Thank you Matthew and Gwenaelle,
I have been trying to figure how to set up an F-test in Glm that would
test the omnibus question of whether there is an interaction between
time and group in a 2x3 design. I understand that I need to set up
some contrasts, set the F button to 1, then click on the buttons next
to the contrasts that I want to include in the F-test. But I have no
idea what contrasts should be included in such a design. Any
suggestions.
Thanks again,
Omar
On Thu, Jan 28, 2010 at 12:11 PM, Matthew Webster
<[log in to unmask]> wrote:
> Hello,
> If you are using the group column in the GLM to set up a design.grp
> file for use with the -e option in randomise, you can safely ignore
> this message.
>
> Many Regards
> Matthew
>> Thank you for your response, it is very re-assuring.
>>
>> With regards to the error messages you said:
>>
>>>It is probably because you've run exactly the model above that is
>>>indeed rank deficient. You just need to add some subjects (and
>>>presumably you've got more than one control) as some additional EVs and
>>>it should not be a problem anymore...
>>
>> Actually, I have run this model in Glm using 24 controls, 15 abstinent
>> subjects, and 5 relapsers. That is when the error message appears. So,
>> adding the EVs in the same way I did with the example dataset does not
>> seem to resolve this error message. Do you have other thoughts about
>> why this could be happening?
>>
>> Thank you so much.
>> Omar
>>
>>
>>
>>
>>> Problem with processing the model. Warning: at least one EV is (close
>>> to) a linear combination of the others. You probably need to alter
>>> your design. (Design matrix is rank deficient - ratio of min:max
>>> eigenvalues in SVD of matrix is 1.21792e-17).
>>>
>>> Warning-design matrix uses different groups (for different variances),
>>> but these do not contain "separable" Evs for the different groups (it
>>> is necessary that, for each EV, only one of the groups has non-zero
>>> values).
>>
>> Hope this helps,
>> Gwenaelle
>>
>>
>
|