### Introduction

### Common Mistakes in Statistical Methods

*Korean Journal of Anesthesiology*(KJA), provide independent instructions to authors, referring to the Uniform Requirements. Nevertheless, statistical errors are strikingly common in medical articles; Altman and Bland [2] estimated that more than 50% of medical reports published at that time included statistical errors. On the other hand, an analysis of 164 articles published in British psychiatry journals showed that 40% of articles included statistical errors [3]. Articles published in Korean journals are not very different. Ko et al. [4] analyzed KJA articles from Vol. 1 in 1981 to Vol. 6 in 1990, and reported that statistical errors were included initially in 97% of the articles and in about 67% of the articles published later. Ahn [5] analyzed KJA articles published in five years starting in 1994 and reported that 60% of the articles included various types of errors.

*Circulation*and reported that the most common statistical error is the inappropriate use of a t-test for a multi-group hypothesis test. This result is consistent with another report which found that the most common statistical error occurred with data to which an ANOVA or paired t-test should be applied but which were tested with Student's t-test [7]. According to work by Olsen [8], of the approximately 141 articles published in the journal

*Infection and Immunity*, 54% included statistical errors. These were found in the detailed statistical methods applied (20%), in the descriptions of the statistical results (22%), or in both (12%). An assessment of the articles published in the Journal of the Korean Medical Association in the 1980s showed that 97.8% of the analyzed articles included one or more statistical errors, including insufficient descriptions of the statistical power of the test method and confidence interval (91.9%), duplicated testing due to the incorrect statistical method applied (65.2%), insufficient descriptions of the statistical method itself (58.2%), and deductions from unreasonable statistical conclusions (52.2%) [9]. In addition, a parametric test method was often applied to a variable for which normality is doubted, or analytical results were omitted without an appropriate explanation despite being described in the methods section. Most of these statistical errors were commonly observed in articles published in the KJA.

### Tips for Avoiding Negative Reviewer Comments on Statistics

### Consulting about statistics should start with the planning and design of the study

### What type of study should I conduct?

### Sample size calculation errors and ethics

### Errors associated with study methods and the applied analytical method

*The data were approximately and normally distributed and thus did not violate the assumptions of the t-test*."

*The number of subjects was small and normality was found to be contrary to the normal distribution test results. A nonparametric test was performed using the Wilcoxon rank-sum test.*"

### Presentations & Interpretations of Results

*British Medical Journal*,

*The Journal of the American Medical Association*, and the

*New England Journal of Medicine*showed that the statistical results described in the abstracts were different from or were not mentioned in the main text in 18 to 68% of the reviewed articles. Because most readers judge the results and values of studies through abstracts before reading the full-text version, this review result may not be regarded as a mere mistake. It is herein emphasized that correctly describing the results is as important as appropriately performing the statistical analysis. When two or more analytical methods are applied, detailed descriptions should be provided about the data set applied to each of the analytical methods. It is not enough simply to say "

*where appropriate*."