I am such a novice in JMP and statistics. I have basic questions for experts here.
The data I have is product dimension. I have ‘min’ dimension data, and ‘max’ dimension data around 30 points for each. And with these data, I need to propose to team min and max dimension specification.
In JMP11, I used Analyze>Distrubution and found most of the data are slightly skewed, and there are some with a few outlier points. I checked continuous fit – normal – fitted normal - goodness of fit: which shows Prob<W less than 0.05.
Traditionally I was told that we just take mean/average then +- 3 STDEV, which I don’t find it appropriate in this case.
If I could be pointed out proper analysis of my data, and how I should be setting specification, or JMP article I should read, I would appreciate very much.
Nan, an alternate approach is to use the data to create a control chart for each potential specification item. The published specification [e.g. to an internal or external customer] is then based on +/- 3 Sigma [Hence 6 Sigma quality].
Update the chart as new data becomes available and use the charts to not only set specifications but to, more significantly, control the process.
Nan, I like to use tolerance intervals for helping to set spec limits. The online (NIST/SEMATECH e-Handbook of Statistical Methods) engineering stats book has a section (end of chapter 7) on both tolerance intervals based on normal distributions and a non-parametric version. I would start there and then you can explore further. You are correct to note that using the mean +/- 3 standard deviations might result in spec limits that you can meet given you might be looking at skewed distributions.
Thank you very much for your suggestion Karen. And the e-Handbook looks very useful.