tfNBS results interpretation from negative contrast vector

Hi experts:

I am new to tfNBS analysis using connectomstats and though probably a silly question, I am seeking help interpreting the “fwe_1mpvalue.csv” output file from a negative contrast vector.

We had two groups in the two column design matrix and we tested two contrast vectors separately: pos [1, -1] and neg [-1, 1] to see which edges were significantly different.

Both the pos and neg contrast “fwe_1mpvalue.csv” output files contained mostly zeroes. However, the “pos_fwe_1mpvalue” file also had values between 0.001-0.040 (which looked to me like normal p-values) while the “neg_fwe_1mpvalue” file contained values ranging up to 0.99+.

Based on other queries in this forum I see that values in the “fwe_1mpvalue.csv” file are calculated as “fwe_pvalue = 1 - p”.

My questions are:

  • does the “fwe_pvalue = 1 - p” apply only to a positive contrast, and would the negative contrast be “fwe_pvalue = -1 - p” ?
  • Or another way of putting it: Would the values in either neg or pos “fwe_pvalue.csv” files need to be above 0.95 to represent a significant difference? And therefore in my example, the significant differences are only observed in the negative contrast since it contains values above 0.95?
  • would the significant p-values in the neg file mean that the second group has increased connectivity in those edges compared to the first group and are the connectivity scores I should be using in the “enhanced.csv” file?

I guess I’m probably confusing myself, but just want to be sure.

Thank you for your time!

Hi @tieg ,

I’m not an expert on this but here’s my understanding -

  1. If you were doing a simple t-test, then yes the sign of the contrast only changes the direction of the t-statistic and your first assumption would hold. I would expect to see this in the uncorrected p-values but there can be some variation from this in the FWE-corrected p-values due to randomness in the permutation distribution.
    However, when I look into my own data and try these two contrasts between groups, I don’t see a 1:1 correspondence between the positive and negative contrast uncorrected p-values (they’re related but not exactly 1-the_other_p). I’m not sure where this comes from I would be careful to conclude that the positive contrast p-values are exactly 1-negative contrast p-values.
  2. The significant differences are only those connections where fwe_1mpvalue>0.95 in the negative contrast, but if you didn’t have a pre-specified directional hypothesis, you should correct for looking at both directions, in which case only connections exceeding fwe_1mpvalue>0.975 should be considered.
  3. Interpretation is correct, but I don’t think the enhanced.csv gives you interpretable connectivity values. To do this, I would find connections exceeding your statistical threshold and then grab the raw connectivity values (from the connectomes that were used as input to connectomestats) for those suprathreshold connections.

Hope this helps,

Thank you so much, yes this is very helpful!