Hello,

I want to ask three questions, relates to second level covariates.


1. In case I would like to perform ANCOVA, with 2 covariates (that I would like to control, both are continuous like age and IQ ), with 2 groups (control/ Parkinson), and want to compare the connectivity between the groups, for example:

control: 11110000
Parkinson: 00001111
age: 25 36 37 38 38 39 30 50
IQ: 98 100 32 93 89 120 90 130

To see if the connectivity is higher among the experiment I will use this contrast: [-1 1 0 0]
Is this right? Is this is the same for adding more continues covariates [-1 1 0 0 0] and so on?

2.Binary covariate:


In case I would like to use a covariate (in order to control it) that is binary (such as the groups), do I use the covariates in the same way? E.g., if I want to see the effect of IQ and control for my two groups:

control: 00001111
Parkinson: 11110000
IQ: 98 100 32 93 89 120 90 130

contrast: [0 0 1]
Is this is right, or is it different in the case of binary covariates?

3. Statistical test:


In the case I define those contrast:

control: 11110000
Parkinson: 00001111
age: 25 36 37 38 38 39 30 50

and check this contrast: [-1 1 0] (Parkinson>control)
I get as a result value of a t-test (T(53)>x). Why don't I get an F value? This is actually an ANCOVA test?

Thank you very much for the answers!
Tal

Tal-

MRIcron just helps you draw the lesion. You will want to use a different tool for statistics. For ANCOVA, you would want to use VLSM, and should direct your questions to that group https://aphasialab.org/vlsm/

Another option would be to conduct a permutation-thresholded Freedman-Lane analysis using NiiStat
  https://www.nitrc.org/plugins/mwiki/index.php/niistat:MainPage
For this, your Excel file would look like this

ID PD AGE IQ
C1 0 25 98
C2 0 36 92
C3 0 37 99
C4 0 38 102
P1 1 39 104
P2 1 30 100
P3 1 50 98
P4 1 33 96

And your analysis would be a [1 0 0] analysis. Since NiiStat is designed for lesion work, we tend to assume that most brain injury impairs behavior, so we have conduct a one-tailed test. Based on the distribution of the data, the two tails may not be symmetrical, so you can have different scores for the positive end of the tail and the negative. I am a huge fan of permutation thresholding, it solves a lot of ills in Neuroimaging.

By the way, NiiStat should report Z-scores not F or T scores. With F and T scores, you need to know the degrees of freedom to interpret them. Computers are extremely efficient at transforming F/T scores to z-scores. Humans are much less good at this. So I let the computers do the hard work of converting the statistical values to this intuitive metric.

Dear Chris, 

Thank you very much for the answer. 

I am not familiar with NiiStat, but sounds very helpfull and I will check it.  You mention here that it's initially meant for lesions. I am working in the field of neuropsyhciatry (schizophrenia data), so no clear lesions. Is this is aimed also for this? 

Last, is there is no way to do those kind of calculations via CONN? 

Best, 
Tal

I was simply pointing out that with lesions we have a strong one-tailed hypothesis, so we traditionally put 0.05 as the threshold and only look at one tail. If you have a two-tailed hypothesis (e.g SZ may result in some regions may showing increased activity, others decreased), you would have a two-tailed hypothesis and use 0.025 as the threshold for each tail. NiiStat is a general linear model tool, and it can analyze voxels, regions of interest and connectomes. It is agnostic regarding the source of the data.

You may also want to look at PALM
  https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/PALM
which is very similar to NiiStat, but does allow different models.

Thank you very much, 

Best, 
Tal