Pairwise comparison in betadisper - when to adjust for small sample size or unequal sample sizes

[ibrila]

Hi,

I was wondering if there is a rule of thumb on when to adjust for small or unequal sample sizes in betadisper? I guess small sample size depends on the context of the study, but not sure how big the difference in sample sizes between groups has to be to warrant bias adjustment?

Thanks

[jarioksa]

Bias adjustment is √[n/(n – 1)], where n is the group size (see ?betadisper). If n is large, the adjustment is nearly 1 and does not adjust results in any way. For n = 5, adjustment is 1.12, and for n = 20 it is 1.03. You may either ponder whether you need to use this adjustment, or you can always use it: it does not change the results if it doesn't matter, and it matters only if it changed the results. See the paper cited in the help page (Stier et al., 2013).

[ibrila]

Thank you for explaining! It seems that using it is the safest option in most cases.

[jarioksa]

I don't remember why bias adjustment was not made default — it was back in 2012. Perhaps to have the same default as the old function and the one used in other software, or because the method was still unpublished when it was implemented in vegan. Perhaps @bbolker remembers (and perhaps he gives his recommendation of the default argument – and I would follow).