Show all posts by user
Dear AFNI users-
We are very pleased to announce that the new AFNI Message Board framework is up! Please join us at:
https://discuss.afni.nimh.nih.gov
Existing user accounts have been migrated, so returning users can login by requesting a password reset. New users can create accounts, as well, through a standard account creation process. Please note that these setup emails might initially go to spam folders (esp. for NIH users!), so please check those locations in the beginning.
The current Message Board discussion threads have been migrated to the new framework. The current Message Board will remain visible, but read-only, for a little while.
Sincerely,
AFNI HQ
History of AFNI updates
Results 2521 - 2550 of 2632
Nic,
Based on the scaling factors, it seems you would have holes due to truncations for values around 0.001 or below. If you suspect something else is going on, you can upload all the data plus the 3dMEMA script, and I'll take a close look.
The intermediate ORIG file is due to an ancient write function, but that's just a temporary step about the file name, and would not have any e
by
gang
-
AFNI Message Board
Nic,
3dMEMA currently outputs files in shorts only. As all the values in the output are supposed to be bounded within [-100, 100], shorts should be good enough. So you are seeing holes in your output? Could you paste the output of the following command?
3dinfo -verb TheOutputFile
by
gang
-
AFNI Message Board
You may consider using 3dLME.R in this case, which requires that you install R (and package 'snow' if you want parallel computing). Your case is covered in the second half of the following page. Add a 'Group' column in the table and use the following two lines:
Model:Group:Time-1
...
SS:marginal
by
gang
-
AFNI Message Board
Focusing on the data structure such as ANOVA is not necessarily the only nor the ideal perspective when running FMRI group analysis. If you provide a clearer picture of your situation, a better solution may be available: Could you list all the statistical tests (contrasts, main effects, interactions, etc.) you would like to run at group level?
by
gang
-
AFNI Message Board
Since your situation is not what I originally thought, you need to abandon the approach to analyzing the effects for quadratic fitting and behavioral measure. Instead use 3dLME.
by
gang
-
AFNI Message Board
I thought the eight blocks were under one condition, but your most recent response seems to indicate that my original assumption was wrong. Then you may just try out 3dLME by using the two columns in your 3dRegAna script as two covariates in 3dLME.R with a model like quadratic + behavioral measure.
by
gang
-
AFNI Message Board
3dRegAna can't model the random effects (the deviation of each subject from the group average). Two approaches:
1) Go back and run individual subject analysis with amplitude modulation (option AM2) in 3dDeconvolve: simultaneous modulations with both quadratic and behavioral measure effects. And go to group level.
2) Use 3dLME directly at group level.
by
gang
-
AFNI Message Board
Marwa Wrote:
-------------------------------------------------------
> In preparing the interaction term for a PPI
> analysis, does it matter which condition you
> assign '1' and which you assign '-1'? I.E., does
> the interpretation remain the same whether A and B
> conditions are labeled 1 & -1 or -1 &1?
Essentially it does not matter as long
by
gang
-
AFNI Message Board
Two approaches:
1) If you have Matlab plus its Statistics toolbox, you can use GroupAna:
2) List all the individual tests you'd like to get. Some of them can be easily analyzed with 3dttest++ or 3dMEMA. The only part that is complicated is the two-way interactions: Get the difference between before scan and after scan, and then dummy code the three groups as covariates and use 3dtt
by
gang
-
AFNI Message Board
> what I want is really the regression effects, but there might be a confound
> from the behavioral measures, so I would like to run the regression and
> remove the impact of the behavioral measures on the regression.
I'm not so sure what you mean by removing the effect of the behavioral measures. If you put a variable in a model, the effect from that variable is accounted fo
by
gang
-
AFNI Message Board
> I would like to remove a covariate of behavioral measures from a quadratic regression.
Do you mean that you want to see the impact of the behavioral measure? If so, you can do
-model 1 : 0 2
However, this is not really necessary because the t-statistic for the behavioral measure is basically what you want, which is just the square root of the F-statistic of the above option.
by
gang
-
AFNI Message Board
3dMEMA might be able to account for the impact of unequal number of trials on the reliability of each subject's effect estimate, but the impact of unequal number of trials should not lead to any difference on the directionality of the contrast at group level.
by
gang
-
AFNI Message Board
> I have done t-tests using the beta values as the input for each participant and
> the resulting t-stats are higher for the 60 trial condition than the 20 trial condition?
