3dLMEr without fixed effects

AFNI Message Board
,
1 
[picchionid@cn0032 ~]$ afni --version
Precompiled binary linux_rocky_8: Jul  6 2026 (Version AFNI_26.2.01 'Gordian III')
[picchionid@cn0032 ~]$ R --version
R version 4.5.2 (2025-10-31) -- "[Not] Part in a Rumble"

Gang,

Hi! How are you? How are data?!

I am trying to do some interesting analyses with 3dLMEr without fixed effects. Of course, I think they are interesting; I hope you think they are interesting.

First, in preparing to continue to take advantage of its -R2 option, I am trying to run a reduced model so that I can calculate an effect size with Cohen's f^2. The full model is this: -model "1+AAT+(1+AAT|Subj)". The reduced model I am trying is this: 1+(1+AAT|Subj), but it will not run and gives this error:

Error in checkForRemoteErrors(val) :
  4 nodes produced errors; first error: replacement has length zero
Calls: aperm ... clusterApply -> staticClusterApply -> checkForRemoteErrors
Execution halted

In this case, I do not need -gltCode because I am only interested in R^2.

Second, I am trying to do an LME equivalent of a single-sample t-test to account for subject clusters by running an intercept-only model: -model '1+(1|Subj)'. When I try, I get the same error. In this case, I would need -gltCode because I need the direction from which the mean is different from zero. I tried a workaround by including a constant factor in -dataTable without success:

...
-model "1+Constant+(1|Subj)" \
-gltCode mean 'Constant : 1*mean' \
...
** Error:
   Quitting due to model test failure...

What do you think?

Sincerely,

Dante

2

What kind of variable is AAT? Quantitative or qualitative? Within- or between-individual?

Gang Chen

3

Gang,

Hi. It is a quantitative (specified so with -qVars), within-individual variable.

Sincerely,

Dante

4

The following model is structurally appropriate:

-model "1+AAT+(1+AAT|Subj)"

In contrast, the reduced model:

-model 1+(1+AAT|Subj)

is problematic: the random effects of AAT need to be anchored to a population-level effect as in the first model above.

The model below is fine if at least some individuals have more than than one input file:

-model "1+(1|Subj)"

The following model is not well-constructed because Constant is not a variable in this context:

-model "1+Constant+(1|Subj)"

Gang Chen