Methods of data analysis in the emergency medicine literature

Abstract

The authors hypothesized that data analysis in the current emergency medicine literature uses relatively few methods and sought to determine the frequency distributions of each method of analysis. The authors defined their population as original contributions in three refereed emergency medicine journals from September, 1985 through July, 1989. Letters to the editor, brief reports, reviews, and case reports were excluded. The authors reviewed 250 randomly selected articles and identified the method(s) of data analysis in each. The absolute frequency distribution of statistics were as follows: descriptive statistics only, 31%; contingency tables, 35% (chi 2, 28.4%; Fisher's exact test, 13.2%; McNemar's test, 0.4%); Student's t-test, 34%; ANOVA/ANCOVA, 12%; regression techniques, 8% (simple linear regression, 4.0%; multiple regression, 3.6%; logistic regression, 1.6%); nonparametric tests, 7% (Mann-Whitney, 2.8%; Wilcoxon, 2.4%; Dunnett, 0.8%; Kolmogorv-Smirnov, 0.4%; Kruskal-Wallis, 0.4%); multiple comparisons, 6% (Scheffé, 4.4%; Newman-Keuls, 2.0%); correlation techniques, 4% (Pearson product-moment correlation coefficient, 2.8%; Kendall's tau, 0.8%; Spearman's rho, 0.4%); confidence intervals, 2%. Correction techniques were used in 9% (Dunn-Bonnferoni, 4.8%; Yates correction, 4.4%). No statistics were found in 2% of the articles reviewed. Five statistical methods account for the vast majority (97% cumulative) of statistical uses in emergency medicine literature. This information should prove useful in deciding which tests should be emphasized in educating emergency physicians.