I assumed that your OP was about individual subject analysis. Are you actually talking about group analysis?
by
gang
-
AFNI Message Board
Jackie,
I'm not so sure what you mean by 'higher activation': higher beta value or t-statistic? Generally speaking, with a bigger sample size (number of trials or subjects), you would get more reliable results, but the effect estimates would remain unbiased. In other words, more trials should lead to little changes in the effect estimates (beta values), but the t-statistic value
by
gang
-
AFNI Message Board
dargonchow Wrote:
-------------------------------------------------------
> Can we use 3ddot or other AFNI program to calculate the phi coefficient between two
> binary maps ?
Not 3ddot. First, you need to get the 2 x 2 table (see ) at each voxel using, for example, 3dcalc. Then you can easily get the Phi value with the formula at that website with 3dcalc.
by
gang
-
AFNI Message Board
By Bonferroni correction I assume you meant family-wise error (FWE) correction. Take a look at
3dClustSim -help | less
by
gang
-
AFNI Message Board
Hi Paul,
> Could i use these two t maps from separate analysis approaches with 3ddot to
> investigate the relationship between GM structure covariance and GM atrophy ?
3ddot with its option '-demean' does seem to fit your scenario. However, I would avoid computing the correlation with t-statistic values, if possible. Instead some direct measure that carries some physical
by
gang
-
AFNI Message Board
kati Wrote:
-------------------------------------------------------
> Hi Gang,
>
> Thanks for the tip. Just to be sure, I only need to add the line:
>
> -concat concat.1D
>
> to 3dDeconvolve? (where concat.1D contains the line: 0 80 160 240) (these are the onset timepts
> of each run). In the documentation for 3dREMLfit, there doesn't seem to be any option
by
gang
-
AFNI Message Board
Sam,
A couple of questions for you:
1) What program did you use to obtain the "functional connectivity maps"?
2) What do you want to use the z-score values at regional level for?
by
gang
-
AFNI Message Board
Hi Kati,
As long as you let 3dDeconvolve/3dREMLfit know those discontinuities, e.g., through -concat, the program does not consider the serial correlation at those disjointed time points. If you want to know the technical details of how this is achieved, you may read Appendix D in the following paper:
by
gang
-
AFNI Message Board
JY Wrote:
-------------------------------------------------------
> I want to get the contrast just based on all #0 brain activations after
> running AM2. However, -gltsym gave me the contrast based on the
> original data w/o separating #0 and #1. Do you know how to get the
> contrast based on #0 brain activations in -gltsym? Thanks.
Try something like
-gltsym 'SYM:
by
gang
-
AFNI Message Board
Hi Panda,
> (1) Which is the correct way when I want to get the areas which is modulated
> by the semantic distance? AM1 OR AM2? What's the difference between the
> modulated areas we got from AM1 and AM2?
If you want to find out the brain regions whose modulation effects are associated with the semantic distance, AM2 is what you looking for. More specifically, it's
by
gang
-
AFNI Message Board
Cindy,
There are two aspects that may tilt the direction of analysis approach. First, what is the proportion of subjects that dropped out for Time 2 across those two groups? More importantly, exactly what statistical tests would you want to obtain from the analysis?
by
gang
-
AFNI Message Board
What 3dfim+ does is essentially a regression analysis between the regressor(s) specified through option -ideal_file and each voxel in the brain.
> if we want to have the maximum cross-correlation for each voxel then we
> have pick the best delay associated to the maximum cros-correlation value.
What delay are you referring to? Could you be a little more specific?
by
gang
-
AFNI Message Board
The behavior of 3dfim+ seems normal: It would only make sense to get the "Best index" when there are more than one ideal file.
by
gang
-
AFNI Message Board
JoJo,
> Do I need to combine 3dDeconvovle and 3dBucket to achieve this?
No, that would not give you statistics for the new contrasts.
> is there any simple magic that I can use the coefficient information in
> Stats and error information in the Header to construct new contrasts in
> the Stats BRIK?
It's possible to use 3dcalc to obtain them if you know the formula
by
gang
-
AFNI Message Board
Aarthi, you have more subjects listed in the covariate file than the ones in the 3dMEMA script. Also, some subjects in the 3dMEMA script are not in the covariate file either.
by
gang
-
AFNI Message Board
> I just noticed that running the following command yields a dataset with 34 degrees of freedom
> in each of the t-stat subbricks. Is this because I didn't use -unequal_variance? Is it going to
> inflate my t-statistics? Should I be worried?
In this regard 3dMEMA is pretty much like two-sample t-test in which the statistics in the results are based on the pooled variance. Eve
by
gang
-
AFNI Message Board