Selection of the Data Distribution Type for the Normal Range Calculation

Abstract

An algorithm is proposed for selection of data distribution type. Distribution models used in this method are normal, log-normal and power-normal (X^P; P=-2-+3). Transformed data according to the distribution type are subjected to the iterative truucation and correction method, and normal range and skewness are calculated for each distribution type. The best fit distribution type is selected according to following criteria: (1) distribution which has small skewness enough by the skewness test (p=0.05),(2) simple distribution type, normal or log-normal distribution, and (3) distribution which has minimum skewnss. The distribution types of 33 chemistry and 8 hematology tests were analyzed by the present method, and distributions of some tests showed negative power distribution. Using computer simulation models, it is assumed that those types of distributions are obtained from mixed distribution models.