MNE permutation cluster test

Support & Discussions
,
1

Hello!

I have 3 groups of participants. PD ON medication, PD OFF medication, and Healthy control (PD group are same participants but 2 different conditions).
I plan to run paired cluster permutation between PD ON and PD OFF as following:

contra_mov_on_power_mu_array = np.array(contra_mov_on_power_mu) # (n_observations, n_channels, times) --> (20, 1, 1025)

contra_mov_off_power_mu_array = np.array(contra_mov_off_power_mu)

contra_mov_on_power_mu_array_T = np.transpose(contra_mov_on_power_mu_array, (0,2,1)) # --> (20, 1025, 1)

contra_mov_off_power_mu_array_T = np.transpose(contra_mov_off_power_mu_array, (0,2,1))


X_contra_mov_mu = np.subtract(contra_mov_on_power_mu_array_T, contra_mov_off_power_mu_array_T).squeeze()
 
print(X_contra_mov_mu.shape)  # --> (20, 1025)

t_contra_mov_mu, clusters_contra_mov_mu, cluster_pv_contra_mov_mu, H0_contra_mov_mu = mne.stats.permutation_cluster_1samp_test(
    X_contra_mov_mu, threshold=None, n_permutations=10000, tail=0)

Based on the inputs I have, the above permutation will run 2 tailed 1samp ttest.

I plan to run unpaired cluster permutation between PD ON and HC or PD OFF and HC as following:

contra_mov_on_power_mu_array = np.array(contra_mov_on_power_mu) # --> (20, 1, 705) (n_observations, n_channels, times)
contra_mov_hc_power_mu_array = np.array(contra_mov_hc_power_mu) # --> (23, 1, 705)

contra_mov_on_power_mu_array_T = np.squeeze(np.transpose(contra_mov_on_power_mu_array, (0,2,1))) # --> (20, 705)

contra_mov_hc_power_mu_array_T = np.squeeze(np.transpose(contra_mov_hc_power_mu_array, (0,2,1))) # --> (23, 705)

X_contra_mov_on_hc_mu = [contra_mov_on_power_mu_array_T, contra_mov_hc_power_mu_array_T]

F_obs_contra_mov_on_hc_mu, clusters_contra_mov_on_hc_mu, cluster_pvals_contra_mov_on_hc_mu, H0_contra_mov_on_hc_mu  = mne.stats.permutation_cluster_test(
    X_contra_mov_on_hc_mu, threshold=None, n_permutations=10000, tail=0)

Based on MNE explanation, when I do not specify stat_fun and put tail=0 for unpaired condition, this would run 1 tailed one-way anova.

I want to run 2 tailed for both paired and unpaired to be consistent. So, I can put stat_fun input in unpaired function (mne.stats.permutation_cluster_test) as ttest_ind to run 2 tailed ttest. But, I understood in that case, MNE does not calculate threshold for tstat.

What is the right approach to handle this to be consistent?

Thank you in advance!

Best,

Shahrzad

2

Hi Shahrzad,

As far as I understand defining tail=0 runs a two tailed F test (2 groups) or t-test (paired sample). I do think if you use the code you provided you are consistent, I do remember (not 100 % sure but pretty confident) that a unpaired two sample t-test is equivalent to a one way ANOVA (or at least gives you the same p-value).

Cheers,

Carina

3

Thank you for your response, Carina!

when I put tail=0 for unpaired cluster permutation (mne.stats.permutation_cluster_test), MNE says:

Ignoring argument "tail", performing 1-tailed F-test

But, do you think still I am consistent between my paired and unpaired comparisons based on the code I provided?

Thank you!

Shahrzad

4

Yes, the 1-tailed F-test is just more handy because we are looking at a variance ratio, so it’s enough to test one side (especially for df = 1).

Hope that helps?