Uses and misuses of common statistical techniques in current clinical biomedical research

摘要: Seventy-four articles published in seven basic biomedical journals 1973 or 1974 provided examples to illustrate current misuse less than optimal use of statistical tests. The two primary New England Journal Medicine and the American Medical Association, are weekly publications whose editorial review is stringent Examples classified three broad areas. General findings include both overuse underuse tests, failure state methods analysis, P-values calculated without setting alpha (a) level misinterpretation null hypotheses, misdefinition confidence interval. Major specific areas discussed (1) chi square testing (2) analysis variance where indicated, (3) non­ utilization nonparametrie tests criteria for parametric fail, (4) paired data, (5) linear regression correlation analyze same data when alone sufficient. Less frequent mistakes were found measures used estimate central tendency, choice one-tailed two-tailed misapplica­ tion coefficients, survey sampling techniques which pro­ duce significant bias, utilize sequential analysis. Many only t-tests unpaired evaluate results other would seem more useful.