nominal, ordinal, or interval; as independent or related samples; as two-sample or more-than-two-sample cases. The descriptive, inferential, and correlative tests used were evaluated for appropriate application and primary errors were identified. Nine percent of the 722 descriptive statistics had major errors, most of which were a description of ordinal data as though they were interval. The incidence of erroneous applications of 394 inferential statistical tests was 78%. Nearly three-quarters of the 308 primary inferential statistical errors involved either use of a test for independent samples on related data (and vice versa) or multiple applications of an uncorrected test to the same data. Only 4% of the 113 statistics of association were considered erroneous, most because the method was not identified. No differences were detected in the incidence of errors in either experimental design or statistical analysis across time or across the two anesthesia journals. Fifteen percent of the 243 articles in both journals at both times were without major errors in statistical analysis. Recognition of potential sources of error should make it easier for investigators to use experimental designs and statistical analyses appropriate to their needs.