To do a complete analysis of the data, the rule is that we work our way left to right across this row of buttons. So we’ve already looked at the session window and we found the active variable. The next thing to check is the residuals. Well, this first one isn’t really residuals at all. This is just simply the Y variable of the experiment in the order that its presently presented. So thats sometimes useful but it’s not a residual. Here’s residuals versus fits. A fit is what the model says should be there and so we’ve got a set of residuals here and another set here and they’re all within the control limits. That looks pretty good.
Here’s residuals by run order. This is the order in which the data appear down the Y column and residuals of course are the error in the model, what the fit says should be going on there versus what actually happened there. We’d like to see those be nice and normally distributed. We can see here and the histogram they look good and none of them are out of limits and in the Anderson Darling test they test as normal. So we’re happy with that. Now, if we go to effects, we have greyed out everything that we have deleted which is most of it but here’s the effect of catalyst, that is that as catalyst goes from 1.7 to 2.3, our Y variable goes up and we can see how much over on the scale over here on the left.
If we want to see that in a scatterplot, then we can do that here and sure enough these don’t show much difference but we’ve got a lower cluster here than we have here. So okay, we’re good. ANOM we’re going to skip for now, that belongs in a different context but I do want to show you optimize. We built a mathematical model of a region of space and so I can now drag this anywhere in that region and over here on the left I can see what my predicted output is. That’s pretty nice. And I can double click in here and then I can enter a precise value. If I wanna know what it is at 2.2 I can enter that and it’ll calculate and just as you may have seen in other places, I can actually enter a lower spec limit and an upper spec limit and click calculate again and I’m not only get Y output variables, I get an estimate of CPK and PPK and that’s extremely handy and when you doing a designed experiments say in a research and design organization you want to be able to hand this over to production you say, okay, well here’s probably about as good as you’re going to do on process capability and that’s very useful.
Couple other general comments before we go. First, its very tempting to think, oh well let’s see, I’ve got to get, if I got four replicates, I’ve got to do this combination four times and so I’ll just to, well I’m, set up all take four data. Well, that’s not really fair. You change the setup and take a data point, change the set up and take a data point, change the set up and take a data point, and that’s good practice. The other thing that you might wanna see, and we’ll touch on this later, is you can enter what are called center points, those will be middle values. We use those to test for curvature. If we find curvature, then we’ll go on to a central composite design and also there’s a very tricky tool called blocking and that’s an advanced topic that we’ll catch later.