Number of Function Calls

[Rovang]

Hi,

I am quite new to PRESENCE and I just started running a single season model. I have started with my most complex model first and I have included the abundance of food as a covariate for occupancy. If I use species A as the food covariate in the model it works fine. But if I use species B or C the model doesn't converge and I get really high beta estimates for occupancy. I am not quite sure why this is happening since the data on the three species is not dramatically different. From the output, the only difference I see is that the "Number of Function Calls" for species B and C is quite a bit higher than for species A.

What does the "Number of Function Calls" refer to? Could this be the problem?

Thanks!
Sarah

[jhines]

Hi,

The number of function calls is the number of times the optimization routine calls the function to compute the likelihood value for the model. The optimization routine can be thought of as a 'trial and error' procedure, where it computes a likelihood value for a set of parameter values repeatedly, until it finds the set of parameter values which yields the lowest negative log-likelihood. It might take only a few function calls for a simple model. Or, it might take thousands of calls for some complicated models. The optimization routine is usually able to detect when it has found the parameter values and stops when this happens.

With sparse data or complicated models, the optimization routine can have difficulty finding the lowest negative log-likelihood value. When this happens, you will usually see a warning message in the output indicating that it could not find the lowest negative log-likelihood, or that the variance-covariance matrix could not be computed. Another indication of a problem is when the standard errors of the estimates are huge or not estimable. If you don't see these things, then the number of function calls is not a problem.

Since you said the model doesn't converge for 2 of the species, I suggest simplifying the model. I usually recommend starting with the simplest model and working my way towards the more complicated ones.

Jim