help > Power analysis for NBS
Showing 1-3 of 3 posts
Display:
Results per page:
Nov 23, 2020  02:11 AM | Paul Thomas
Power analysis for NBS
Hello all,

I was wondering if anyone knows how to go about doing a power analysis for either experiments conducted with NBS to determine power, or just generally to determine whether, given a sample size, desired effect size, etc., one has adequate power. For example, is it as simple as conducting a power analysis on the statistical test used on each edge? I guess I am confused on how to determine the effect size of an NBS-defined subnetwork. Any input is appreciated.

Thank you,

Paul
Nov 24, 2020  12:11 AM | Andrew Zalesky
RE: Power analysis for NBS
Hi Paul, 

I suggest performing the sample size calculation for a single connection using existing power calculators for univariate tests. This will most likely yield a conservative estimate of the required sample size because evidence that accumulates across multiple connections comprising a subnetwork is not considered in this calculation (although this is not guaranteed). 

Exact power calculations are challenging in these circumstances. 

The average effect size across a subnetwork could be used as a crude effect size estimate for the univariate sample size calculation. Remember that the effect size estimate will most likely be inflated when inferred from a small sample. 

Andrew 

Originally posted by Paul Thomas:
Hello all,

I was wondering if anyone knows how to go about doing a power analysis for either experiments conducted with NBS to determine power, or just generally to determine whether, given a sample size, desired effect size, etc., one has adequate power. For example, is it as simple as conducting a power analysis on the statistical test used on each edge? I guess I am confused on how to determine the effect size of an NBS-defined subnetwork. Any input is appreciated.

Thank you,

Paul
Nov 24, 2020  03:11 PM | Paul Thomas
RE: Power analysis for NBS
Andrew,

Thank you very much for responding so quickly. I think your suggested method will be more than adequate. I do not have such a strong background in statistics so I didn't know if that would yield a good estimate or not. Thanks again for the clarification.

Best,

Paul