elusive bug in IMa2

[jhey]

Sorry to do this to folks, but there seems to be a problem in IMa2.
I can't find the cause as of yet, and so I figured I'd post here.

The problem is that a bias is introduced to the estimated posterior
distributions:
1) for 3 or more populations
2) when the prior on migration is high

By 'high' I mean that if you knew about what the estimated value is
for 4Nu, (call it q) then an upper bound on migration should
ideally not be much higher than 2/q

This is quite similar to the guidlines given on p48 of the IMa2
manual, under 'Getting Started Checklist', but for now those
guidelines are more emphatic.

Of course this is tricky, since you don't know the true values of
4Nu. (but again see discussion in the manual).

sorry about this - will try to get it resolved as soon as possible,
but so far it has proved elusive.

jhey

[Michael Sorenson]

Hi Jody,

Thanks for the heads-up. Does the bias have a consistent direction?

Thanks, Mike

> --
> You received this message because you are subscribed to the Google
> Groups "Isolation with Migration" group.
> To post to this group, send email to isolation-wi...@googlegroups.com
> .
> To unsubscribe from this group, send email to isolation-with-mig...@googlegroups.com
> .
> For more options, visit this group at http://groups.google.com/group/isolation-with-migration?hl=en
> .
>
>

________________
Michael Sorenson
Associate Professor
Department of Biology
Boston University
5 Cummington St.
Boston, MA 02215

(617) 353-6983
FAX: (617) 353-6340

[julien]

Hmmmm... does that issue still stand while using exponential prior
(instead of upper bounds) on migration rates ?

Thanks

Julien

[Fra]

Dear Jody,

thank you for this announcement despite the fact it sounds bad.
I would like to be sure having understood your comment properly,
because I am wavering over the meaning of 1) and 2) . Would you say a
bias appears when we are working with 3 (or more) populations AND when
the prior on migration is high or would you say both conditions are
exclusive ?

Sorry for this tricky question but I work on a model with 2
populations and I fixed migration priors to 0.1 for a 4Nu of 20, I
feel stressed and would be sure before going to far in my analyses.
Could you, please, confirm they are not exclusive?

Anyway, I hope you will find a solution.

Kind Regards

François