Dear AFNI experts and users,
After running the 3dLME as below, I find a significant result in “-gltLabel 8 after_before_in_anode_now_or_notnow”. Does anyone know how to extract the effect of ‘after-before’ in anode for each individual in the current model? Then I can use them for scatter plot.
Thanks.
All the best.
Rujing
Does anyone know how to extract the effect of ‘after-before’ in anode for each individual in the current model?
By “anode” do you mean a voxel or an anatomically or statistically defined region? First, define your voxel coordinates or a mask for the region, then extract the values from each subject using 3dmaskdump -xbox (or -ibox) or 3dmaskave -mask
Hi Gang,
Nope, the “anode” did not mean a voxel or an anatomically or statistically defined region. The “anode” is one condition of within-subject variable “CiJi”. Similarly, the “QainHou” (two level: after vs before) and the “Contrast3” (two level: now vs notnow) variables are within-subject variables.
In the current LME model, effect of this Lable is significant: -gltLabel 6 after_before_in_anode_now -gltCode 6 ‘CiJi : 1anode QainHou : 1after -1before Contrast3 : 1now’ . That is to say, in the “anode” “now” condition, there was a significant difference between “after” and “before” conditions. Therefore, I tried to extract the data in the “anode” “now” condition for “after” and “before” condition separately, then data were fed into a paired t-test. However, t-test revealed that group comparison between “after” and “before” conditions on the data was not significant. I guess one reason for ‘not significant in the t-test, but significant in the LME’ is that: in the LME, the model was modelling many effects, including interactions and main effects of ‘CiJi’, ‘QainHou’, ‘Contrast3’, and main effects of age, education. In contrast, in the t-test, these effects were not modelled. Is that correct?
So is there any method to extract the only/simple effect of ‘after’ - ‘before’ in the ‘anode’ ‘now’ condition while regressing out interactions and main effects of ‘CiJi’, ‘QainHou’, ‘Contrast3’ (these were within-subject variables)?
Thanks.
All the best.
Rujing
In the current LME model, effect of this Lable is significant: -gltLabel 6 after_before_in_anode_now -gltCode 6 ‘CiJi : 1anode QainHou : 1after -1before Contrast3 : 1now’ . That is to say, in the “anode” “now” condition, there was a significant difference between “after” and “before” conditions. Therefore, I tried to extract the data in the “anode” “now” condition for “after” and “before” condition separately, then data were fed into a paired t-test.
The proper way to do this is to add two more specifications:
Then combining with the original one -gltCode 6 ‘CiJi : 1anode QainHou : 1after -1before Contrast3 : 1now’, you should be able to sort out the details. Similarly for the other contrasts in your 3dLME script.
I guess one reason for ‘not significant in the t-test, but significant in the LME’ is that: in the LME, the model was modelling many effects, including interactions and main effects of ‘CiJi’, ‘QainHou’, ‘Contrast3’, and main effects of age, education. In contrast, in the t-test, these effects were not modelled. Is that correct?
You have other explanatory variables plus more factor levels in the LME model. In contrast, your paired t-test does not take those variables into consideration. These are two different models.
Then combining with the original one -gltCode 6
‘CiJi : 1anode QainHou : 1after -1before
Contrast3 : 1now’, you should be able to sort out
the details. Similarly for the other contrasts in
your 3dLME script.
Is there any method to plot coefficient, SD/SE of this effect '-gltCode … ‘CiJi : 1anode QainHou : 1after Contrast3 : 1*now’?
Thanks.
All the best.
Is there any method to plot coefficient, SD/SE of this effect '-gltCode … ‘CiJi : 1anode QainHou : 1after Contrast3 : 1*now’?
Each post hoc inference through -gltCode gives you two values per voxel: effect estimate and t-statistic. You can directly obtain the standard error (dividing effect estimate by t-statistic). Once you have the standard error, I think there is a way to plot the result at the voxel level although I don’t know the detail on the AFNI GUI. For ROI-level plotting, do it with a graphical tool after extracting the resutls.
Each post hoc inference through -gltCode gives you
two values per voxel: effect estimate and
t-statistic. You can directly obtain the standard
error (dividing effect estimate by t-statistic).
Once you have the standard error, I think there is
a way to plot the result at the voxel level
although I don’t know the detail on the AFNI GUI.
For ROI-level plotting, do it with a graphical
tool after extracting the resutls.
Only considering the ROI-level plotting.
1, 3dLME return Z value, and how can it be transformed into t value?
2, The standard error and effect estimate could be used to bar plot with error bars. Could this bar-errorbar plot be transformed into scatter-point plot? Could I use residuals plus the effect estimate to measure the individual datapoint for scatter-plot?
3dLME return Z value, and how can it be transformed into t value?
In this case, standard error is beta divided by Z-value
Could this bar-errorbar plot be transformed into scatter-point plot? Could I use residuals plus the effect estimate to
measure the individual datapoint for scatter-plot?
If you really want to show scatterplot with data of a hierarchical/multilevel structure, things might be much more complicated than simple models such as conventional regression. See an example here:
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.