Handling Non Normal Data

2 Replies

Handling Non Normal Data

Posted by Siddharth Sanghvi on Aug 29, 2018 12:05 am

Hello Members,
I have had frequent run-ins with non normal data and this continues. This time the scenario is different and I need your advice on handling such data.

I was trying to perform the capability analysis of an injection molding process. I have 24 data points. I started by checking the normality of the data and it turned out to be non normal. I tried to find the best distribution on Minitab  but none matched since the p value for all distributions was far less than 0.05. The data could not be transformed using the Johnson's transformation also. While reviewing the data I realized that the values are closely stacked and very close to each other. In fact there were a few points with the same value. Looking at the data I could realize the process is stable, operating very close to the median with none crossing specification limits. How can I perform the capability analysis using this data?

Regards,
Siddharth Sanghvi
     
 

Re: Handling Non Normal Data

Posted by Ernest Phoon on Aug 29, 2018 10:47 am

Dear Sidhharth,

When I was handling Injection molding I preferred to look at the output instead of the parameters which as you have found are quite well controlled via the closed loop systems our new machines are capable of, per

"While reviewing the data I realized that the values are closely stacked and very close to each other. In fact there were a few points with the same value."

This creates problems for statistical programs like Minitab as there is hardly any variation to speak of and falls into the category of "inadequate measurement units" which you may like to search to get a better insight into what you are dealing with.

Yours sincerely,
Ernest

 

Re: Handling Non Normal Data

Posted by John Vandenbemden on Aug 29, 2018 6:23 pm

I do have a table that I have used for such situations.  If you are interested I can send it to you for your use.  I normally do not circulate these types of documents on an open response.  Contact me at jfviii@fuse.net if you are interested.
Best Regard
John
John Vandenbemden