Abstract
A manual calculation method of the normal range from the data which contain a certain number of abnormal data was presented. The data are summarized in a frequency table, and the classes with the frequency less than one fifth of the maximum frequency are eliminated (initial truncation). Cumulative sum of the remaining classes is calculated (truncated cumulative sum), and 24.4 percentile (X24) and 75.6 percentile (X76) arecalculated from the truncated cumuiative sum. The lower limit (RL) and the upper limit RUof the normal range are calculated as follows: d=x76-x24, RL=x24-d, RU=x76+d. Classes with the class mark less than RL and those with the class mark larger than RU are eliminated from the original frequency table, and the remaining classes are subjected to repeated calculations as described above, until RL and RU give constant values. This method was extended to the log-normal distribution processing without log calculation. The only difference from the normal distribution processing as described above is RL and RU calculation, which are calculated as follows: r=x76/x24, RL=x24/r, RU=x76 r. The calculated range was less affected by contamination of abnormal data up to 20 per cent.
