why p-value correction

[Alexandra_Bastkowska]

Hi,

I am just wondering why we need the p-value correction for permanova, where I for example compare just two groups and the overall microbial composition on a specific distance metrics? Likewise for alpha diversity?

Alex

[timanix (Timur Yergaliyev)]

Hello!

You are right - we don't need p-value correction when there are only 2 groups. In that case, q-value will be equal to the p-value.

If you have only two groups, ignore anything in the pairwise section and report the p-value from the general test (upper table).

[Alexandra_Bastkowska]

Hi Timur,

thanks so much for your reply.

When does it make sense to apply it?
scenario a) If i have a factor with multiple groups, for example treatment with 3 groups, then it makes sense to include the corrected p-values and qiime2 does it already.

scenario b) If i have an experiment with multiple factors (e.g. BMI,gender, age, treatment) and would perform permanova for each of them, would then p value correction be necessary? if this is the case, can qiime calculate the corrected p-value for each factor?
scenario c) If I have just one permanova model with all the factors in the one model, then p-value correction may not be necessary, because permanova tests one factor after accounting the variance of the previous factors?

[timanix (Timur Yergaliyev)]

Glad to help!

@Alexandra_Bastkowska

scenario a) If i have a factor with multiple groups, for example treatment with 3 groups, then it makes sense to include the corrected p-values and qiime2 does it already.

That is correct

@Alexandra_Bastkowska

scenario b) If i have an experiment with multiple factors (e.g. BMI,gender, age, treatment) and would perform permanova for each of them, would then p value correction be necessary? if this is the case, can qiime calculate the corrected p-value for each factor?

No, if you test multiple factors, you don't need to correct P-values since you answer different biological questions. You correct P-values from multiple comparisons of various levels within a single factor.

@Alexandra_Bastkowska

scenario c) If I have just one permanova model with all the factors in the one model, then p-value correction may not be necessary, because permanova tests one factor after accounting the variance of the previous factors?

Right, you don't need to correct in that case. But if you got a significant p-value for one of the factors, and that factor consists of more than 2 levels (for example, 3 different diets for factor Diet), and you want to know which diets are different, then you can perform pairwise comparisons and correct p-values of them.

For example, you run adonist test for Diet and Sample type and discovered, that Diet is significant. You can report that p-value directly. Then, if you have more than 2 diets, you can run PERMANOVA test in pairwise mode, and report corrected p-values for pairwise comparisons.

Best,