Possible error in the design matrix for robust design

[annaren]

Hi
I going through chapter 16 on the Robust Design and I am currently using the sample data set rd_complex1.INP. On page 22 the manual shows the coding for the DM when p=c parameters but I think that there is a column missing : p session 1 when there is p but no c. If you put this column in you get the correct number of parameters and other values that the chapter then preceeds with. Is this correct? I just wanted to check that I am understanding the workings of the Robust Design.

[Bill Kendall]

Thanks for pointing this out. I can understand the confusion. The row for p1 is not displayed in the figure. You say you added a column, so maybe you assigned a different beta for p1. I suspect that is exactly what Evan did in this case, and the column you mention is probably the left-most one in the figure, with all zeroes for the rows listed. Therefore in the results you see that the estimate for p1 is different from any of the other estiamates.

So I believe you have the mechanics of it exactly right. However, based on another part of your comment you might slightly misunderstand one aspect. You refer to "p session 1 when there is p but no c". You are partly confusing sessions (primary periods) with occasions within sessions (secondary periods). Session 1 has recapture (c) parameters but in this model they are assumed equal to the p's. For each session, the p for the first occasion within that session has no matching p. We'll try to clarify the discussion for this example of time variation. There is time variation across primary periods, but the PIM structure indicates the p's are assumed constant within a primary period.

[cooch]

Thanks for pointing this out. I can understand the confusion. The row for p1 is not displayed in the figure. You say you added a column, so maybe you assigned a different beta for p1. I suspect that is exactly what Evan did in this case, and the column you mention is probably the left-most one in the figure, with all zeroes for the rows listed. Therefore in the results you see that the estimate for p1 is different from any of the other estimates.

Indeed - I accidentally 'cut off' the top of that particular figure. I've just uploaded a corrected version, which deals with that - and a few other minor issues related to the 'complex' example. Note that in rendering the new version of Chapter 16, and this particular example, I generated a new version of complex_rd.inp, which is available in the zip archive containing all the example files.

So I believe you have the mechanics of it exactly right. However, based on another part of your comment you might slightly misunderstand one aspect. You refer to "p session 1 when there is p but no c". You are partly confusing sessions (primary periods) with occasions within sessions (secondary periods). Session 1 has recapture (c) parameters but in this model they are assumed equal to the p's. For each session, the p for the first occasion within that session has no matching p. We'll try to clarify the discussion for this example of time variation. There is time variation across primary periods, but the PIM structure indicates the p's are assumed constant within a primary period.

Correct. We will clarify this text at some point, but examination of the PIM structure makes it pretty obvious (to us) what is going on.