3dLME Longitudinal Analysis

AFNI Message Board
1

Hi AFNI Crew,

I am attempting to use 3dLME to test if an interaction between symptom severity and age predict our brain measure in a longitudinal sample (scan 1 and follow-up scan 2). Our design is purely within-subjects (aside from a ‘sex’ nuisance covariate) with a model and data table similar to the following:

-model “SYMP_SEV*AGE + SEX + COV”
-ranEff “~1”

subj SYMP_SEV AGE SEX COV InputFile
_019 0.907720004 12.66 F 1377.5 ${data}/019_SCAN1
_019 1.176118983 15.23 F 1328.14 ${data}/019_SCAN2
_020 -0.925335639 15.68 M 1560.22 ${data}/020_SCAN1
_020 -1.147441677 14.63 M 1567.69 ${data}/020_SCAN2
_021 -1.316514248 11.04 F 1467.67 ${data}/021_SCAN1
_021 -0.740970281 12.09 F 1452.89 ${data}/021_SCAN2 \

…

_118 0.627853985 17.68 M 1804.6 ${data}/118_SCAN1
_118 -0.474358615 18.77 M 1833.68 ${data}/118_SCAN2
_122 0.78376315 14.24 F 1364.45 ${data}/122_SCAN1
_122 0.107800324 15.37 F 1385.81 ${data}/122_SCAN2

  1. Should we include a ‘scan number’ variable? Originally we thought this variable would be redundant, as this was a naturalistic follow-up and the only difference between our ‘pre’ and ‘post’ measures is the passage of time (which we assume is captured by the difference in ‘AGE’ between scan1 and scan2). However, we are unsure how the model is aware that each measure is within-subject and varies between scan1 and scan2.

  2. Would we also model SUBJ (as well as -ranEFF “~1 + SUBJ”) to indicate all continuous repeated measures within SUBJ?

Thanks for your help!
Taylor

2

Taylor,

  1. Should we include a ‘scan number’ variable?

If you don’t explicitly model the variable of scans, 3dLME would simply assume that the two scans have the same intercept other than the age effect, which is probably what you want. However, you may want to consider change the random effect specification to

-ranEff “~1+AGE”

  1. Would we also model SUBJ (as well as -ranEFF “~1 + SUBJ”) to indicate all continuous repeated measures within SUBJ?

No, 3dLME automatically assumes that the pivotal unit is subject. You need to change the label ‘subj’ in the data table to ‘Subj’.

A couple of further comments:

  1. Make sure to properly center SYMP_SEV and AGE first (https://afni.nimh.nih.gov/sscc/gangc/centering.html), then create another column that is the product of the two. And use the following

-model “SYMP_SEV+AGE + SYMP_AGE + SEX + COV”

  1. Make sure that you explicitly declare all the quantitative variables with option -qVars.
3

Thanks for your help Gang!

Just as a point of clarification, would the main effect of SYMP_AGE in your proposed model (the product of properly-centered SYMP_SEV and AGE by SUBJ) be equivalent to the interaction SYMP_SEV*AGE in my original model?

Best,
Taylor

4

would the main effect of SYMP_AGE in your proposed model (the product of properly-centered
SYMP_SEV and AGE by SUBJ) be equivalent to the interaction SYMP_SEV*AGE in my original model?

Probably equivalent if you also properly center the variables in your original model, but I never verified it myself.