An Evaluation of the Statistical Techniques used in the 1995-1996 editions of the Korean Journal of Clinical Pathology

Park, Ile Kyu; Kang, Jung Oak; Kim, Think You; Keum, Dong Geuk

Abstract

BACKGROUND
Quantitative experiment and analysis of the result with statistical techniques are an essential part of the medical article for acquiring objective confidence. But errors on application, calculation, and interpretation of statistics and insufficient explanation of the used statistical technique deprive the reader of reliance on the article. We identified the statistical errors that were commonly encountered and which researchers and readers should recognize in the Korean Journal of Clinical Pathology (KJCP) in order to improve the quality of the statistics in the article.
METHODS
We identified the frequency of the statistical errors from the 193 articles in the 1995 - 1996 editions of the KJCP. There were seven kinds of statistical errors that were most frequently observed. Each different kind of error in the same article were counted separately, but two or more of the same kind of error in an article were counted as one.
RESULTS
Seventy-five statistical errors were identified. Seven kind of the most common errors and the observed numbers of the every kind of error were as follow : 1. Using P value without the statistical name (12), 2. Performing t test instead of the analysis of variance (ANOVA) test in comparing the means of three or more groups (11), 3. Omitting the F test on the unpaired t test in small different sized samples (10), 4. Mentioning the statistical technique without using it (9), 5. Mentioning the predictive value without prevalence (7), 6. Not performing multiple comparison after the significant ANOVA test (5), 7. Not using the P value with the correlation coefficient (r) (5).
CONCLUSIONS
There were 75 statistical errors in the 1995-1996 editions of the KJCP. Not mentioning the name of the statistical technique used was the most frequently observed error. The authors' careful application of the basic statistics would be the real solution of the problem.