Closed cdiener closed 3 years ago
Hi @cdiener -- maximum likelihood estimates aren't unbiased (only consistent). So this isn't really surprising to me, and we can't "fix" this without invalidating our hypothesis tests. That's my rapidfire 2c on this :) Amy
Ok, I see, Thanks! Yeah, I guess that is maybe a limitation of the data.
Hi @cdiener ,
I don't have much to add to that, so I'm going to close this issue. But feel free to re-open if you have a follow up question!
Bryan
Hi,
I noticed that the estimation of model parameters seems to struggle a bit with converging when phi >> mu. In particular for the following example:
You get a good estimation for mu and phi eventually but it requires a very large n (>1000) and seems to be biased before convergence (underestimation). Do you know a strategy to improve on that? We observed that this often hampers the power for some taxa and we get phi>mu quite often for negative samples (estimated from healthy untreated mice fecal samples).
EDIT: had a typo in the code