Hi Todd,
I'm a bit confused about temp derivatives as well. Please do correct me if i'm wrong but
this is what i understood correctly or incorrectly from current and previous posts.
Using temporal derivatives will increase the statistics of the main EV by correcting
SLIGHT mismatches. the derivative will regress out mismatch so that the unexplained
data doesnt creep into the residuals and decrease the z-stats. it wont affect the PE value
of the main EV. However,your PE value and hence z-scores will be reduced if there is a
quite a bit of shift between the model and response.
So , we can take the information from the temporal derivative to construct a better
model. if there is a mismatch , then you can see it in the timeseries and peristimulus
plots in the 1st level FEAT report. i.e look at the difference in shape of the full model fit
and Main EV model fit. You can also , as Vince suggested, combine the Main EV and
derivative in the form of an f-stat. The plots should again show you an adjusted fit. But
you cant do a group analysis in GLM using f-test copes, which is what Vince and Eugene
stated. As an alternative you can create Latency maps as in the henson paper. You can
also specify the derivative contrast to be calculated along with the Main EV. This can
show the regions where there is a mismatch(latency maps can show you where and how
much) ,which might be interesting eg: differences in timing between primary and
secondary areas of the brain. You can also feed the individual temp derivative images to
the 2nd level. I'm not sure about the information that the group mean derivative image
gives but may be it represents the consistent mismatch between model and stimulus
timing.
-Vish
|