When goodness-of-fit values are displayed, as in the distribution fitting comparison chart, p-values are displayed for some combinations of ranking methods and fitted distributions. These express the degree to which the actual fit conforms to a theoretical fit for that fitting test and distribution (see Goodness Of Fit for more information). When the Chi-square method is used, p-values are displayed for all continuous and discrete distributions. P-values are also displayed for the following continuous distributions when the Anderson-Darling or Kolmogorov-Smirnov methods are used: normal, exponential, logistic, maximum extreme, minimum extreme, uniform, gamma, Weibull, and lognormal. P-values for the other distributions are under development.
Since p-values for Anderson-Darling and Kolmogorov-Smirnov statistics are influenced by the number of data points being fitted, an adjustment formula is used to arrive at the asymptotic Anderson-Darling and Kolmogorov-Smirnov statistic for a given sample size. The quality of fitted parameters and the calculated p-value deteriorates as the sample size decreases. Currently, Crystal Ball needs at least 15 data points for fitting all the distributions.
To run fittings on multiple data sets, use the Batch Fit tool.