help > ROI statistics in NBS
Showing 1-6 of 6 posts
Feb 9, 2022 04:02 PM | Patrick McCunn
ROI statistics in NBS
Hi Andrew
Thanks for this tool
I just want to confirm my thought process below. I have done whole brain analysis and am currently beginning to look at specific ROI connectivity. For example assume i have 5 ROI's and i want to see only the differences in connectivity between groups in ROI #3. Am i correct to assume i could simply 0 all the other entries and analyze the resulting matrices similar to the matrix below? (x being numeric connectivity values)
0 1 2 3 4 5
1 0 0 x 0 0
2 0 0 x 0 0
3 x x x x x
4 0 0 x 0 0
5 0 0 x 0 0
Thanks for this tool
I just want to confirm my thought process below. I have done whole brain analysis and am currently beginning to look at specific ROI connectivity. For example assume i have 5 ROI's and i want to see only the differences in connectivity between groups in ROI #3. Am i correct to assume i could simply 0 all the other entries and analyze the resulting matrices similar to the matrix below? (x being numeric connectivity values)
0 1 2 3 4 5
1 0 0 x 0 0
2 0 0 x 0 0
3 x x x x x
4 0 0 x 0 0
5 0 0 x 0 0
Feb 10, 2022 05:02 AM | Andrew Zalesky
RE: ROI statistics in NBS
Hi Patrick,
yes - your understanding is perfectly correct.
Setting a connection's values to 0 for all subjects ensures that connection is not considered in statistical testing. Note that it needs to be set to zero in every subject.
best,
Andrew
Originally posted by Patrick McCunn:
yes - your understanding is perfectly correct.
Setting a connection's values to 0 for all subjects ensures that connection is not considered in statistical testing. Note that it needs to be set to zero in every subject.
best,
Andrew
Originally posted by Patrick McCunn:
Hi Andrew
Thanks for this tool
I just want to confirm my thought process below. I have done whole brain analysis and am currently beginning to look at specific ROI connectivity. For example assume i have 5 ROI's and i want to see only the differences in connectivity between groups in ROI #3. Am i correct to assume i could simply 0 all the other entries and analyze the resulting matrices similar to the matrix below? (x being numeric connectivity values)
0 1 2 3 4 5
1 0 0 x 0 0
2 0 0 x 0 0
3 x x x x x
4 0 0 x 0 0
5 0 0 x 0 0
Thanks for this tool
I just want to confirm my thought process below. I have done whole brain analysis and am currently beginning to look at specific ROI connectivity. For example assume i have 5 ROI's and i want to see only the differences in connectivity between groups in ROI #3. Am i correct to assume i could simply 0 all the other entries and analyze the resulting matrices similar to the matrix below? (x being numeric connectivity values)
0 1 2 3 4 5
1 0 0 x 0 0
2 0 0 x 0 0
3 x x x x x
4 0 0 x 0 0
5 0 0 x 0 0
Feb 14, 2022 03:02 AM | Patrick McCunn
RE: ROI statistics in NBS
Awesome thank you Andrew! I have a second (much longer) question if
that is ok. I'm currently running an analysis on two groups using
age and sex as covariates. I have used a design matrix similar to
that below with group 1 and group 2 represented in the first two
columns, age (in days: demeaned) in the third column as well as sex
(demeaned - equal number in groups) in the fourth:
1 0 45 1
1 0 -33 -1
0 1 22 -1
0 1 -11 1
and contrasts of [1 -1 0 0] as well as [-1 1 0 0] with a two-sided t-test. I have found one network with significant connections and exported the resulting matrix for further analysis. I interpret these as functional connectivity differences when the effects of age and sex are controlled.
Now I would like to determine if the effects I am observing are dependent on sex and/or age. To test this, is it correct to look for interaction effects? My plan was to use a design matrix as below with column 1 = intercept, column 2 = group (experimental vs control), age (in days: demeaned) in the third column as well as sex (demeaned - equal number in groups) in the fourth:
1 1 45 1
1 1 -33 -1
1 -1 22 -1
1 -1 -11 1
with the following contrasts:
[0 0 1 0] and [0 0 -1 0] to test the positing/negative interaction effect of age
[0 0 0 1] and [0 0 0 -1] to test the positing/negative interaction effect of sex
Multiple comparison correction will be taken into account. My hope is to be able to determine if, for example, older females have unique functional connectivity changes not seen in younger males or younger females.
Sorry for the long question, any advice you have is greatly appreciated.
1 0 45 1
1 0 -33 -1
0 1 22 -1
0 1 -11 1
and contrasts of [1 -1 0 0] as well as [-1 1 0 0] with a two-sided t-test. I have found one network with significant connections and exported the resulting matrix for further analysis. I interpret these as functional connectivity differences when the effects of age and sex are controlled.
