3dLMEr -glfCode interactions between categorical and quantitative variable

AFNI Message Board
1

AFNI version info (afni -ver): Precompiled binary macos_13_ARM_clang: Jan 24 2024 (Version AFNI_24.0.02 'Caracalla')

Hello!

I am running a lme model with 1 categorical fixed variable (Category) and 2 quantitative fixed variables (age, motion) , as well as 2 categorical (Subj, exemplar) and 1 quantitative random variables (age). I would like to use -gltfCode to examine the interaction between specific levels of the 1 categorical fixed variable (e.g., Category: Digit vs. Letter) and age. However, the way I have specified it below seems to ignore the "age" specification and the results area the same as the comparison of "Digit vs. Letter". Is there a way to obtain a categorical x quantitative interaction using -glfCode?

Thanks for the help,

Andrew

3dLMEr -prefix 3dLMEr_output_Longit_Category_censor_exemplars \
-jobs 8 -bounds -2 2 -mask /Volumes/ANDY/NumberProcessing/ref/mask/BN_Atlas_HaskinsPeds_NL_template_2.5_mask+tlrc.HEAD \
-model "category*age+motion+(1+age|Subj)+(1|exemplar)" \
-qVars "age,motion" \
-qVarCenters "5.34,0.1" \
-glfCode DigitvLetterNovel "category : 1*Digit -1*Letter & 1*Digit -1*BACS1" \
-glfCode DotsvLetterNovel "category : 1*Dots -1*Letter & 1*Digit -1*BACS1" \
-glfCode DigitvLetter "category : 1*Digit -1*Letter" \
-glfCode DotsvLetter "category : 1*Dots -1*Letter" \
-glfCode DigitvDots "category : 1*Digit -1*Dots" \
-glfCode DigitvLetterNovelvAge "category : 1*Digit -1*Letter & 1*Digit -1*BACS1 age" \
-glfCode DotsvLetterNovelvAge "category : 1*Dots -1*Letter & 1*Digit -1*BACS1 age" \
-glfCode DigitvLettervAge "category : 1*Digit -1*Letter age" \
-glfCode DotsvLettervAge "category : 1*Dots -1*Letter age" \
-glfCode DigitvDotsvAge "category : 1*Digit -1*Dots age" \
-SS_type 3 \
-dataTable \
Subj    category  exemplar    age   motion   InputFile \
2

Some of the post hoc estimates would be better specified using -gltCode instead of -glfCode, as this would provide more informative directionality in the comparisons. For example:

-glfCode DigitvLetter "category : 1*Digit -1*Letter" \  
-glfCode DotsvLetter "category : 1*Dots -1*Letter" \  
-glfCode DigitvDots "category : 1*Digit -1*Dots" \

could be revised as:

-gltCode DigitvLetter "category : 1*Digit -1*Letter" \  
-gltCode DotsvLetter "category : 1*Dots -1*Letter" \  
-gltCode DigitvDots "category : 1*Digit -1*Dots" \

Additionally, the reason the age effects did not work is due to the following lines:

-glfCode DigitvLetterNovelvAge "category : 1*Digit -1*Letter & 1*Digit -1*BACS1 age" \  
-glfCode DotsvLetterNovelvAge "category : 1*Dots -1*Letter & 1*Digit -1*BACS1 age" \  
-glfCode DigitvLettervAge "category : 1*Digit -1*Letter age" \  
-glfCode DotsvLettervAge "category : 1*Dots -1*Letter age" \  
-glfCode DigitvDotsvAge "category : 1*Digit -1*Dots age" \

which should instead be specified as:

-glfCode DigitvLetterNovelvAge "category : 1*Digit -1*Letter & 1*Digit -1*BACS1 age :" \  
-glfCode DotsvLetterNovelvAge "category : 1*Dots -1*Letter & 1*Digit -1*BACS1 age :" \  
-gltCode DigitvLettervAge "category : 1*Digit -1*Letter age :" \  
-gltCode DotsvLettervAge "category : 1*Dots -1*Letter age :" \  
-gltCode DigitvDotsvAge "category : 1*Digit -1*Dots age :" \

Also, do you think that age might influence motion? If so, directly including motion as a covariate could be problematic. For further discussion on this issue, see this blog post and the associated paper.

Gang Chen