Hi Nicholas,
What you're doing seems proper to me (but see below regarding calculation of
the SE).
Rarefaction error windows should coalesce around the endpoint as you get to
larger randomized samples. And you only have limited numbers of richness values
to draw from. You can confirm this
Nicholas,
Your approach looks perfectly OK. However, rarefied sampling is normally done
*without* replacement.
It looks to me that a similar analysis could be done with vegan::specaccum
function. However, that uses sampling without replacement.
About your specific questions: you can use