I am running 3dDespike on AFNI version 16.2.16. For testing purposes, I am running my analysis on the same dataset on two OS (CentOS 6.5 and CentOS 7.2). Comparing the outputs from the two OS, I’ve noticed that there are differences in the values. Meanwhile, the input files for each subject on both OS are identical. The differences are minor, however they pose some concerns for subsequent analyses.
I was wondering if anyone can advise on what might contribute to these differences in the 3dDespike outputs?
To be sure, are both of those CentOS systems running the
exact same version of AFNI, the exact same set of binaries?
Can you report “afni -ver” on each system?
I ask because there was a change made to 3dDespike around
that time which would indeed lead to slightly different results.
The program was made to be much faster. Either result should
be considered “correct”, but they are indeed a little different.
Why do you say “The differences are minor, however they pose some concerns for subsequent analyses”?
Minor differences should not produce major changes in the results of subsequent analyses. If minor differences at one stage make a big difference in the final brain mapping results, then there is something else wrong with the analysis – such as bad data or other bad inputs (e.g., a very ill-conditioned experimental design).