Differences in model ranking between Presence and Unmarked

[leara...@gmail.com]

Hi all,

I am normally a Presence user but decided to use Unmarked for an analysis because I want to run the same set of models with many different occupancy data sets which would take forever in Presence. First I wanted to compare model ranking between the two programs with one of the data sets just to see if I would get similar results. Given some of the user group comments I read in other threads I didn't necessarily expect to get the same AIC values but I would hope to have similar rankings and similar inferences. To test this I ran single season models in both programs with constant occupancy and standardized covariates on probability of detection (forest characteristics such as tree density and canopy cover). The occurrence data and covariates were identical between programs. I got the same AIC value for the null models between the programs but different AIC values for the models with covariates on pdet. More concerning to me is that the model rankings were different (as well as deltaAIC and model weights). Now admittedly the data is sparse (lots of zeros in the occurrence data set). Is it possible that the differences are due to difference in convergence criteria? And if so, is one program "better" than the other? If I get different results between the programs and Presence is better, then there isn't much point in continuing with Unmarked. Any insight would be appreciated!

Lea

Presence
Model	AIC	deltaAIC	AICwgt	Model likelihood	no.Par.	"-2*LL"
psi(.),p(dist_edge)	185.33	0	0.9055	1	3	179.33
psi(.),p(tree<10)	192.03	6.7	0.0318	0.0351	3	186.03
psi(.),p(crown)	192.63	7.3	0.0235	0.026	3	186.63
psi(.),p(tree>10)	193.55	8.22	0.0149	0.0164	3	187.55
psi(.),p(canopy_stdv)	194.29	8.96	0.0103	0.0113	3	188.29
psi(.),p(.)	194.89	9.56	0.0076	0.0084	2	190.89
psi(.),p(canopy)	195.22	9.89	0.0064	0.0071	3	189.22
Unmarked
	nPars	AIC	delta	AICwt	cumltvWt
psi(.)p(dist_edge)	3	194.73	0	0.246	0.25
psi(.)p(.)	2	194.89	0.16	0.227	0.47
psi(tree)p(crown_diameter)	3	195.44	0.7	0.173	0.65
psi(tree)p(canopy)	3	196.48	1.74	0.103	0.75
psi(.)p(tree>10)	3	196.86	2.12	0.085	0.83
psi(.)p(tree<10)	3	196.89	2.15	0.084	0.92
psi(tree)p(canopy_stdv)	3	196.89	2.16	0.084	1

[Jeffrey Royle]

hi Lea,

 I'm sure the two have been compared before (on the email list too) and they should produce consistent results.

 I'd be interested to see the summary output for the unmarked model

psi(.)p(tree<10)

 Just to see if the output appears sensible..  Be sure to standardize your covariates.

regards

andy

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