It is difficult to get good estimates of an FIR model from a block design, as I also said in response to you other email. However, I would expect to get better results in the event-related data, although you only have 12 trials, so it might be that you are still lacking sufficient statistical power to get good estimates. Does it improve if you have less FIR functions (e.g., 10 functions, each one 2 seconds long)?
As I mentioned in the other email, you might do a lot better with a more restricted basis for your HRF, such as a 3 component basis from FLOBS. This will definitely increase the quality of the estimates.
All the best,
> On 25 Aug 2017, at 19:47, Yuqi Liu <[log in to unmask]> wrote:
> Dear FSL experts,
> I have been playing with FIR modeling method in FSL for our data from multiple experiments, but I always get weird peristimulus plot however I set up the analysis.
> In the first experiment, we delievered vibrotactile stimuli to subjects in an 30-seconds on, 30-seconds off block design, repeating the on-off cycle for 4 times within each run. I tried FIR multiple times with different time windows (12 seconds, 24s, 30s, 60s) and number of basis function (given 3-second's TR). However, after concatenate the PE files in order and extract time series from a mask, I always get sawtooth plot that jumped between positive and negative values from time points to time points, which doesn't make sense to me.
> In the second experiment, we deliever vibrotactile stimuli to subjects in a more event-related manner: 3-seconds on, 18~20-seconds off, for 12 trials in each run. I modelled HRF with 20 FIR functions spanning 20 seconds (TR=1s), expecting a good-looking HRF given more separated trials. However, after concatenating the 20 pe files in order, I still got sawtooth-shaped plot.
> I am very confused by this, does anyone has any clue on why this would be?
> Thank you very much,