Error in fitting models

[Amael Hinojo Antille]

Dear all,

Using secr.fit function I found a warning message in some of the models I was trying to fit:

M0C <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~1, sigma~1))
MbC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~b, sigma~1))
MtC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~t, sigma~1))
MbtC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~b+t, sigma~1))
MTC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~T, sigma~1))
MBC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~B, sigma~1))
MbkC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~bk, sigma~1))
MBkC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~Bk, sigma~1))
MKC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~K, sigma~1))
MK_NMC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~K, sigma~1),method="Nelder-Mead")
MkC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~k, sigma~1))
Mh2C <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~h2, sigma~1))
Mh2_NMC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~h2, sigma~1),method="Nelder-Mead")
Mh2_NM_SLC <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~h2, sigma~1),method="Nelder-Mead",start=list(D=0.00012,g0=0.24,sigma=3200))
Mh2_NM_SM0C <-secr.fit(roe_capthist, mask=mask_prox, model = list(D~1, g0~h2, sigma~1),method="Nelder-Mead",start=M0)

Here the warning message delivered:

Warning message:
In secr.fit(roe_capthist, mask = mask_prox, model = list(D ~ 1,  :
  possible maximization error: nlm returned code 3. See ?nlm

 

This is the information given by R about ?nlm : Non linear minimization warning code 3: Last global step failed to locate a point lower than estimate. Either estimate is an approximate local minimum of the function or steptol is too small

Do someone know the meaning of this warning message and if it really affect the results of the fitted models?

Thank you very much for your attention and help,

Amael

[Murray Efford]

It has not been a problem in the cases I've examined: proceed with caution. If some variances are missing a model can be re-fitted with model = 'none' and start=previousmodelfit.

You seem to be trying to fit an exhaustive list of detection models. What about all the combinations involving variation in sigma? Basically, I think this is not good practice.

--
You received this message because you are subscribed to the Google Groups "secr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to secrgroup+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.