Home > iGrafx Process for Six Sigma Methodology and Reference > Fit Data Reference
|
This functionality is only available in Process for Six Sigma and above. |
MINITAB computes an Anderson-Darling goodness of fit statistic for the Normal, Weibull, LogNormal and Exponential distributions. For more on the Anderson-Darling test see the Minitab StatGuide™. In general, the lower this value is the better the distribution fits the data, but the interpretation of the value is distribution dependent -determining whether a particular value for this statistic is significant at a given confidence level depends upon the distribution as well as the number of data values. Because of this you should rely on your visual interpretation of the curves as well as the statistic in making your judgment about which distribution to use. Experience with the statistic suggests that for large values of N, a few hundred data points or so, the agreement between the statistic and visual interpretation is quite good. However, you should think of the Anderson-Darling statistics displayed in the dialog as a clue, not an answer.
JMP computes different goodness of fit measures which are as follows:
Normal Distribution - Shapiro-Wilk Test (if n < 2000) or Kolmogorov-Smirnoff_lillifors Test (n >= 2000)
Weibull Distribution - Cramer-von Mises W Test
LogNormal Distribution - Kolmogorov's D Test
Exponential Distribution - Kolmogorov's D Test
iGrafx Process for Six Sigma supports the following distributions for Fit Data:
Uniform
Normal
Exponential
LogNormal
Weibull
To choose the best distribution, use the Plot check boxes and click a graph to show:
Probability Density Function (PDF)–The graph of frequency (how often the number will be this value) on the y axis and the value itself on the x axis.
Cumulative Distribution Function (CDF)–This is the integral of the PDF. It shows the probability that a random value from the distribution will be less than the value on the x-axis. The y-axis ranges from 0 to 1.
Residuals–The difference between the probabilities calculated by the CDF for the distribution and the probabilities computed from the measured values.
Related Topics