Confidence intervals?

[eddie....@gmail.com]

Hi Dr. Excoffier,
I have read through the posts here about bootstrapping as well as recently published papers that use fsc2, but have a question about generating CI around point estimates. Its my understanding that the point estimates with the highest likelihood are used to simulate new SFS's which are then used to generate CI around the original estimate from the observed data. But since the estimate of parameters from the observed data is conducted (usually) 100 times, the user already has a distribution of parameter estimates. Is there a reason that I am unaware of why one would not want to use this distribution that is from your observed data as CI around the point estimates with the highest likelihood?
Thanks so much!
Ed Myers

[Laurent Excoffier]

Hi,

the distribution of parameters obtained from the non maximum-lhood solutions cannot be used for CI inference, as it is not representing evolutionary variance, but due to some variance in the optimization procedure.

The procedure you describe above is a parametric bootstrap procedure, but you could also use a non-parametric (block) bootstrap procedure (probably better). 

L