Practical Concerns when Conducting Capability Studies

There are both positive and negative aspects to capability estimates. For example, Cp and Cpk estimates are highly sensitive to the assumption that you are sampling from a normal distribution—that is, most of the data points would be concentrated around the average (mean), forming a bell-shaped curve.

Furthermore, sampling from a stable system is essential to obtaining meaningful estimates of process performance for future production.

Many quality practitioners report solely the numerical values of the capability estimates. Others, however, note that the capability estimates are themselves merely statistics, or point estimates of the true capability of a process. As such, the use of confidence intervals for the true capability values may also be reported.

When sampling from stable, but non-Normal distributions, other strategies to obtain meaningful capability estimates may be appropriate, including:

• Transforming the data to be approximately well modeled by a Normal distribution.
• Using an alternative probability distribution, such as Weibull or Lognormal distributions.

Contributed by Keith M. Bower, a statistician and webmaster of www.KeithBower.com,