Error running t-test statistics (paired-samples / independent t-test)

Bugs
1

Dear Francois,

In a previous edition of Brainstorm I was able to perform t-tests without any problems. However, with with the current version I get an error when I try to compare to conditions (each has 56 subject averages) using a paired samples t-test.

My pipeline:

  • calculated the time-frequency decomposition of source activity over single trials
  • averaged the time-frequency files, baseline-normalized them (z-transform) and extracted a time-window of interest (300-500 ms post feedback)

Next, I put condition 1 in the “Files A box” and condition 2 in the “Files B box”. Via the pipeline editor I click on "run–>Test–>Parametric Test: Paired.
The paired-test box further ask you whether you would like to average across time, signals, frequency. Whenever I don’t tick these boxes Brainstorm encounters an error: “Cannot process source maps, or any file that does not have explicit row names. Could not find a common list of rows: trying to average directly the matrices”. However, when I do tick the boxes “average across time, signals. frequency”, the t-test runs fine and the output is generated as a intra-subject file.

The same error occurs with testing the ‘differences of means’ (subjects averages of condition A vs subject averages of condition B).

It seems as a bug to me, since I was able to run this t-test previously and correct for signals, frequency, time AFTER running the t-test via the “statistics” tab in the Brainstorm GUI.

I hope you have any idea what might be causing this error?

Kind regards,
Melle van der Molen

PS. comparing the two conditions using Fieldtrip’s non-parametric cluster-based permutation test works fine - however, here I do opt for the controlling for signals, frequency and time - due to the computational demands whenever you don’t opt for these controls (I run into out of memory errors in that case).

2

Hi Melle,

You are right, there were some issues with the way the new test functions handle the time-frequency maps computed on full cortex.
The messages you reported should be warnings and not errors. I fixed this and added more explanations in the message.
This option “Match signals between files using their names” in the process panel does not apply to full cortex time-frequency maps (because in these files, the signals are not “named” but just supposed to correspond to the surface vertices).
Uncheck this option to skip the warning.

Additionally, I think there is some confusion here:

  • The three “average” checkboxes in the tests options actually AVERAGE the data BEFORE testing.
  • The three “control over dimension” checkboxes in the Stat tab are related with the number of multiple comparisons considered for the FDR or Bonferroni corrections for multiple comparisons, AFTER testing.
    Please see the new statistics tutorial for help with these two aspects:
    http://neuroimage.usc.edu/brainstorm/Tutorials/Statistics#Correction_for_multiple_comparisons

Also note that parametric t-tests are maybe not adapted to process time-frequency power values.
This is currently under evaluation in Brainstorm, I cannot give you more precise directions right now.
Updates will soon be available on this tutorial page: http://neuroimage.usc.edu/brainstorm/Tutorials/Workflows

Francois