[Data transformation and normalization].

When we analyze measured values by statistical techniques, we usually use a parametric method. It is necessary for measured values to show a normal distribution. Therefore, we must confirm that the distribution is normal, whereby a histogram shows that the distribution of data points is symmetrical above and below the mean. When measured values do not show a normal distribution, power transformation (square root transformation, logarithmic transformation) must be performed. We can evaluate the presence of normality by three methods: viewing a histogram, through skewness and kurtosis values, and Kolmogorov-Smirnov method's p-value. Because Kolmogorov-Smirnov's method is influenced by outliers, attention is necessary regarding the interpretation of p-values. For example, if we calculate reference intervals from clinical testing data, we calculate a parameter(mean and standard deviation) and set X +/- 2SD to upper and lower limits, respectively. When we evaluate the reference intervals, a range including 95% of the central part of sample is important. Identification of the distribution type based on the diagonal linear pattern of normal quantile plots may be the most reliable in my experience.