Hi everyone! I analyze the statistics that arise during the production and packaging of tablets (pills). Data analysis is performed in several steps. Detecting outliers (Grubb's test) Determining, if the distribution is normal (Shapiro-Wilk test) Tolerant intervals Plotting the Levey Jennings graph, PP/PPK Turn on the test for the first 3 special reasons Sometimes there are errors in entering data, sometimes the problem is in the methodology. Of course, this affects the distribution. We find the reasons. But sometimes the data is not normal, in principle. It can be not symmetrical. One time it hided mixture of 2 normal distributions and so on... In JMP, as I understand, your can check, what distribution is closer to yours (FIT ALL). (1) The standard calculation PP/PPK is not suitable for an abnormal distribution, is it? Ok, we recalculate PPK using percentile method (it uses median, not average (by JMP) (2) But what about graph. Which chart should I use, to receive UCL & LCL for Non normal data? I came to the conclusion that and graph must calculate UCL & LCL also from median. I use Levey Jennings with fixed average equal to the median. Sometimes median and average are very close, and the difference between borders is almost not felt. But there are also differences. (3) Or, is it enough to recalculate the PPK and plot usual Levey Jennings chart using average and use testing for special causes? (4) Or maybe non normal data can't be tested in such way? *P.S. XBar, IMR I don't use for today, for this year. We are at the beginning of applying statistical methods. Сauses that are not visible in the routine process are revealed and eliminated. Please help me answer the questions in bold. (1-4) Please give a rating to my method. Offer your work algorithm for Normal data & Non normal Thank you in advance!!!
... View more