Attribute capability indices - Cpk and/or Ppk value for an attribute type of feature

[Bill Ryan]

I've been asked if it's possible to compute a Cpk and/or Ppk value for an attribute type of feature. My from the hip response was "No it doesn't make any sense". As I thought more on it I began questioning that answer. I understand that rating the attribute with a Likert scale (for example) might make some sense, but is it possible AND make sense to figure an index for a pass/fail type of requirement?

I understand that I could just backtrack from a table if I knew the PPM value, but, once again, that value doesn't make any sense to me. Maybe I'm just overthinking (maybe it's just a "brain freeze" type of thing, but I've really gotten myself confused ).

Any help is, as always, most appreciated.

[Jim Wynne]

Originally Posted by Bill Ryan

I've been asked if it's possible to compute a Cpk and/or Ppk value for an attribute type of feature. My from the hip response was "No it doesn't make any sense". As I thought more on it I began questioning that answer. I understand that rating the attribute with a Likert scale (for example) might make some sense, but is it possible AND make sense to figure an index for a pass/fail type of requirement?

I understand that I could just backtrack from a table if I knew the PPM value, but, once again, that value doesn't make any sense to me. Maybe I'm just overthinking (maybe it's just a "brain freeze" type of thing, but I've really gotten myself confused ).

Any help is, as always, most appreciated.

The inherent problem of attributes analysis is that it tells you nothing about proximity to specifications limits. You know only that a thing is "good" or "not good." For this reason, the concept of Cpk makes no sense, because the whole idea of Cpk is to compare the process mean to the specification limits. Anyone who tells you he can calculate Cpk from pass/fail data is blowing smoke. On the other hand, if you know the percent defective (or number of defects) why would you need a capability index value?
Also, there are a number of reasons why it's not a good idea to go backwards from PPM to Cpk, not the least of which is that the PPM value in question may have been arrived at under dubious assumptions that conflict with what Cpk is meant to convey. There are no universally accepted methods for computing PPM as there are with Cpk, so it's easy to get into an apples-and-oranges comparison with misleading results.

[Statistical Steven]

Originally Posted by JSW05

The inherent problem of attributes analysis is that it tells you nothing about proximity to specifications limits. You know only that a thing is "good" or "not good." For this reason, the concept of Cpk makes no sense, because the whole idea of Cpk is to compare the process mean to the specification limits. Anyone who tells you he can calculate Cpk from pass/fail data is blowing smoke. On the other hand, if you know the percent defective (or number of defects) why would you need a capability index value?
Also, there are a number of reasons why it's not a good idea to go backwards from PPM to Cpk, not the least of which is that the PPM value in question may have been arrived at under dubious assumptions that conflict with what Cpk is meant to convey. There are no universally accepted methods for computing PPM as there are with Cpk, so it's easy to get into an apples-and-oranges comparison with misleading results.

I agree in principle that Cpk is meanigless for attributes data. But then again, any capability index is meaningless unless the specifications are meaningful. Therefore, if the attribute specification is meaningful (i.e. less than 2 defects) I think you can use a PPM like comparison to monitor the process.

[Jim Wynne]

Originally Posted by Statistical Steven

I agree in principle that Cpk is meanigless for attributes data. But then again, any capability index is meaningless unless the specifications are meaningful.

An index might be less than meaningful for practical purposes while still mathematically correct, in other words.

Originally Posted by Statistical Steven

Therefore, if the attribute specification is meaningful (i.e. less than 2 defects) I think you can use a PPM like comparison to monitor the process.

I think you're condusing specifications with expected yields, or AQL. The "specification" in this context is a limit value for evaluation of each piece--"must be between x and y," or "must be x maximum" or "y minimum". IMO, PPM doesn't make much sense unless you're actually going to make a million of something. In most cases, the percent defective should suffice, because the tacit message being sent when using attributes inspection is that you don't care about being more exact.

[Statistical Steven]

Originally Posted by JSW05

In most cases, the percent defective should suffice, because the tacit message being sent when using attributes inspection is that you don't care about being more exact.

I always say it is not that one does not care...but rather...does not care to find a measurement that is more exact.

[Arvind]

If you wish to convert Defect per Million opportunities- DPMO to capability index , you may use following formula from Schmidt and Launsby

Sigma= 0.8406+ SQRT(29.37-(2.221*Ln(DPMO)))
Capability= Sigma/3.

This equation holds good for defects less than 55 % which is quite acceptable in most cases.

Arvind

[dylin123]

It is good .I have been interested in it

[Bill Ryan]

Thanks for the responses. I think JSW's first response was what my brain was trying to tell me but, at the time, it just didn't click.