Dear colleagues,
I calculated an AMOVA using Arlequin (10.000 bt) on a microsatellite dataset (12 sats). There are 4 more or less differentiated subpopulations. The data is fine, I did vast checks and a lot of other analyses.
In my AMOVA however, I end up with the following percentage of variance explained:
Among subpops: 1.69 (significant)
Among inds within subpop: -2.43 (not significant)
Within inds: 100.73 (not significant)
I realise this is a strong indicator for good admixture, and no real subpop strcuture. Yet I am a little confused about the negative value:
- Is this totally fine, an artefact of computation?
- Or does it indicate some problems with the AMOVA?
Thank you for your help!
EDIT: formatting
I have had negative values before, but not with Arlequin and not with microsats (RADseq-derived SNPs can produce similar patterns). Having such a large proportion of variance within individuals suggests there are lots of unique alleles per individual. That seems odd though. Is there a lot of missing data? I think with a negative value and a variance above 100, there is some computational issues or data formatting trouble. Sorry can't be of more help.
Hi Florian
Negative values in AMOVA are just a complication of the statistical approach, They should be considered 0 (zero), unless they are large. In the latter case they probably mean some problem in the input file.
Cheers
Pablo
Thank you all for your answers!
I understand these values, while a little weird, probably being the consequence of intern calculations.
So with having the data checked by all means and using alternative analyses as well to support interpretations, one should be fine.
In this instance, The AMOVA percentage of variance can be explained as:
Among subpops: 1.69 (significant)
This seems Ok.
Among inds within subpop: -2.43 (not significant)
Negative valuesare taken as zero.
Within inds: 100.73 (not significant)
More than 100 , is explained as: as sum of all 3 above values is 100, and as among inds within subpop: -2.43 is taken as zero, so
Within inds is (100- 1.69) 98.31.
Does anyone know of a reference stating that negative values can be taken as zero? I can't seem to find anything other than in discussion groups with no real reference.
Small negative values, I agree. I'm reviewing an MS now that reports a small negative phiSC = -1.168, accounting for a highly significant (p < 0.001) -23% of the total variance. This bothers me, as adding this to the other two components makes them 49% and 74% of the total. And, I admit that I don't really understand how a variance can be negative (covariance, sure).
- Badges
- Science topic
- Similar topics
- Filing