AN ASSESSMENT OF THE STATISTICAL PROCEDURES USED IN ORIGINAL PAPERS PUBLISHED IN THE SAMJ DURING 1992

Objective. To assess the statistical procedures used in original paper s published in the SAMJ. Design. Descriptive study based on a random s ample of 100 papers from the 153 papers with methodological content th at were published in the SAMJ during 1992. Results. This review showed that 34% (95% CI (25%; 43%)) of papers used no statistical procedure at all or used simple descriptive statistics only. In sampling methods , there was a predominance of the use of the period sampling method as opposed to probability sampling methods, Inappropriate statistical me thods were used in 15% (6%; 24%) of papers, while in 16% (9%; 23%) sta tistical procedures and in 13% (6%; 20%) the sampling methods used cou ld not be identified. Inaccurate graphical methods were used in 17% (6 %; 28%) of papers, Confidence intervals and power calculations are use d far too infrequently, in 33% (19%; 47%) and 11% (3%; 19%) of appropr iate papers respectively. If the Journal's readers are at least famili ar with simple descriptive statistics, contingency table analysis, sim ple epidemiological statistics, t-test procedure and confidence interv al calculation and interpretation, they will have a complete understan ding of the statistical content of 60% of original articles published in the Journal. Conclusion. Guidelines for the statistical treatment o f reported data and the statistical review of-articles before publicat ion will assist substantially in improving the quality of statistical analysis. More intensive use of available biostatistical and epidemiol ogical expertise at the study design and analysis stages is needed to shift the emphasis from descriptive research to analytical investigati on.