Bayesian version of one-sample t-test

AFNI Message Board
1

AFNI version info (afni -ver): Precompiled binary linux_rocky_8: Jun 29 2023 (Version AFNI_23.1.10 'Publius Helvius Pertinax')

Dear Colleagues,

Hi. Does AFNI have a Bayesian version of a one-sample t-test?

My current plan is to use this.

3dttest++ \
    -prefix output_tt_"$region""$regsiz"_"$prepro"_"$statin".nii \
    -singletonA 0.0 \
    -setB ...

In this scenario, I am equally interested in all three outcomes: significantly negatively different from zero, not significantly different from zero, and significantly positively different from zero. In other words, I am only choosing a null hypotheses of 0.0 for the sake of convenience. I plan to create a tricolor map for my manuscript with these three outcomes. However, for the not-significantly-different-from-zero outcome, it seems to me at present that I cannot conclude such because one cannot prove the null in the Null Hypothesis Decision Strategy.

Sincerely,

Dante

2

Dante,

Your 3dtttest++ script is equivalent to

3dttest++ \
    -prefix output_tt_"$region""$regsiz"_"$prepro"_"$statin".nii \
    -setA ...

I am equally interested in all three outcomes: significantly negatively different from zero, not significantly different from zero, and significantly positively different from zero. In other words, I am only choosing a null hypotheses of 0.0 for the sake of convenience. I plan to create a tricolor map for my manuscript with these three outcomes. However, for the not-significantly-different-from-zero outcome, it seems to me at present that I cannot conclude such because one cannot prove the null in the Null Hypothesis Decision Strategy.

Everything in statistics is probabilistic, which means that certainty cannot be proven for anything except for trivial cases, including non-null hypotheses, regardless of the modeling framework.

Does AFNI have a Bayesian version of a one-sample t-test?

You may want to try the program RBA which can perform Bayesian analysis on a list of brain regions, rather than at the voxel level.

Gang

1 Like
3

Gang,

Hi! How are you?! I would like to do a voxel- rather than region-wise analysis, so I must assume the answer to my question is "no." Thank you anyway. I hope you are well.

Sincerely,

Dante

P.S.: Thank you for correcting me. I should have said that one can only reject or fail to reject the null hypothesis.

P.P.S.: Your code and my code are equivalent. However, I like to write my code with the default options specified so that if I ever need to use it again with a value other than 0.0, I do not need to reread the help file.