Dear Rebeka,

The RRPP::pairwise function does not adjust P-values.  It reports the P-values 
as a percentile of the observed values in the distributions of random values 
from the permutation procedure.  One can adjust their pairwise level of 
significance if they wish — there is an option for that to illustrate the 
critical value in the distribution — but there is no attempt to “correct” 
P-values based on a particular philosophy.

Furthermore, one should be careful about correcting P-values that have a finite 
limit based on the arbitrary number of permutations performed.  For example, 
with 1,000 random permutations a P-value of 0.001 is the lower limit.  For 
10,000, 0.0001 is the lower limit.  One could make a judgement about 
“significance”, influenced perhaps by using fewer random permutations than 
would be required to find significance.   

As for whether P-value corrections are necessary for statistics with 
distributions generated from a resampling procedure, I have never had a strong 
opinion whether they should or should not be performed, especially when you are 
generating 15 pairwise differences between means, presumably among 6 means, in 
every random permutation of the data, for the same data.  (There might be some 
programs or functions that generate 15 separate distributions from 15 different 
permutation procedures of 15 different data sets, maybe by subsetting the data 
by particular groups.)  Because resampling procedures are not universally the 
same, it is difficult to have a blanket opinion about what pairwise 
significance levels should be.  Perhaps worrying less about an arbitrary 
criterion and focusing more on the effect sizes from the tests is a better way 
to go.

Best,
Mike

> On Nov 29, 2021, at 3:11 PM, Rebeka Rmoutilová <[email protected]> wrote:
> 
> Hi!
> 
> I have shape data with groups by sex and age (young, middle and old). I used 
> procSym function to perform Procrustes ANOVA on the data and now I want to 
> use a post-hoc test to find which groups are different. Totally, there is 15 
> combinations of sex and age groups, but I am interested only in some of the 
> comparisons to find whether there is sexual dimorphism inside the age groups 
> or whether the age groups differ within males or females. So for example, I 
> am not interested whether there is a difference between young males and old 
> females which leads to only 9 comparisons.
> 
> There are two functions I could probably use: permudist in Morpho package and 
> pairwise in RRPP package. Both functions compute differences between all 
> possible pairs of groups applying a stricter p value correction than if I 
> make only the 9 comparisons. Can I test the differences with one of those 
> functions without any p value correction and then correct it for 9 
> comparisons using the p.adjust function? I see that at least in the permudist 
> function, it is possible to use no adjustment method, but I am not sure 
> whether it is a correct approach. Additionally, is it necessary to use p 
> value correction when I use permutations in the permudist function?
> 
> Rebeka
> 
> -- 
> You received this message because you are subscribed to the Google Groups 
> "Morphmet" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to [email protected] 
> <mailto:[email protected]>.
> To view this discussion on the web visit 
> https://groups.google.com/d/msgid/morphmet2/8450baec-0ec3-4b1f-be42-ed6fef8ab811n%40googlegroups.com
>  
> <https://groups.google.com/d/msgid/morphmet2/8450baec-0ec3-4b1f-be42-ed6fef8ab811n%40googlegroups.com?utm_medium=email&utm_source=footer>.

-- 
You received this message because you are subscribed to the Google Groups 
"Morphmet" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/morphmet2/8A7D3FEF-812C-4E6E-9C92-AB516FBC4FE0%40gmail.com.

Reply via email to