We are performing site maintenance. Some features may not not be available at this time.

Dear Joaquim & Anton

I'd like to generate a mean map and comparisons which control for activation associated with a covariate. I'm not sure whether these are in the output I've already generated but I'm not sure which ones they are or whether there is something else I need to do. 

Specifically, I have two groups of tasks but the control conditions range from rest to high level so I'd like to include this as a covariate of no interest in the mean maps for each group separately and comparison between them.

I've run a comparison between groups with a covariate but couldn't tell whether any of the input controlled for the covariate. I have the same issue with interpreting the multiple regression output files which I've tried running with "group" as a binary predictor and my covariate (varies between 1 and 4) as another predictor.

Thanks in advance for your help,
Jo

Hi Jo,

I have a similar problem; I am running a group analysis with a binary independent and a binary covariate (of no interest). A similar question has been asked before on this forum and from that I gather that if you want to look at a group difference controlling for your covariate, where your group IV is the first variable and covariates second, third etc then you should look at:

10m00 (in the case of 1 covariate). 

Best, 

Jurriaan

Hi Jurriaan

Thanks for your reply. I think I have seen the forum post you mentioned but do you have a link to it / remember the title please just in case I missed it.

I wasn't confident whether the output map you mentioned represents the effect of the variable controlling for the covariate or simply the main effect of the variable, ignoring the covariate (mine is 4 ordered groups). If anyone could confirm what each of the maps represent I would really appreciate it as I don't want to misinterpret.

Many thanks,
Jo

https://www.nitrc.org/forum/message.php?...

I have run analyses with and without the covariate and the results are different for 1m0 (without covariate) as compared to 10m00 (with covariate) - hence in the latter case I think the effect of the covariate is controlled. I also tested it by entering the same dummy variable as group variable and as covariate in the same analyses (then you expect to see no significant voxels, which was confirmed). 

As I understand it the following model is fitted  (in the case of 1 binary group variable and 1 binary covariate):

SDMz = c + b1*D1 + b2*D2

Where c is intercept and b's are partial regression coefficients of categorical variables (ie. D1, D2; one controlled for the other). 10m00 then means: difference in SDMz caused by first variable (setting D1 to 1 vs 0; so the first position before and after m) while controlling for the second dummy (second position before and after m: always 0). 

I would actually also appreciate it if an expert could confirm whether what I concluded above is correct. 

Best

Dear Jo and Jorge,

first of all, thank you for your great participation in the forum.

Your interpretation is correct: 10m00 means the effect of the first moderator (in this case, group comparison) controlling for the second moderator (in that case, the covariate of no interest).

Hope this helps,

Joaquim

Thanks so much for your reply Joaquim.

Does running a linear model comparing groups with a covariate also give the mean analysis for each group separately, controlling for the covariate?

If so am I right in thinking these are:

name_10_z - group coded as 1 controlling for the covariate
name_00_z - group coded as 0 controlling for the covariate

Many thanks,
Jo

Dear Jo,

Specifically,

name_10_z - group coded as 1, when the covariate is 0 (or minimum)
name_00_z - group coded as 0, when the covariate is 0 (or minimum)

I'm sorry for this confusion,

Joaquim

Hi all,

ANCOVAs are always a bit confusing for me. Thus, I would like to clarify some additional aspects.

You state that:


However, this description has a substantially different meaning than "group coded as 1, when the covariate is controlled for". For instance, I would like control for the mean age of study participants. Am I right to assume that in a scenario with the values of: 18, 28, 28, 29, 32, 36, 80 years, the contrast "name_10_z" reflects the "group coded as 1, when the mean age is 18"?

How would one proceed to actually control for age-related aspects? And how would I best deal

Thank you very much for the clarification, and Merry Christimas,
Marco

Dear Marco,

I apologize for the confusing syntax, I hope to make it easier in the future...

Your assumption is right: "name_10_z" reflects the "group coded as 1, when the mean age is 18". For current SDM, the contrast of interest is assumed to be "10m00", which would reflect the difference between the group coded as 1 and the group coded as 0, when their mean age is the same (18, or any other). Unfortunately, it is not possible to contrast "group coded as 1, when the mean age is the mean age of all samples".

Hope this helps,

This makes perfect sense now.

Thank you very much!