[R-sig-ME] Evaluating Z matrix

[Ben Bolker]

Pantelis Hadjipantelis <kalakouentin at ...> writes:

> 
> Hello,
> 
> I am going through lme4's book chapters and among the things that are
> slightly in disparity with the current state of things is the structure of
> Zt. Namely in chapter 3, section 2, Fig 3.4 presents the images of the Z^T
> matrices for the models:
> (fm06 <- lmer(Reaction ~ 1 + Days + (1 + Days|Subject), sleepstudy,
> REML=FALSE)); image(fm06 <at> Zt)
> (fm07 <- lmer(Reaction ~ 1 + Days + (1|Subject) + (0+Days|Subject),
> sleepstudy, REML=FALSE)); image(fm07 <at> Zt)
> and while in the book they appear different now they appear the same.
> I understand that the "important bit" is the L matrix that does remain
> "unchanged" but getting a different Z matrix than before does seem odd.
> What is the reason of this change in the generation of Z and Λ from the
> random-effects terms?

  There have been several major changes in the internal structures
of the lme4 package in the last couple of years. The structure
of Z has simply changed as a side effect. In fact, the
development version of lme4 has reverted to a behaviour more like
the version from 2010, where (x|grp) and (1|grp)+(0+x|grp) give
differently structured Z matrices.

> If lme4 book is not the most update to date reference about the generation
> of Z in lme4, what resource is?

   Unfortunately, the code -- specifically the lme4:::mkReTrms function.
But you should feel free to keep asking questions here.


> Just to clarifty I am using: MEMSS_0.9-0, lme4_0.999999-0, Matrix_1.0-11.
> lme4's book version is that of June 25, 2010.
> 
> Thank you in advance.
> 
> All best,
> Pantelis Hadjipantelis