Now I would like to determine if the effects I am observing are dependent on sex and/or age. To test this, is it correct to look for interaction effects? My plan was to use a design matrix as below with column 1 = intercept, column 2 = group (experimental vs control), age (in days: demeaned) in the third column as well as sex (demeaned - equal number in groups) in the fourth:
1 1 45 1
1 1 -33 -1
1 -1 22 -1
1 -1 -11 1
with the following contrasts:
[0 0 1 0] and [0 0 -1 0] to test the positing/negative interaction effect of age
[0 0 0 1] and [0 0 0 -1] to test the positing/negative interaction effect of sex
Multiple comparison correction will be taken into account. My hope is to be able to determine if, for example, older females have unique functional connectivity changes not seen in younger males or younger females.
Sorry for the long question, any advice you have is greatly appreciated.
Feb 14, 2022 07:02 AM | Andrew Zalesky
RE: ROI statistics in NBS
Hi Patrick,
to test for interaction effect, additional columns need to be added to the design matrix.
A group by age interaction can be added by multiplying the 2nd and 3d column to get:
45
-33
-22
11
If you append the above column as the 5th column of your design matrix, the contrast to test for an interaction would then be:
[ 0 0 0 0 1] or [0 0 0 0 -1]
The same approach can be followed for the group by sex interaction. I.e. a 6th column can be added by multiplying the 2nd and 4th columns
I hope this is clear.
The contrasts that you have currently suggested will test for a main effect of age and sex, but will not test whether group 1 connectivity is higher in females, but not males (for example).
Andrew
Originally posted by Patrick McCunn:
to test for interaction effect, additional columns need to be added to the design matrix.
A group by age interaction can be added by multiplying the 2nd and 3d column to get:
45
-33
-22
11
If you append the above column as the 5th column of your design matrix, the contrast to test for an interaction would then be:
[ 0 0 0 0 1] or [0 0 0 0 -1]
The same approach can be followed for the group by sex interaction. I.e. a 6th column can be added by multiplying the 2nd and 4th columns
I hope this is clear.
The contrasts that you have currently suggested will test for a main effect of age and sex, but will not test whether group 1 connectivity is higher in females, but not males (for example).
Andrew
Originally posted by Patrick McCunn:
Awesome thank you Andrew! I have a second (much
longer) question if that is ok. I'm currently running an analysis
on two groups using age and sex as covariates. I have used a design
matrix similar to that below with group 1 and group 2 represented
in the first two columns, age (in days: demeaned) in the third
column as well as sex (demeaned - equal number in groups) in the
fourth:
1 0 45 1
1 0 -33 -1
0 1 22 -1
0 1 -11 1
and contrasts of [1 -1 0 0] as well as [-1 1 0 0] with a two-sided t-test. I have found one network with significant connections and exported the resulting matrix for further analysis. I interpret these as functional connectivity differences when the effects of age and sex are controlled.
Now I would like to determine if the effects I am observing are dependent on sex and/or age. To test this, is it correct to look for interaction effects? My plan was to use a design matrix as below with column 1 = intercept, column 2 = group (experimental vs control), age (in days: demeaned) in the third column as well as sex (demeaned - equal number in groups) in the fourth:
1 1 45 1
1 1 -33 -1
1 -1 22 -1
1 -1 -11 1
with the following contrasts:
[0 0 1 0] and [0 0 -1 0] to test the positing/negative interaction effect of age
[0 0 0 1] and [0 0 0 -1] to test the positing/negative interaction effect of sex
Multiple comparison correction will be taken into account. My hope is to be able to determine if, for example, older females have unique functional connectivity changes not seen in younger males or younger females.
Sorry for the long question, any advice you have is greatly appreciated.
1 0 45 1
1 0 -33 -1
0 1 22 -1
0 1 -11 1
and contrasts of [1 -1 0 0] as well as [-1 1 0 0] with a two-sided t-test. I have found one network with significant connections and exported the resulting matrix for further analysis. I interpret these as functional connectivity differences when the effects of age and sex are controlled.
Now I would like to determine if the effects I am observing are dependent on sex and/or age. To test this, is it correct to look for interaction effects? My plan was to use a design matrix as below with column 1 = intercept, column 2 = group (experimental vs control), age (in days: demeaned) in the third column as well as sex (demeaned - equal number in groups) in the fourth:
1 1 45 1
1 1 -33 -1
1 -1 22 -1
1 -1 -11 1
with the following contrasts:
[0 0 1 0] and [0 0 -1 0] to test the positing/negative interaction effect of age
[0 0 0 1] and [0 0 0 -1] to test the positing/negative interaction effect of sex
Multiple comparison correction will be taken into account. My hope is to be able to determine if, for example, older females have unique functional connectivity changes not seen in younger males or younger females.
Sorry for the long question, any advice you have is greatly appreciated.
Feb 14, 2022 09:02 PM | Patrick McCunn
RE: ROI statistics in NBS
Thanks Andrew
I believe I'm following you but im struggling slightly with the interpretation. So following the example again(column 1= intercept, column 2=group, column 3=age, column 4=sex, column 5=group x sex) if I created the following design matrix
1 1 45 1 1
1 1 -33 -1 -1
1 -1 22 -1 1
1 -1 -11 1 -1
(Male =1, Female =-1 for this example). And then ran a t-test with the contrast [0 0 0 0 1].
