confidence intervals on richness/diversity estimators

I’m wondering about the confidence intervals reported for the various richness & diversity estimators. I actually have replicate samples in my dataset, but I get much lower richness and diversity estimates if I use means from across replicates, compared to bulking replicates into a composite category and estimating richness/diversity on that. Is it legitimate to estimate richness from grouped replicates and use the confidence intervals to determine significant differences between treatments?

When you pool the data you are estimating the richness of the pool and not the individual sample. So if you pool sequences from a bunch of soil samples, you are estimating the richness of soil, in general. When you only do it with one sample, it’s the richness of that particular sample. Also, I suspect that because you have insufficient sampling in each sample that the richness of the individual samples will be less than the composite. Another added factor is that the samples are probably very different and don’t overlap much in their membership. Just guessing…