Hi, I'm Annie Dudley Zengi, the developer for control charts in JMP . In version 17, I added the Laney P' Prime and U' Prime Control Charts.
Just a little reminder to everybody, the control charts are intended to show the stability of your process, how well you can reliably make the same part of the same size over and over again over time.
Let's think about, in order to do that, you need to have a constant variance. An attribute chart, either Binomial or Poisson, assume a constant variance.
But what happens if your variance is not constant? What happens if the temperature changes or the humidity level changes, or the voltage changes, or all kinds of little things change in your process, and you can't rely on the variance being constant over time?
This is called over dispersion, or in rare instances, under dispersion. A one parameter distribution, the Binomial and/ or the Poisson can model over dispersion or under dispersion.
Now, Laney recognizes this and proposed that we normalize the data, account for varying subgroup sizes, and then compute a moving range. One thing to keep in mind with Laney control charts is the over dispersion and under dispersion happen typically when you have either large lot sizes or you have unequal lot sizes. It causes the traditional attribute charts to incorrectly characterize your process.
Let's take a look at a sample data that has, although these lot sizes are not terribly large, it does have some that are unequal in size. We'll open up washers and go to Control Chart Builder. We're going to model our number of parts defective.
I will change my chart type to attribute because this is only available in the attribute control charts. I'll pull in my lot size and the lot. Then recall that lot size 2 is the varying lot size. I'm going to pull this down into the lot size grouping here.
Now, we notice we have varying control limits and we have a couple of points out of control. I'm going to turn on our test beyond limits. We see we have two points out of control.
Now, we're looking at a Poisson chart. Also understand that Laney's P' Prime and U' Prime control charts only apply when the statistic is a proportion . Next, we need to change our statistic to a proportion. Again, we notice that those same two points are beyond the limits.
Now, if we change our Sigma so that we can choose a Laney P' Prime, we notice that now that we change it to proportion. We have four options here. Not only just the regular binomial and Poisson, we also have the Laney P' Prime and Laney U' Prime.
Now, we're looking at a U' chart right now of number of defective, so I'm going to change it to a Laney U' Prime just for consistency . You notice now those two points that were out of control are now in control. It's not indicating that our process is unstable at this point.
We'll take a little look at the formulas. There's our standardized values. We compute a moving range of those standardized values. We sum them up and then we compute an average . That Sigma sub Z is what is inserted into the formula for both the P' Prime limits and the U' Prime limits.
These look just like the formulas for the U' and the P' charts, only with that Sigma sub Z in there . That Sigma sub Z approaches one as over dispersion goes away . Keep in mind that U' Prime and P' Prime charts will match your U' and P' charts, respectively, if there is no over dispersion, and that this is available when the chart type is attribute chart and the statistic is proportion.
Thank you very much for your time. I hope you found this talk helpful, and feel free to reach out to me with any questions. Thank you.