Hello All,

My question is related to data that ends up being heavily right skewed. We have an insertion force test that measures the force required to insert products into our device. The upper specification is 2 pounds and the majority of the data is around .1 and then skewed right. Obviously this data set fails the normality assessment. When running the capability study regardless of normality the Cpk and Ppk values are aroung 9-10. I realize this is not valid due to the normality, I include that information to show how far the data set is from the upper specification of 2.

When I use the Box-Cox to normalize the data I am able to due so however the new capability assessment ends up being 1-1.25 Cpk/Ppk.

My question is as follows: Is there a better assessment test I should be using to determine capability on a right skewed dataset?

I am running Minitab 14.

Thanks for any help.

Regards,

Jon

Try running a distribution ID plot. You can find it under Stat/Reliability-Survival and determine if any of the supported distribution models fit well.

Alternatively, you can look perfrom a simple % defective calculation. Since process capability 'indexes' are no more than estimates for determining past perfromance (how bad was it) or forcasting future performance if nothing cahgnes (How bad will it be) the effort used to get a statistically more elegent answer is probably not worth the added value of 'better' answer. While the statistics can be fun, they aren't necessarily contributing to the goal...

Of course, your time frame / # of tests used shoudl be large enought o be representative and any estimate you make for forcasting purposes is only good if the time period used was stable and in control...

Hello All,

My question is related to data that ends up being heavily right skewed.....The upper specification is 2 pounds and the majority of the data is around .1 and then skewed right.... When I use the Box-Cox to normalize the data I am able to due so however the new capability assessment ends up being 1-1.25 Cpk/Ppk.

My question is as follows: Is there a better assessment test I should be using to determine capability on a right skewed dataset?
Jon

There is a more severe issue on your case, The lack of lower spec limit, beware of Cpk or Ppk for one sided specs, try better Cpm (an indicator who takes the target in account) with the data transformated by box-cox.

Box Cox was demostrated to have a good behabiur in different conditions :agree1: