I am trying to evaluate a design with 4 conditions. two of the conditions will occur 32 times and two of the conditions will occur 12 times per run. There will be a total of three runs. I would also like to present the stimuli at sub-TR intervals to increase the sampling of the hemodynamic response. there will be 225 images per run at a TR of 2000 ms. I want to be able to sample at 1 second intervals. I used thefollowing syntax to generate my response functions to test the design. However, I noticed a few things that I did not understand. First, it appears to me that all of the even and odd coefficients are closely correlated to one another. For example, h0 and h2 and h4 all seem to have similar values whereas h1, h3, and h5 all seem to have similar values. In the following examples this difference can be substantial (especially for the 12 trial condition). I reran RSFGEN 3 times with different random seeds. Why are odd coefficients correlated with odds and evens with evens? What am I doing wrong?
FIRST RUN
RSFgen \
-nt 450 -num_stimts 4 -seed 957821451 -one_file -prefix 1RUN2ptr2\
-nreps 1 32 -nreps 2 12 -nreps 3 32 -nreps 4 12
3dDeconvolve \
-nodata \
-num_stimts 4 \
-stim_file 1 "1RUN2ptr2.1D[0]" -stim_maxlag 1 5 -stim_label 1 V1 -stim_nptr 1 2 \
-stim_file 2 "1RUN2ptr2.1D[1]" -stim_maxlag 2 5 -stim_label 2 I1 -stim_nptr 2 2 \
-stim_file 3 "1RUN2ptr2.1D[2]" -stim_maxlag 3 5 -stim_label 3 V8 -stim_nptr 3 2 \
-stim_file 4 "1RUN2ptr2.1D[3]" -stim_maxlag 4 5 -stim_label 4 I8 -stim_nptr 4 2 \
-xout > 1RUN2ptr2
RESULTS
Stimulus: V1
h[ 0] norm. std. dev. = 0.2413
h[ 1] norm. std. dev. = 0.3303
h[ 2] norm. std. dev. = 0.2394
h[ 3] norm. std. dev. = 0.3249
h[ 4] norm. std. dev. = 0.2476
h[ 5] norm. std. dev. = 0.3220
Stimulus: I1
h[ 0] norm. std. dev. = 0.5063
h[ 1] norm. std. dev. = 0.4645
h[ 2] norm. std. dev. = 0.4891
h[ 3] norm. std. dev. = 0.4354
h[ 4] norm. std. dev. = 0.4532
h[ 5] norm. std. dev. = 0.4334
Stimulus: V8
h[ 0] norm. std. dev. = 0.3011
h[ 1] norm. std. dev. = 0.2530
h[ 2] norm. std. dev. = 0.3002
h[ 3] norm. std. dev. = 0.2488
h[ 4] norm. std. dev. = 0.2969
h[ 5] norm. std. dev. = 0.2461
Stimulus: I8
h[ 0] norm. std. dev. = 0.4621
h[ 1] norm. std. dev. = 0.4080
h[ 2] norm. std. dev. = 0.4669
h[ 3] norm. std. dev. = 0.4081
h[ 4] norm. std. dev. = 0.4790
h[ 5] norm. std. dev. = 0.4082
SECOND RUN
RSFgen \
-nt 450 -num_stimts 4 -seed 986427531 -one_file -prefix 1RUN2ptr2\
-nreps 1 32 -nreps 2 12 -nreps 3 32 -nreps 4 12
RESULTS
Stimulus: V1
h[ 0] norm. std. dev. = 0.2916
h[ 1] norm. std. dev. = 0.2708
h[ 2] norm. std. dev. = 0.2991
h[ 3] norm. std. dev. = 0.2723
h[ 4] norm. std. dev. = 0.3008
h[ 5] norm. std. dev. = 0.2739
Stimulus: I1
h[ 0] norm. std. dev. = 0.5240
h[ 1] norm. std. dev. = 0.3731
h[ 2] norm. std. dev. = 0.5272
h[ 3] norm. std. dev. = 0.3817
h[ 4] norm. std. dev. = 0.5203
h[ 5] norm. std. dev. = 0.3783
Stimulus: V8
h[ 0] norm. std. dev. = 0.2856
h[ 1] norm. std. dev. = 0.2656
h[ 2] norm. std. dev. = 0.2830
h[ 3] norm. std. dev. = 0.2556
h[ 4] norm. std. dev. = 0.2867
h[ 5] norm. std. dev. = 0.2598
Stimulus: I8
h[ 0] norm. std. dev. = 0.4568
h[ 1] norm. std. dev. = 0.4293
h[ 2] norm. std. dev. = 0.4475
h[ 3] norm. std. dev. = 0.4282
h[ 4] norm. std. dev. = 0.4566
h[ 5] norm. std. dev. = 0.4323
THIRD RUN
RSFgen \
-nt 450 -num_stimts 4 -seed 125347698 -one_file -prefix 1RUN2ptr2\
-nreps 1 32 -nreps 2 12 -nreps 3 32 -nreps 4 12
RESULTS
Stimulus: V1
h[ 0] norm. std. dev. = 0.2621
h[ 1] norm. std. dev. = 0.2941
h[ 2] norm. std. dev. = 0.2576
h[ 3] norm. std. dev. = 0.2864
h[ 4] norm. std. dev. = 0.2682
h[ 5] norm. std. dev. = 0.2896
Stimulus: I1
h[ 0] norm. std. dev. = 0.3993
h[ 1] norm. std. dev. = 0.4892
h[ 2] norm. std. dev. = 0.3981
h[ 3] norm. std. dev. = 0.4759
h[ 4] norm. std. dev. = 0.4058
h[ 5] norm. std. dev. = 0.4931
Stimulus: V8
h[ 0] norm. std. dev. = 0.2631
h[ 1] norm. std. dev. = 0.2760
h[ 2] norm. std. dev. = 0.2655
h[ 3] norm. std. dev. = 0.2765
h[ 4] norm. std. dev. = 0.2642
h[ 5] norm. std. dev. = 0.2789
Stimulus: I8
h[ 0] norm. std. dev. = 0.3809
h[ 1] norm. std. dev. = 0.5644
h[ 2] norm. std. dev. = 0.3835
h[ 3] norm. std. dev. = 0.5606
h[ 4] norm. std. dev. = 0.3863
h[ 5] norm. std. dev. = 0.5352
I also noticed that the estimates seemed better and did not seem to be correlated when I ran RSFGEN without trying to sample at sub-TR resolutions.
FIRST RUN
RSFgen \
-nt 225 -num_stimts 4 -seed 987654321 -one_file -prefix 1RUN1ptr1\
-nreps 1 32 -nreps 2 12 -nreps 3 32 -nreps 4 12
3dDeconvolve \
-nodata \
-num_stimts 4 \
-stim_file 1 "1RUN1ptr1.1D[0]" -stim_maxlag 1 5 -stim_label 1 V1 \
-stim_file 2 "1RUN1ptr1.1D[1]" -stim_maxlag 2 5 -stim_label 2 I1 \
-stim_file 3 "1RUN1ptr1.1D[2]" -stim_maxlag 3 5 -stim_label 3 V8 \
-stim_file 4 "1RUN1ptr1.1D[3]" -stim_maxlag 4 5 -stim_label 4 I8 \
-xout
Stimulus: V1
h[ 0] norm. std. dev. = 0.2071
h[ 1] norm. std. dev. = 0.2102
h[ 2] norm. std. dev. = 0.2120
h[ 3] norm. std. dev. = 0.2119
h[ 4] norm. std. dev. = 0.2141
h[ 5] norm. std. dev. = 0.2146
Stimulus: I1
h[ 0] norm. std. dev. = 0.3178
h[ 1] norm. std. dev. = 0.3138
h[ 2] norm. std. dev. = 0.3160
h[ 3] norm. std. dev. = 0.3180
h[ 4] norm. std. dev. = 0.3147
h[ 5] norm. std. dev. = 0.3144
Stimulus: V8
h[ 0] norm. std. dev. = 0.2136
h[ 1] norm. std. dev. = 0.2117
h[ 2] norm. std. dev. = 0.2098
h[ 3] norm. std. dev. = 0.2087
h[ 4] norm. std. dev. = 0.2100
h[ 5] norm. std. dev. = 0.2112
Stimulus: I8
h[ 0] norm. std. dev. = 0.3192
h[ 1] norm. std. dev. = 0.3190
h[ 2] norm. std. dev. = 0.3193
h[ 3] norm. std. dev. = 0.3260
h[ 4] norm. std. dev. = 0.3282
h[ 5] norm. std. dev. = 0.3257
SECOND RUN
RSFgen -nt 225 -num_stimts 4 -seed 957628431 -one_file -prefix 1RUN1ptr1 -nreps 1 32 \
-nreps 2 12 -nreps 3 32 -nreps 4 12
Stimulus: V1
h[ 0] norm. std. dev. = 0.2088
h[ 1] norm. std. dev. = 0.2099
h[ 2] norm. std. dev. = 0.2127
h[ 3] norm. std. dev. = 0.2134
h[ 4] norm. std. dev. = 0.2146
h[ 5] norm. std. dev. = 0.2122
Stimulus: I1
h[ 0] norm. std. dev. = 0.3200
h[ 1] norm. std. dev. = 0.3197
h[ 2] norm. std. dev. = 0.3180
h[ 3] norm. std. dev. = 0.3132
h[ 4] norm. std. dev. = 0.3125
h[ 5] norm. std. dev. = 0.3119
Stimulus: V8
h[ 0] norm. std. dev. = 0.2056
h[ 1] norm. std. dev. = 0.2060
h[ 2] norm. std. dev. = 0.2032
h[ 3] norm. std. dev. = 0.2061
h[ 4] norm. std. dev. = 0.2108
h[ 5] norm. std. dev. = 0.2101
Stimulus: I8
h[ 0] norm. std. dev. = 0.3116
h[ 1] norm. std. dev. = 0.3116
h[ 2] norm. std. dev. = 0.3100
h[ 3] norm. std. dev. = 0.3114
h[ 4] norm. std. dev. = 0.3112
h[ 5] norm. std. dev. = 0.3142
Based on this analyses it would seem as though it would be better to not sample at sub-TRs. This seems to be a trade-off issue increased temporal resolution but a more "variable" estimate. Is this accurate?
My other questions deal with constraints on the design. If I do sample at sub-TR levels is it better to ensure that half of the trials start with the TR and the other half start exactly one second later to get optimal
sampling? Or should I totally randomize the design? Relatedly, if I want to use a constraint like at least one "blank" trial between every condition is there a way to totally randomize the design? I saw on the website to use something like the following syntax
RSFgen \
-nt 450 -num_stimts 4 -seed 987654321 -one_file -prefix TEST2 \
-nreps 1 32 -nreps 2 10 -nreps 3 32 -nreps 4 1 \
-nblock 1 2 -nblock 2 2 -nblock 3 2 -nblock 4 2
but that places two ones in the column and I would have to go and edit one of them out depending on whether half of the trials should start mid-TR (the first constraint?).
Sorry for the long post and thanks in advance for the help!!!