Am I correct in understanding that if this uncovers significant networks, this would indicate higher connectivity in males/lower connectivity in females for group 1 and/or the inverse relationship for the group assigned -1? Therefore I would look to descriptive statistics to define the directions and relations in greater detail?
Lastly could I take this one step farther and look for a group x sex x age interaction by multiplying column 2, column 3 and column 4 together
1 1 45 1 45
1 1 -33 -1 33
1 -1 22 -1 22
1 -1 -11 1 11
with contrast [0 0 0 0 1] or [0 0 0 0 -1]? In this case the direction of interactions would again be determined through descriptive statistics.
Sorry for the barrage of questions, I really appreciate your help!
I believe I'm following you but im struggling slightly with the interpretation. So following the example again(column 1= intercept, column 2=group, column 3=age, column 4=sex, column 5=group x sex) if I created the following design matrix
1 1 45 1 1
1 1 -33 -1 -1
1 -1 22 -1 1
1 -1 -11 1 -1
(Male =1, Female =-1 for this example). And then ran a t-test with the contrast [0 0 0 0 1].
Am I correct in understanding that if this uncovers significant networks, this would indicate higher connectivity in males/lower connectivity in females for group 1 and/or the inverse relationship for the group assigned -1? Therefore I would look to descriptive statistics to define the directions and relations in greater detail?
Lastly could I take this one step farther and look for a group x sex x age interaction by multiplying column 2, column 3 and column 4 together
1 1 45 1 45
1 1 -33 -1 33
1 -1 22 -1 22
1 -1 -11 1 11
with contrast [0 0 0 0 1] or [0 0 0 0 -1]? In this case the direction of interactions would again be determined through descriptive statistics.
Sorry for the barrage of questions, I really appreciate your help!
Feb 14, 2022 10:02 PM | Andrew Zalesky
RE: ROI statistics in NBS
Hi Patrick,
yes your interpretation is correct.
A signification interaction effect would indicate higher connectivity in males in group 1 and lower connectivity in females in group 1 (but connectivity in group 1 overall could still be significantly different from group 2). You could extract the mean values of connectivity for the network associated with the interaction and plot them out to better understand the effect.
Three-way or higher interaction can be difficult to interpret and be sure that your sample size is adequate for high order interactions. I probably would stick to two-way interactions.
Andrew
Originally posted by Patrick McCunn:
yes your interpretation is correct.
A signification interaction effect would indicate higher connectivity in males in group 1 and lower connectivity in females in group 1 (but connectivity in group 1 overall could still be significantly different from group 2). You could extract the mean values of connectivity for the network associated with the interaction and plot them out to better understand the effect.
Three-way or higher interaction can be difficult to interpret and be sure that your sample size is adequate for high order interactions. I probably would stick to two-way interactions.
Andrew
Originally posted by Patrick McCunn:
Thanks Andrew
I believe I'm following you but im struggling slightly with the interpretation. So following the example again(column 1= intercept, column 2=group, column 3=age, column 4=sex, column 5=group x sex) if I created the following design matrix
1 1 45 1 1
1 1 -33 -1 -1
1 -1 22 -1 1
1 -1 -11 1 -1
(Male =1, Female =-1 for this example). And then ran a t-test with the contrast [0 0 0 0 1].
Am I correct in understanding that if this uncovers significant networks, this would indicate higher connectivity in males/lower connectivity in females for group 1 and/or the inverse relationship for the group assigned -1? Therefore I would look to descriptive statistics to define the directions and relations in greater detail?
Lastly could I take this one step farther and look for a group x sex x age interaction by multiplying column 2, column 3 and column 4 together
1 1 45 1 45
1 1 -33 -1 33
1 -1 22 -1 22
1 -1 -11 1 11
with contrast [0 0 0 0 1] or [0 0 0 0 -1]? In this case the direction of interactions would again be determined through descriptive statistics.
Sorry for the barrage of questions, I really appreciate your help!
I believe I'm following you but im struggling slightly with the interpretation. So following the example again(column 1= intercept, column 2=group, column 3=age, column 4=sex, column 5=group x sex) if I created the following design matrix
1 1 45 1 1
1 1 -33 -1 -1
1 -1 22 -1 1
1 -1 -11 1 -1
(Male =1, Female =-1 for this example). And then ran a t-test with the contrast [0 0 0 0 1].
Am I correct in understanding that if this uncovers significant networks, this would indicate higher connectivity in males/lower connectivity in females for group 1 and/or the inverse relationship for the group assigned -1? Therefore I would look to descriptive statistics to define the directions and relations in greater detail?
Lastly could I take this one step farther and look for a group x sex x age interaction by multiplying column 2, column 3 and column 4 together
1 1 45 1 45
1 1 -33 -1 33
1 -1 22 -1 22
1 -1 -11 1 11
with contrast [0 0 0 0 1] or [0 0 0 0 -1]? In this case the direction of interactions would again be determined through descriptive statistics.
Sorry for the barrage of questions, I really appreciate your help!