Hi,
These are quite general and tricky questions, which is probably why no-one has yet replied.
I would say that generally you want the second design if you are "correcting for" the different covariates, since you want them to be able to explain (correct for) group differences in a straightforward way and not have different/disconnected slopes/intercepts per group.
I cannot say anything general and definitive about statistical power myself.
However, there is a potential for problems with the relatedness of the subjects (siblings and non-sibling groups), especially given the different numbers of subjects in the groups, but I'm not entirely sure what the best approach would be for this.
Maybe someone else on the list could make a suggestion here?
I hope this helps.
All the best,
Mark
On 14 Apr 2013, at 09:04, Alex <[log in to unmask]> wrote:
> Hello List,
>
> We are working a project to explore the changes in OCD and their siblings. There are total 69 subjects of 28 controls, 26 patients and 15 siblings of patients. I did comparing each pair of the three groups as design_subs69.png.
>
> Currently, I want a unified ANCOVA design as design_subs69_ancova.png.
>
> My question is what is the difference in statistical power and outcomes given by FEAT commands?
>
> I know there would be some questions on setting models for this sample regarding the ocd-sibs closeness than that of ocd-con in generation, do you have more suitable approaches for addressing this issue? Or what is the common way of dealing with patients-sibs-controls data mining in this field?
>
> Any of your inputs are greatly appreciated!
>
> Best Regards,
> Xi-Nian Zuo
> <design_subs69.png><design_subs69_ancova.png>
|