An assessment of the statistical procedures used in original papers published in the SAMJ during 1992

Abstract

Objective: To assess the statistical procedures used in original papers published in the SAMJ.

Design: Descriptive study based on a random sample of 100 papers from the 153 papers with methodological content that 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 methods were used in 15% (6%; 24%) of papers, while in 16% (9%; 23%) statistical procedures and in 13% (6%; 20%) the sampling methods used could not be identified. Inaccurate graphical methods were used in 17% (6%; 28%) of papers. Confidence intervals and power calculations are used far too infrequently, in 33% (19%; 47%) and 11% (3%; 19%) of appropriate papers respectively. If the Journal's readers are at least familiar with simple descriptive statistics, contingency table analysis, simple epidemiological statistics, t-test procedure and confidence interval calculation and interpretation, they will have a complete understanding of the statistical content of 60% of original articles published in the Journal.

Conclusion: Guidelines for the statistical treatment of reported data and the statistical review of articles before publication will assist substantially in improving the quality of statistical analysis. More intensive use of available biostatistical and epidemiological expertise at the study design and analysis stages is needed to shift the emphasis from descriptive research to analytical investigation.