Richly Parameterized Linear Models: Additive, Time Series, and Spatial Models Using Random Effects

This page is now updated fairly regularly, until recently only for Errata (alas) but more recently with Follow-on and Discussion.

I've added a section (on 26 September 2014) called "Credit where credit is due" describing instances in which another person had previously done something that I put in my book, or something really close, and I didn't give them credit because I didn't know of their work. To all these people, I apologize for my slack scholarship.

Links

Sample chapters.

Datasets.

Unpublished manuscripts cited in the book.

Errata: Regular, small errors are listed on this page. For three larger errors, I've re-written the sections as follows:

Section 17.1.1, p. 368 through the top of p. 369, on additive models: The derivation here is wrong, though the result is correct: you can't re-express this model's restricted likelihood in a scalar form except for the special cases noted. Download the corrected version here, posted 21 April 2015. This error invalidates Chapter 17's first two regular exercises and the corrected version replaces them with two new exercises. Download the new exercises here, posted 21 April 2015. Correcting the error in Section 17.1.1 required a couple small downstream corrections, which I've posted with the regular, small errors listed on this page.
Section 17.1.3, on the spectral approximation to a Gaussian process, had quite a few errors. Thanks to my PhD student Maitreyee Bose for finding most of the errors and figuring out how to make the theory work. Rather than ask you to fix the errors one by one, I've made a corrected version of Section 17.1.3; download the corrected version here, posted 30 April 2014. (This supersedes previous versions posted 14 February and 7 February 2014.)
Section 17.2.2, on the approximate restricted likelihood for the DLM puzzle: The theory given here was correct (well, I haven't found any errors yet) but my research assistant Ellie Duffy found an error in my code for computing the approximate one-step case deletion diagnostics. (Thanks, Ellie!) Correcting the error makes little qualitative change for most j but big changes for small j which is, of course, where the most interesting things are happening. Download a re-written version of this section here. The correct results are much cooler than the erroneous ones I put in the book, and I've also had another year to think about it, with help from Ellie, so I understand it better. If you want a more terse version, here are the lecture transparencies from which I re-wrote this section.

Credit where credit is due. All cases described on this page arose from my insufficiently diligent scholarship, and I hereby issue a blanket apology. I will include these citations, with suitable changes in the text, in the second edition, if there is one.

Follow-on and Discussion. This page includes

Things people sent me that were interesting, e.g., further examples of things mentioned in the book.
Class projects attempting something I suggested in the book.
The latest iteration on random effects old and new (Chapter 13).
[That's it so far -- I'll expand this list as needed.]

Computer code. Putting my computer code on a web site feels like selling my elementary-school art class paintings on eBay, but my editor at C&H suggested this, so I'll do it sooner or later. (But embarrassing myself publicly in this manner is a low-priority item, so don't hold your breath.)