Numerical convergence error with negative significant digits

questions concerning analysis/theory using program PRESENCE

Numerical convergence error with negative significant digits

Postby Lea » Tue Feb 23, 2016 6:59 pm

Hi all,
I am looking at the effect of different landscape types on occupancy at different scales (100, 250, 500, 1000, and 2000 m buffers). I am running identical models at each scale. The top models seem to be fine (no error messages) at every scale except 500 m. At 500 m I get the following error message:

Numerical convergence may not have been reached. Parameter estimates converged to approximately -14.01 significant digits.

From other posts it seems that this message typically isn't a problem as long as there are >2 significant digits. Given that I have negative significant digits I conclude this is a problem. I have tried supplying initial starting values from the identical models at different buffer sizes and I still get the error message. The beta values given by these models seems in line with the estimates at the other buffer sizes but some of the betas have zero SE.

Given that the data is very similar between the different buffer sizes and the betas seem reasonable why is this error only showing up at the 500 m scale? There are a fair number of zeros associated with the land use type that is giving me grief but I would think that this would be more of a problem at the smaller buffer sizes as there are more zeros.

Anyone have any insight?
Lea
 
Posts: 33
Joined: Thu Oct 24, 2013 1:09 pm

Re: Numerical convergence error with negative significant di

Postby Lea » Tue Feb 11, 2020 7:01 pm

Hi all,
I am still struggling with this but haven't received an answer.
Here is an example of the error message.
**** Numerical convergence may not have been reached. Parameter estimates converged to approximately -15.16 significant digits.

I find that models with this kind of error message have SE of zero, similarly when I get ***** neg. std.err(s) in VC matrix I find that SE of beta is -1.#IND00, but I am wondering if the parameter estimate could still be considered reliable?

Thanks,
Lea
Lea
 
Posts: 33
Joined: Thu Oct 24, 2013 1:09 pm

Re: Numerical convergence error with negative significant di

Postby darryl » Tue Feb 11, 2020 7:15 pm

Hi Lea,
Check out the PRESENCE FAQ: viewtopic.php?f=40&t=665
Hope you enjoyed WCH9! ;-)

Cheers
Darryl
darryl
 
Posts: 495
Joined: Thu Jun 12, 2003 3:04 pm
Location: Dunedin, New Zealand

Re: Numerical convergence error with negative significant di

Postby Lea » Thu Feb 13, 2020 1:55 pm

Hi Darryl,
Thanks for response, I will give that a try. I had a great time at WCH, what a great conference and New Zealand was fantastic. I would have liked a chance to get caught up but I only saw you on the run!
Cheers,
Lea
Lea
 
Posts: 33
Joined: Thu Oct 24, 2013 1:09 pm


Return to analysis help

Who is online

Users browsing this forum: No registered users and 9 guests

cron