# Does random effects allow me to do repeated measures?

I am analysing some behaviour data from several participants in Matlab. I used to do "repeated measures" in SPSS.

But is this the same as adding "subject" as a factor, and choosing it as a random effect?

i.e. my normal repeated measures data matrix x looks like this

                 condition A      condition B
subject 1          x(1,1)           x(1,2)
subject 2          x(2,1)           x(2,2)
...                 ...              ...
subject 7          x(7,1)           x(7,2)


so for Matlab, I tried

COND = [1 2                   SUBJ = [1 1
1 2        and                2 2
...                           ...
1 2]                          7 7]


then ran

anovan(x(:), {COND(:), SUBJ(:)}, 'random',2 )

Now, this does not seem to give me the same results as SPSS, for repeated measures... Perhaps I have misunderstood the 'random' parameter - I thought you could construct a "repeated measures" test from it?

-
 Please do not cross-post here and on SO. – mbq♦ Nov 28 '11 at 17:25 sorry @mbq, someone told me I should post here but didn't know how to migrate questions – Sanjay Manohar Nov 28 '11 at 17:35 Ok; for the future, you can ask mod to do this by flagging question. – mbq♦ Nov 28 '11 at 17:38

But is this the same as adding "subject" as a factor, and choosing it as a random effect? Yes. These two SPSS codes gives identical results for the RM-factor "cond":

/*Wide data structure: cond as 2 variables*/.
glm cond_A cond_B /wsfactor= cond 2 /wsdesign= cond.

/*Long data structure: dependent var score and factor cond with 2 levels*/.
glm score by cond subject /random= subject /design= cond subject.


I can't speak of MATLAB.

P.S. Long data structure is like this:

subject cond score
1       A    1.2
1       B    0.5
2       A    3.1
2       B    1.4
3       A    0.8
3       B    1.8
...

-
 Thanks! That is helpful. But what about interaction terms e.g. cond x subject? should they included or excluded from the regression? The funny thing is matlab reports main effect of subject p<0.00001 - do I just discard this as irrelevant, or should I somehow stop matlab from making this contrast? – Sanjay Manohar Nov 29 '11 at 11:52