News
On June 1, we’re asking you to select a content label when starting a new topic in the Discussions area. Read more to find out why.
Choose Language Hide Translation Bar
Highlighted
Level III

## Control Chart builder, Non Normal data, Control Limits

Hi,

I have bounded data to Zero. Sure not normal distributed. How can I get the Control Limits? Most appreciated is a visualiation directly in the Control Chart Builder and an interactive solution. Any Help out there?

Thanks Peter

1 ACCEPTED SOLUTION

Accepted Solutions
Highlighted
Level IV

## Re: Control Chart builder, Non Normal data, Control Limits

I scripted my way around this problem and have submitted this to Laura Lancaster at the last discovery summit in Copenhagen. My approach is what Hadley mentioned: figure out which quantiles are analagous to the classical control limits, find out where those quantiles are for a given type of curve (gamma, log normal etc) and then draw all of that in the run chart platform. Manually I calculate alarms for test 1,2&5 + OOS rate and PPM values. It stores the control chart script as a script in your Data Table so that you can re-run with the hard coded control limits when there are new data points. Make sure you have figured out which fit you want to use before launching the script.

Regards,

Paul

4 REPLIES 4
Highlighted
Staff

## Re: Control Chart builder, Non Normal data, Control Limits

Hi Peter,

One possibility is to consider the quantiles as analogous to standard deviations.  In non-normally distributed data, the 84.1% quantile minus the 15.9% quantile encompasses 68.2% of the data, and therefor can perhaps be thought of as 1 std on either side of the median.  The +/-3s control limits could therefore be determined using the 99.9% and 0.1% quantiles, approximately.  Custom quantiles can be set using the redhotspot in the Distribution report window.

Highlighted
Level III

## Re: Control Chart builder, Non Normal data, Control Limits

good approach, thanks
Peter
Highlighted
Level IV

## Re: Control Chart builder, Non Normal data, Control Limits

I scripted my way around this problem and have submitted this to Laura Lancaster at the last discovery summit in Copenhagen. My approach is what Hadley mentioned: figure out which quantiles are analagous to the classical control limits, find out where those quantiles are for a given type of curve (gamma, log normal etc) and then draw all of that in the run chart platform. Manually I calculate alarms for test 1,2&5 + OOS rate and PPM values. It stores the control chart script as a script in your Data Table so that you can re-run with the hard coded control limits when there are new data points. Make sure you have figured out which fit you want to use before launching the script.

Regards,

Paul

Highlighted
Level III

## Re: Control Chart builder, Non Normal data, Control Limits

That pushes me forward. Many thanks for doing all that effort and share it with us.
Peter