I believe there should be a choice for forming an interaction between
the covariate and the factor of interest. This will split the
covariate between the levels of your medication condition so you will
end up with 5 colulmns. You should then be able to make a contrast of
[0 0 0 -1 1] which would look at the effect of the covariate on
different levels of medication.
-----
Darren Gitelman
On Fri, Dec 5, 2008 at 5:03 AM, Andreas Hahn
<[log in to unmask]> wrote:
> Dear Jonathan and SPMers,
>
> Yes, that answers my first question (and I assume it is the common way how
> to include a covariate into an ANOVA). Thanks!
>
> However, I'm still confused how to set the covariate if I do NOT want to
> correct for its effect (e.g. to include it like a regressor variable):
> If my contrast is
> -1 1 0 1 (instead of -1 1 0 0)
> I get the difference between my first two medications - but how to interpret
> the effect of my covariate?
> Is it then correct to say that the difference between the medications is
> modulated by the covariate?
>
> From the mailing list archive (Andrew Holmes, 6th June 2000, "Covariate
> correlations") I found out that a correlation between condition and
> covariate is calculated with a contrast like
> 0 0 0 1
> However, in my case I want to see the DIFFERENCE between the two medications
> in dependency of my covariate, but setting the condition to 0 does not
> depict the difference, does it?
>
> Thanks for your help,
> Andreas.
>
>
> Jonathan Peelle schrieb:
>>
>> Hi Andreas
>>
>>
>>>
>>> For example, the covariate is age and I believe that age differences
>>> between
>>> subjects would influence my results. Is the covariate then set 0, i.e.
>>> does
>>> this mean that the results are corrected for age?
>>>
>>
>> Any covariates you enter have their effect when you estimate your
>> model. If you have age as a covariate, the variability in your data
>> that can be attributed to age effects should be accounted for by this
>> covariate, (hopefully) leaving your non-age regressors (in this case,
>> medication group) free of age effects.
>>
>> What contrast you look at is independent of how you estimated your
>> model. In your case, since age was in the model, looking at a
>> contrast between medication groups have you have suggested
>>
>> 1 -1 0 0
>>
>> should give you the difference between groups after age has been
>> accounted for, which is what you want.
>>
>> Hope this helps,
>>
>> Jonathan
>>
>>
>>
>
> * Original message *
>
> Dear SPM experts!
>
> I have a question regarding the contrast setting of the covariate in
> an ANOVA.
>
> My ANOVA design has 1 factor including 3 levels (3 different
> medications) and 1 covariate.
> So if I want to calculate the difference between medication group 2
> against group 1 I would set the contrast as
> -1 1 0 0
> However, I don't really understand how to set the covariate or how
> to interpret the result...
> For example, the covariate is age and I believe that age differences
> between subjects would influence my results. Is the covariate then
> set 0, i.e. does this mean that the results are corrected for age?
> And more importantly, if the covariate has an effect which I do NOT
> want to correct for but take its effect into account within the
> results, would the conrast than be
> -1 1 0 1
> ?
>
> Thanks in advance,
> Andreas.
>
>
>
|