Non-normal Data Cpk

P

Priss80

The data that collected for analysis found not normal by Normality test in minitab. Can i proceed to culculate the CPK using the "capability analysis normal" in minitab or use the "capability analysis non-normal" tab?
The same data i plug into excel to calculate the CPK, the result is similar to the result from "capability analysis normal" . CAn anyone help to explain why?
The other question is what should i do if found the data collected is non-normal?

:thanx:
 

Bev D

Heretical Statistician
Leader
Super Moderator
What do your Customer requirements say about non-normal data?

General advice: fi your data are non-normal proceeding with a capability index calculation based on normality would be foolish. That is why Minitab provides you an alternative approach. The reason that manual calculations in EXCEL match the 'normal data' approach in minitab is that you and minitab are using the same formula. Minitab's formula for non-normal data adjusts for the non-normality; it is a different formula.
 

Miner

Forum Moderator
Leader
Admin
There are two options for you in Minitab:
  1. Capability Analysis (Normal Distribution) > Transform, select Box-Cox power transformation or Johnson transformation.
  2. Capability Analysis (Nonnormal Distribution) > Fit distribution... (run Individual Distribution Identification first).
I recommend option 2. Option 1 transforms the specification also and becomes very difficult to understand and explain to your stakeholders.
 
P

Priss80

In minitab, there is a within (red curve) and overall line (black dotted curve) in capability analysis. What does it mean?
 

Miner

Forum Moderator
Leader
Admin
The Within curve is based on the Within-subgroup variation only, while the Overall curve is based on the combined Within-subgroup and Between-subgroup variation. The Within is essentially short-term variation while the Overall is long-term variation assuming that the samples were collected over a long period of time, and are not consecutive samples.
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
It is important to understand the correct interpretation of the "normality test". Read this as a great reference: https://elsmar.com/Forums/blog.php?b=169

I ALWAYS do a curve fit first, using Distribution Analyzer - it is way easier to use and interpret the output than Minitab (since it is designed to do one thing.) Then I THINK to determine which of the scenarios make the most sense (since the statistics can NOT do that for you):

Is it normal, and should it be?
Is it non-normal and should it be?


If it is the distribution that makes sense, continue on to calculate capability (If non-normal, Cpk usually does not apply. If you are in automotive, you already know that from AIAG PPAP 4th edition Section 2.2.11.5.)
 
Top Bottom