I agree with Paul. Another possibility might be if the right
hemisphere activation was very variable in the controls (I have no
idea how likely this is though), since this could firstly explain the
lack of significance in the controls test, and secondly, due to a
pooled variance estimate being used, the lack of a significant
difference between groups. But I think Paul's answer is more likely
the case, unless you check and find that actually the t-values for
controls' right hemisphere are in fact well below the threshold.
An interesting paper which talks about issues like this is Jernigan et
al, 'More "mapping" in brain mapping...'
Sophie Lafaille wrote:
> using SPM2 I'm looking at some random effects analyses.
> One sample T-test
> when looking at task X within a group of 15 controls, we get very
> significant left hemisphere activation with T-values of 10-14 (0.05 FWE)
> when looking at task X within a group of 15 patients, we get very
> significant bilateral activations with T-values of 9-15 (0.05 FWE)
> when I enter all these con files into a 2 sample t-test to look at
> controls vs. patients for task X I get nothing at a corrected level and
> only start seeing the pattern of activation that makes sense at a 0.05
> see attached, I've shown one sample data on top, and then the
> subtraction on bottom.
> am I doing something wrong or is the subtraction of either controls from
> patients or vice-versa removing a lot of the right hemisphere activation
> that I'm looking for and can only been seen at a much much lower
> uncorrected p-value.
> thank you, sophie
> Sophie Josee Lafaille, B.Sc.(Hons.), M.Sc.
> Research Officer II, Speech Fluency Laboratory 1059
> Department of Speech-Language Pathology, University of Toronto,
> 500 University Avenue, Rm 160, Toronto, Ontario, M5G 1V7
> Phone: 416-946-8635 Fax:416-978-1596
> We don't stop playing because we grow old, we grow old because we stop
> - Anonymous