How long should you continue identifying the distribution of your visual data?
Yukio Takahashi, President, BioStat Research Institute, Inc.
Identifying the distribution to which the obtained data belongs is the first step in statistical analysis. Many methods have been provided to visually check the distribution of data using JMP's "Univariate Distribution". Traditionally, the main focus was on judging whether a normal distribution fit was possible. JMP's "Univariate Distribution" also supported fitting probability density functions to histograms for some time. In recent versions, when "All" is selected in "Fit Continuous Distribution", many distributions are listed in order of the magnitude of the -2 log likelihood and the AICc (Akaike Information Criterion Corrected), and it is possible to judge the distribution fit using AICc. It was meaningful to evaluate the ranking of distribution fits using AICc for various data. Since the log likelihood is output, it seems that the maximum likelihood method is used, so I used Excel to retest the calculation method of JMP. When considering what distribution fit is appropriate using AICc, it seems that this is a good opportunity to move away from the traditional ambiguous visual judgment. A similar function is available for the "univariate life span" model, so I would like to discuss fitting a new distribution to it as well.
