Control Charts in Minitab

Thanks thats great, last question on this, I’m using a regular capability analysis for this but the data isn’t normal. It gives me a a good representation so I can see how often we hit our taget on a daily basis.

Does this sound OK??? Many thanks

Thanks for you help Trish.

I agree with Joel Smith. It depends on the source data. If the data is pass/fail, then use a binomial analysis. If it’s defects per unit, a Poisson analysis. And if the data is continuous data, then rule out special causes such as: look at the process for different levels of an X effecting it (humps in the data are a good sign), discrete data pretending it is continuous data, or evidence that the process is unstable (you can use graphical tools). If none of those are the case then try to identify the distribution using (on Minitab 15… don’t know where it is on 16) Stat>Quality Tools>Identify Individual Distribution. If it doesn’t fit any of those distributions, look at the process for different levels of an X effecting it (humps in the data are a good sign), discrete data pretending it is continuous data, or evidence that the process is unstable (you can use graphical tools).

I also have a caution on your % data in your control chart. Find out what the root data is; how was that percentage calculated. There might be binomial or other continuous data behind the scenes that can give you a more effective control chart to use for identifying patterns and trends.

Chris-

Graph the data. If it is not highly skewed and the sample sizes are wildly variable (like 200 parts one day and 2000 the next) then just use the I-Chart. It is robust to many distributions of data and works well unless you have one of those criteria I just described.

Capability Analysis is much more sensitive to distribution, but again graph the data or even better use a probability plot, and see if you have a good distribution fit. The Normal distribution approximates the binomial for good sample sizes, so if your data is in control you might be surprised how good that fit is. And if it is good, then go ahead and use Normal Capability Analysis.

You’ll be in good shape if you can do those things with the data you have, and can then focus your efforts on solving problems instead of finding only slightly better statistical methods.

Good luck!

Joel