I am doing cpk analysis and did continuous fit-->fit all to find the true distribution for the data, then run cpk. JMP provided the best option, then I did Goodness of fit to confirm by P-value, however, P <0.001 which reject the distribution JMP suggested. I assume, this means JMP do not have the proper distribution stored. My question is if I can just use the distribution to run CPK, based on the lowest AICc in JMP? Or what will be the option here for me to get correct CPK? Thanks.
Thanks for the histogram and the box plots. The quantile plot that I was asking for would help understand why the best model still fails the goodness of fit test because it plots the data with axes that have been transformed to linearize the CDF and the data markers. The human visual system is very good at detecting location and linear patterns. Here is an example using the :weight variable in the Big Class data table. The Distribution platform selected the Lognormal distribution model based on AICc and then I produced both the QQ and PP plots. (The first plot is quantiles and the second is probabilities, both from the data versus from the model.) The straight line (reference) is the CDF for the Lognormal distribution. Inspect the markers against this reference line for 'agreement" / "disagreement" or "goodness of fit."
What does your data look like in these plots versus the selected (and fitted) SHASH distribution model?