p-value of non significant results

[LeSensduMouvement]

Hello, I'm currentely using SPM1D package with Python, performing several paired t-tests with my data. I would like to include p-values of non significant results to my results discussion, as it is possible with 0D statistics. Although I can find p-values of significant results, I couldn't find it for non significant results. I was wondering if I missed something or if in SPM, p-value of non significant results weren't available or pointless. Maybe could you advise me more relevant data from SPM analysis to discuss non significant results?

Thank you!

[0todd0000]

Please refer to issues #222 and #191 for relevant discussions.

To summarize those discussions:

• Usually only cluster-level p-values are reported, but clusters do not exist when p > alpha.
• It is indeed possible to calculate p-values for non-cluster quantities, and the most common quantity is the maximum test statistic value. This probability can be calculated using the rft1d.t.sf function as demonstrated in this script, but note that (a) this probability is generally valid only for p < 0.5 and (b) it is different from cluster-level probabilities so it cannot be directly / meaningfully compared to cluster-level results.

[ShengxingFu]

Hello， I used One-dimensional SPM two-way ANOVA with repeated measures on one factor to assess differences between groups across times. However, there was no significant interaction effect of group and time or group effect. I would like to include a specific P-value of the non-significant interaction effect of group and time or non-significant group effect in my study. Could you tell me how to calculate the P-value？

Thank you very much!

[0todd0000]

Provided you have used parametric inference, please use rft1d.t.sf to calculate p-values for specific test statistic values. BUT note that these p-values may not be accurate when they are large (e.g. p>0.4).

Below is a modification of the anova2 example script

import spm1d

def pmax_f(spm):
    from spm1d import rft1d
    zmax = spm.z.max()  # maximum test statistic value
    df   = spm.df       # degrees of freedom
    Q    = spm.Q        # number of domain nodes
    fwhm = spm.fwhm     # smoothness estimate
    p    = rft1d.f.sf(zmax, df, Q, fwhm)  # RFT survival function
    return p

dataset = spm1d.data.uv1d.anova2.SPM1D_ANOVA2_2x3()
y,A,B   = dataset.get_data()
spms    = spm1d.stats.anova2(y, A, B, equal_var=True).inference(0.05)
pmax    = [pmax_f(spm)  for spm in spms]
print( pmax )

This yields the result:

[0.3677, 0.009, 0.3114]

These are p-values for the maximum F-values from each of the anova2 effects.

[ShengxingFu]

Hello, Thanks for your help. However, there are many rft1d codes in the downloaded MATLAB source code from https://spm1d.org/index.html#. Which is the related rft1d code?

[0todd0000]

rft1d.f is appropriate for ANOVA. "f" means the "F statistic".