Warning message help: bufferbiascheck

questions concerning anlysis/theory using program DENSITY and R package secr. Focus on spatially-explicit analysis.

Warning message help: bufferbiascheck

Postby cdevens226 » Mon Oct 05, 2015 10:07 am

Hi Everyone,
While I'm no stranger to all sorts of errors on my secr journey, I've come upon a new one having to do with the buffer and bias check. I've been getting errors like:

(1)
Code: Select all
Warning message:
In bufferbiascheck(output, buffer, biasLimit) :
  predicted relative bias exceeds 0.01 with buffer = 15000


and..

(2)
Code: Select all
Warning message:
In bufferbiascheck(output, buffer, biasLimit) :
  could not perform bias check
> secr.fit(Overberg_90_Phase1, model=list(g0 ~ b, g0 ~ T), CL=FALSE, trace = FALSE, buffer=20000, verify = FALSE)

Error in nlsModel(formula, mf, start, wts) :
  singular gradient matrix at initial parameter estimates
In addition: Warning messages:
1: In secr.fit(Overberg_90_Phase1, model = list(g0 ~ b, g0 ~ T), CL = FALSE,  :
  at least one variance calculation failed
2: In bias.D(buffer, temptrps, detectfn = output$detectfn, detectpar = dpar,  :
  bias.D() does not allow for variable effort (detector usage)


all having to do something with bias issues. I did some homework on the bias.D command and suggest.buffer, but I didn't get very far with those script without running into other questions and errors.

Can anyone explain what is possibly wrong in my data or script to get this error? What must I look to fix in my script? The first warning above saying "predicted relative bias exceeds 0.01 with buffer = 15000" didn't change no matter what buffer I tried... whether it was 1000 or 30,000!?

Through dumb luck, I'm gotten past the second set of errors and warnings but still getting bufferbiascheck warnings and my density results are definitely a bit to high to be right!

Any thoughts or guidance would be amazing!!!

Thanks in advance,
Carolyn
cdevens226
 
Posts: 10
Joined: Mon Mar 03, 2014 1:10 pm
Location: Pretoria, South Africa

Re: Warning message help: bufferbiascheck

Postby murray.efford » Mon Oct 05, 2015 4:02 pm

Hi Carolyn

The bias check is a helpful (?) extra step added by secr.fit to warn you if your 'buffer' setting appears to be truncating the detection function in a way that would result in a biased estimate of density.

You can suppress that step by setting biasLimit = NA in the call to secr.fit.
"predicted relative bias exceeds 0.01 with buffer = 15000" didn't change no matter what buffer I tried... whether it was 1000 or 30,000!?

I hope that at least 15000 changes to 1000 or 30 000?

There is nothing inherently wrong with the code you have shown us, so the problem is in the specifics. My guess is that there is a bad data point that drags the fitted sigma to an enormous value. Also, these problems are more likely with the hazard-rate detection function 'HR' because it has a long tail. I would troubleshoot by suppressing the bias check (that may also take care of the singular gradient error).

Murray
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand

Re: Warning message help: bufferbiascheck

Postby cdevens226 » Wed Apr 19, 2017 8:03 am

Hi Murray!
My apologies, I looked back and just realized you had answered this question for me a couple years ago! I can't seem to figure out how to delete this post! sorry about that...

I just ran it with the biasLimit = NA and it was 100% and same results!

best,
carolyn
cdevens226
 
Posts: 10
Joined: Mon Mar 03, 2014 1:10 pm
Location: Pretoria, South Africa


Return to analysis help

Who is online

Users browsing this forum: No registered users and 8 guests

cron