ORIGINAL POSTING: I understand process capability for variable data. What happens when you calculate process capability for attribute data? In many cases the attribute data may refer to a one sided specification limit only, so converting the DPMO to a Z level and then an equivalent Cp or Pp seems erroneous. One of the books I read said that process capability for attribute data should be defined in terms of the CL of the relevant attribute control-chart (c, n..)? Why do we bother to convert Cp and Pp to Sigma levels? Why not just work with Cp or Pp? How does the timming of the data collection influence the calculation? I get the impression that if I've collected long-term data the sigma calculator assumes a 1.5 sigma shift and then corrects it for the shift to give me an equivalent short-term sigma ratting. What if I've corrected the process and I want to predict the future sigma ratting?
RESPONSE: When someone asks me a question like this I often return with a question asking them: how long them have been involved in the quality industry? Their typical response is " __ years." I then respond saying that if you are confused about this, don't you think that your CEO and line worker would also be confused?
Upon getting a positive response, I then suggest that organizations do not use Cp, Cpk, Pp, Ppk, or sigma quality metrics in their organization. These metrics can be very confusing and, in addition, their actual value can be very much a function of how the sampling is conducted; e.g., from a given process one person could get a dramatically different Cpk value than other person simply by having a different sampling plan.
The approach I suggest is using an infrequent subgrouping/sampling plan for both attribute and continuous data situations. Data is plotted as a 30,000-foot-level control chart. If the data are in-control/predictable and the data are continuous, we would report a process capability/performance metric as the estimated percentage or PPM non-conformance. If the data are attribute (and in-control/predictable at the 30,000-foot-level), the process capability/performance metric would usually be the center of the 30,000-foot-level chart.
Community metadata Could not be loaded. No pre-configured community forum parent found for the incoming fid: . You need to add this forum into custom_code/community.sql
Send to friend
Comments
forrestbreyfogle 6/8/2004
ORIGINAL POSTING: I understand process capability for variable data. What happens when you calculate process capability for attribute data? In many cases the attribute data may refer to a one sided specification limit only, so converting the DPMO to a Z level and then an equivalent Cp or Pp seems erroneous. One of the books I read said that process capability for attribute data should be defined in terms of the CL of the relevant attribute control-chart (c, n..)? Why do we bother to convert Cp and Pp to Sigma levels? Why not just work with Cp or Pp? How does the timming of the data collection influence the calculation? I get the impression that if I've collected long-term data the sigma calculator assumes a 1.5 sigma shift and then corrects it for the shift to give me an equivalent short-term sigma ratting. What if I've corrected the process and I want to predict the future sigma ratting?
RESPONSE: When someone asks me a question like this I often return with a question asking them: how long them have been involved in the quality industry? Their typical response is " __ years." I then respond saying that if you are confused about this, don't you think that your CEO and line worker would also be confused?
Upon getting a positive response, I then suggest that organizations do not use Cp, Cpk, Pp, Ppk, or sigma quality metrics in their organization. These metrics can be very confusing and, in addition, their actual value can be very much a function of how the sampling is conducted; e.g., from a given process one person could get a dramatically different Cpk value than other person simply by having a different sampling plan.
The approach I suggest is using an infrequent subgrouping/sampling plan for both attribute and continuous data situations. Data is plotted as a 30,000-foot-level control chart. If the data are in-control/predictable and the data are continuous, we would report a process capability/performance metric as the estimated percentage or PPM non-conformance. If the data are attribute (and in-control/predictable at the 30,000-foot-level), the process capability/performance metric would usually be the center of the 30,000-foot-level chart.
Reference: Implementing Six Sigma, 2nd ed: Section 1.19 (page 45), 11.3-11.5 (page 258), sec 11.16 (page 283).
Hope this helps.
Forrest Breyfogle
512-996-8288
forrest@smartersolutions.com