Multi-season models: Trouble estimating std errors on betas

[Seth]

Hello,

I've been working with some long-term detection data and trying to analyze them in a multi-season framework in PRESENCE. The trouble I am having is with my betas. My models seem to run and I get reasonable parameter estimates, but frequently (particularly if I include more than one site covariate) the standard errors do not estimate, I get outputs of -1.#IND00. Does this simply mean I don't have enough data for the complexity of the models I'm trying to fit?

Thanks so much for any help you can give.

-Seth

[jhines]

Hi Seth,

If your parameter estimates are near a boundary (zero or one), you'll sometimes get undefined standard errors like that. Of course, if a real parameter is near zero or one, then the associated beta parameters are near minus or plus infinity. In practice, once the sum of the beta's for a parameter gets above 50 (or below -50), then the parameter is approaching a boundary and you might see those standard errors.

Another way of getting those standard errors is when you don't have enough data to estimate the parameter as a function of those covariates. If you have two covariates for a parameter, you need to have all combinations of the covariates in the data. For example, you can't have a binary covariate and a continous covariate where the data contains a range of values for the continous covariate when the binary covariate is zero and only 1 value when the binary covariate is 1.

Jim