Recent changes in arm (arm > 1.1-8)

There are some changes in arm. Most of them are related to the big changes lme4 has made. If your arm version number is > arm_1.1.8, then you might want to read the followings (or if you are a reader of Data Analysis Using Regression and Multilevel/Hierarchical Models, you should read this):

New functions:

1. multicomp.plot(): mcplot() is the wraper for it. It plots the significant differnece of a simulated array. For example, see Figure 5 in this paper.

2. traceplot(): we made traceplot into a generic function. So now it identifies “mcmc.list” and “bugs” objects. The old traceplot() in coda will be masked. But the functionality of it will be passed into arm as the S4 method.

Changes related to lme4.

The new releass of lme4 (version # > 0.999375-16) has many changes. See the detailed here. Basically, it adopts new alogrithm, which makes lme4 quicker and more robust. Here are some changes that affects arm since some functions in arm depend on lme4:

1. lmer(): now lmer() breaks into lmer(), glmer(), and nlmer() (linear, generalized linear, nonlinear and generalized nonlinear mixed-effects models). So if you are fitting a generalized linear model, instead of using lmer() and specify the link function, you should use glmer() and specify the link function.

2. mcsamp(): mcsamp() depends on mcmcsamp() in lme4. It is a way to simulate the mixed model fitted by lme4. Whereas arm does provide a sim() function that does the same thing, mcsamp() can deal with the sigularity problem of the cov-variance matrix. The most common error message related to the sigularity problem when you use sim() to simulate a “mer” model is “Sigma is not postive definite!” However, the new mcmcsamp() contains some problems. It is still under development. So the new arm takes out mcsamp(). We will put it back when mcmcsamp() works properly. This means if you used to use mcsamp() to do the simulation, you would not consider upgrading to the newest arm (arm > 1.1-8).

2 thoughts on “Recent changes in arm (arm > 1.1-8)

  1. Do you happen to know if the problems with mcmcsamp in lme4 have been resolved? Either mcmcsamp or mcsamp would be helpful for some analyses. Thanks for your help.

Comments are closed.