# 3dLME question

Hello,

We have within-subject design where participants are scanned on multiple visits (4 times per participant). Time between visits is 2 years. Some visits are missing for a few participants. We’ll call this variable ‘time’.

We also collected two quantitative variables (“weight”, “age”) during each visit, so we have two time-variant covariates to add to the model

1. Can 3dLME handles missing data for -qvars? For example, we have imaging data for participant 5, visits 1, 2,3,4, but are missing behavior data for visit 2.

2). With respect to the model, we are also interested in each of the two qvars effect on the imaging data. Would our model formula look like the following: "-model ‘timeweightage’?

1. If so, would this also provide the interaction terms of timeweight and timeage, weightage, and timeweight*age in the results?

Thanks

Michael

Can 3dLME handles missing data for -qvars? For example, we have imaging data for participant 5,
visits 1, 2,3,4, but are missing behavior data for visit 2.

No, there is no easy solution under the conventional statistical framework. I may implement an approach at the ROI level into the program RBA in the future.

With respect to the model, we are also interested in each of the two qvars effect on the imaging data.
Would our model formula look like the following: "-model ‘timeweightage’?

It depends on your prior knowledge and your research hypothesis. One possibility is to only account for linearity for each of the two quantitative variables:

-model ‘timeweight+timeage’

If you are interested in capturing the interaction between weight and age, create another quantitative variable Z, which is the product of the two after they are centered. Then you have

-model ‘timeweight+timeage+time*Z’

Thanks Gang.

We want to simplify our model and only include 1 qvar, weight.

The model would then be: -model ‘time*weight’

To investigate the investigate the effect of the interaction, we would compare the slopes for weight at each level of time, correct?

-gltlabel 1 ‘weight_ V1-V2’ -gltcode 1 'Time: 1V1 -1V2 weight : ’

-gltlabel 2 ‘weight_ V1-V3’ -gltcode 2 'Time: 1V1 -1V3 weight : ’

.
.
.

If this is correct, a significant contrast for weight_V1 - V2 from the above glt would mean the effect of weight is more correlated with the imaging data at V1 than V2? Would that be correct?

Thanks.

Michael

If this is correct, a significant contrast for weight_V1 - V2 from the above glt would mean the effect of weight is
more correlated with the imaging data at V1 than V2? Would that be correct?

Yes, that seems right. To further partition the interaction effect, you may consider adding three more lines

-gltcode 3 'Time: 1V1 weight : ’
-gltcode 4 'Time: 1
V2 weight : ’
-gltcode 5 'Time: 1*V3 weight : ’

Thanks Gang.