help
help > RE: Contrast Interaction Group x Time
Jan 7, 2017 11:01 PM | Andrew Zalesky
RE: Contrast Interaction Group x Time
Hi Tom,
The contrast you give is not quite right to test for an interaction.
To test for an interaction, you need to add a new column to your design matrix, which is the multiplication of the group and time columns. So the new columns would be 0 1 0 1 0 0 0 0. You will then need to remove the group (first) column. The design matrix will be rank deficient if you don't remove the group column.
The contrast is then just: [... 0 0 0 1 0 0 0 ...]. That is, zeros everywhere, except a 1 for the column comprising the multiplication of the group and time columns.
I suggest using F-test instead of t-test when evaluating the interaction. t-test is perfectly fine, but in the case of a t-test, you should test both 1 and -1 in the contrast.
Andrew
Originally posted by tom parker:
The contrast you give is not quite right to test for an interaction.
To test for an interaction, you need to add a new column to your design matrix, which is the multiplication of the group and time columns. So the new columns would be 0 1 0 1 0 0 0 0. You will then need to remove the group (first) column. The design matrix will be rank deficient if you don't remove the group column.
The contrast is then just: [... 0 0 0 1 0 0 0 ...]. That is, zeros everywhere, except a 1 for the column comprising the multiplication of the group and time columns.
I suggest using F-test instead of t-test when evaluating the interaction. t-test is perfectly fine, but in the case of a t-test, you should test both 1 and -1 in the contrast.
Andrew
Originally posted by tom parker:
Dear NBSers,
I have 2 groups of subjects (controls and patients), who were scanned at 2 time-points (baseline, after 4 years).
These 2 groups differ significantly in age and gender.
I would like to see if there is an interaction between group and time, while controlling for age and gender differences.
My hypothesis is that controls remain relatively stable, whereas patients show decreased connectivity over time.
This is an example of what my design matrix looks like:
Subjects Group Time Age Gender Regressor_Subj1 Regressor_Subj2 Regressor_Subj3 Regressor_Subj4
control01_baseline 1 0 50 1 1 0 0 0
control01_4years 1 1 54 1 1 0 0 0
control02_baseline 1 0 45 1 0 1 0 0
control02_4years 1 1 49 1 0 1 0 0
patient01_baseline 0 0 36 0 0 0 1 0
patient01_4years 0 1 40 0 0 0 1 0
patient02_baseline 0 0 34 1 0 0 0 1
patient02_4years 0 1 38 1 0 0 0 1
Does the contrast: 1 -1 0 0 0 0 0 0
allow me to test for Group x Time interaction?
Thank you so much!
I have 2 groups of subjects (controls and patients), who were scanned at 2 time-points (baseline, after 4 years).
These 2 groups differ significantly in age and gender.
I would like to see if there is an interaction between group and time, while controlling for age and gender differences.
My hypothesis is that controls remain relatively stable, whereas patients show decreased connectivity over time.
This is an example of what my design matrix looks like:
Subjects Group Time Age Gender Regressor_Subj1 Regressor_Subj2 Regressor_Subj3 Regressor_Subj4
control01_baseline 1 0 50 1 1 0 0 0
control01_4years 1 1 54 1 1 0 0 0
control02_baseline 1 0 45 1 0 1 0 0
control02_4years 1 1 49 1 0 1 0 0
patient01_baseline 0 0 36 0 0 0 1 0
patient01_4years 0 1 40 0 0 0 1 0
patient02_baseline 0 0 34 1 0 0 0 1
patient02_4years 0 1 38 1 0 0 0 1
Does the contrast: 1 -1 0 0 0 0 0 0
allow me to test for Group x Time interaction?
Thank you so much!
Threaded View
| Title | Author | Date |
|---|---|---|
| tom parker | Jan 7, 2017 | |
| Andrew Zalesky | Jan 7, 2017 | |
| tom parker | Jan 9, 2017 | |