Back that up. What exactly is wrong, how is it not addressed in the 16 section-long discussion, and how is my main argument affected? Take another look at Fig. 5A, for example. "A better alternative is the Poisson log-normal distribution" (O'Hara 2005), but assuming it invalidates H and D.

Fisher's alpha goes down, not up, as more stems are randomly drawn from across the BCI plot (upper green line). But it goes up if stems are drawn in order going SW-to-NE across the plot (lower). Chao 1 rises steeply (red). CEGS doesn't (upper blue) while capturing the area signal (lower blue).

Methods matter! BCI tree diversity partitioned by size class (5 cm DBH running window). CEGS (blue): flat, then a sudden drop ~40 cm. Chao 1: lower, different, and higher variance. CBR/alpha/theta (orange/purple/pink): ~50% increase on the way up to ~25 cm. But diversity should always decline.

The BCI manuscript is now done. The log series and ZSM fit the overall 50 ha data poorly. They predict the 1 ha subplot data just as poorly (last post) and underestimate beta diversity (new graph). CEGS is of course best. Manuscript available upon request: comments would be very much appreciated.

CEGS wins again. Predictive test based on the 50 ha BCI tree plot dataset of Hubbell/Foster/Condit (thanks). CEGS fits to species counts in 1 ha plots better predict counts in 9 plots to the east (blue). PLN wins once (red). Log series loses every time. Likelihood ratio cutoff is 10:1.

No problem. It's CEGS. Here's a "subsample at random and estimate richness" analysis of the classic BCI data set of Condit et al. (1996). CEGS asymptotes almost immediately. OPR is much the same. ACE and Chao 1 are way off, consistent with your results. Great paper, I totally agree with it.