On 2/24/12, Paola Valsasina <[log in to unmask]> wrote:
> Dear Donald,
> This thread that wasn't generated by me, but I am interested in this
> question as well, in particular to question 1) from Stefania.
> I always have the doubt on how to introduce categorical covariates when I
> perform second level analyses in SPM.
> Suppose e.g. that I want to compare GM between controls and patients,
> accounting for sex (categorical variable).
> Is it better that I do a full factorial design with a single factor
> (controls/patients) and I add as covariate a gender vector all made by
> zeroes and ones, or that I do a full factorial design with two factors
> (controls/patients and males/females), insert in four separate cells male
> controls, female controls, male patients, female patients, and then perform
> the comparison between controls and patients by setting up a contrast like 1
> 1 -1 -1?
The two models that you described are quite different. If you test the
group differences in the first model, then you are testing the
difference in controls males and patient males (e.g. the y-intercept
of each group). You have assumed that the effect of being female is
the same in both groups.
In model 2, you allow the the effect of gender to be different in both groups.
I'll also suggest another model, add a single covariate of 1s and -1s
to the model with patients and controls. This model will then test
whether the groups - if they were balanced for males/females -- are
different from each other. This assumes the effect of being
male/female is the same for both groups. This can be thought of as
covariate-adjusted group means.
> Which is the difference between these two models? Are they both correct, or
> is one better than the other?
Depending on the assumptions that you want to make and what you would
like to test, different models would be most appropriate. Model 2
would allow you to also test for the interaction of group and gender,
which might be preferable. If there is a reason why there is an
imbalance in gender between males/females, then model 3 could be
problematic because you are testing whether or not the groups are the
same if the genders were balanced. My personal preference would be
The most important thing in the process is the transparency. You need
to clearly state the model that you test and the interpretation of the
model -- whether it is testing the group means, the mean of the males,
or the mean of the group as if the genders were balanced
(covariate-adjusted group means).
Hope this helps.
> Kind regards
> -----Messaggio originale-----
> Da: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] Per
> conto di MCLAREN, Donald
> Inviato: venerdì 24 febbraio 2012 15.10
> A: [log in to unmask]
> Oggetto: Re: [SPM] Use of covariates
> The answer depends on the model. If the models are the same, then
> programs will produce the same results. If the models are different,
> then the results will be different.
> The General Linear Model will always produce the same results!!! The
> differences in results is due to different models.
> Since you are using a full factorial, I assume that this is a between
> subject design. To get the same results as SPSS, you will want to set
> the variance to be equal between levels of your factors. This will
> suppress the variance correction in SPM.
> (1) All columns are treated the same way. If you have 4 columns for
> ethnicity in SPSS, then you need 4 columns in SPM. If you have one
> column in SPSS, then you need 1 column in SPM. When you say you have a
> ordinal variable in SPSS, then it treats it as N variables (each
> variable represents one group). You can also verify this by creating
> the dummy variables yourself. There are a few variations on the model
> that are equivalent (see examples in FSL).
> (2) Yes. You believe that the there variables account for variance in your
> (3) Generally speaking, yes. However, in brain data it is more
> complicated because not every covariate will be significant in every
> voxel. My solution to this is to keep the variable in the model if
> there are any significant effects of the variable.
> On 2/24/12, Stefania Tognin <[log in to unmask]> wrote:
>> Dear SPM’ers,
>> I have few questions regarding the use of covariates in the statistical
>> analysis with SPM. It is not clear to me if covariates in SPM are treated
>> the same as if using a stats software (e.g. SPSS).
>> I want to perform an Full factorial ANOVA with SPM on structural data (1
>> factor 3 levels) and I want to control for (“remove”) the effect of age
>> (continuous variable), gender (categorical, 2 levels), handedness
>> (categorical, 3 levels) and ethnicity (categorical, 4 levels).
>> 1) Under what assumptions is it correct to use categorical nuisance
>> variables (e.g. gender, handedness) given that using stats software as for
>> example SPSS, covariates should be continuous and not categorical?
>> 2) If variables as for example age, gender or handedness do not differ
>> across groups it is still correct to remove their effects, assuming that
>> being right-handed, left-handed or ambidextrous (or male and female, older
>> or younger) could result in brain differences?
>> 3) To justify the use of a covariate should I use statistical criteria
>> theoretical criteria that take into account the fact that we are working
>> with the human brain?
>> Thank you very much and I do apologize if they sound very basic questions.
> Best Regards, Donald McLaren
> D.G. McLaren, Ph.D.
> Postdoctoral Research Fellow, GRECC, Bedford VA
> Research Fellow, Department of Neurology, Massachusetts General Hospital
> Harvard Medical School
> Website: http://www.martinos.org/~mclaren
> Office: (773) 406-2464
> This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
> HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
> intended only for the use of the individual or entity named above. If the
> reader of the e-mail is not the intended recipient or the employee or agent
> responsible for delivering it to the intended recipient, you are hereby
> notified that you are in possession of confidential and privileged
> information. Any unauthorized use, disclosure, copying or the taking of any
> action in reliance on the contents of this information is strictly
> prohibited and may be unlawful. If you have received this e-mail
> unintentionally, please immediately notify the sender via telephone at
> 406-2464 or email.
Best Regards, Donald McLaren
D.G. McLaren, Ph.D.
Postdoctoral Research Fellow, GRECC, Bedford VA
Research Fellow, Department of Neurology, Massachusetts General Hospital
Harvard Medical School
Office: (773) 406-2464
This e-mail contains CONFIDENTIAL INFORMATION which may contain PROTECTED
HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which is
intended only for the use of the individual or entity named above. If the
reader of the e-mail is not the intended recipient or the employee or agent
responsible for delivering it to the intended recipient, you are hereby
notified that you are in possession of confidential and privileged
information. Any unauthorized use, disclosure, copying or the taking of any
action in reliance on the contents of this information is strictly
prohibited and may be unlawful. If you have received this e-mail
unintentionally, please immediately notify the sender via telephone at
406-2464 or email.