Using individual and occasion covariates in the same DM

[simone77]

First of all thanks for being so clear.
I have been doing some trials as a consequence of your suggestions in order to estimate these "difficult" parameters.
I believe that the focus of these posts has changed significantly and that would be better to open a new topic, for this reason I have posted a new one here:
http://www.phidot.org/forum/viewtopic.php?f=1&t=1978

[simone77]

Just a last commentary on this.
The first question on this topic was about the huge similarity in terms of deviance between two models ({phi(cov1+weight*t) p(t)} and {phi(cov2+peso*t) p(t)} ) where I just changed the environmental covariates and maintained the weight variable.
As suggested by Abreton, the way I built the DM was incorrect for redundancy issues.
Now I have tried to build again the two DM in the right way, like DM1 in the above post and I still get very similar deviances (2191.4255 vs 2191.9983).
I believe this should mean that the weight covariate is explaining much more variance than environmental covariates do.

[abreton]

I assume phi4 and phi5 are still giving you trouble in these models? Assuming covariate 1 and 2 are not strongly correlated (if they are then drop one of these from your analysis), then, assuming CJS was the most appropriate parameterization for your data, then I would likely proceed by building all possible models including cov1, cov2 and weight. If biology (and the data) supported 2-way (or 3-way) interactions then I'd incorporate these interactions into the all possible models. See MARK Menu, Run>Subset of DM Models for a quick solution for building 'all possible'.

No doubt several models would be within ca. 6 AICc units. I'd proceed by model averaging...and make inferences based on the model averaged (unconditional) estimates of the parameters. For the cov1 and cov2 effects, I'd also model average betas associated with these effects and use a log odds ratio to quantify/visualize effect size.

I've never used the model deviance to make inferences about explained variance...and note, I don't think you could say 'weight' is responsible for the realized deviance unless you compare models without weight.

andre

[simone77]

Thank you for your answer, as always it has been quite useful to me, I will check the options you talk about.
Also thanks to make me realize that

..I believe this should mean that the weight covariate is explaining much more variance than environmental covariates do..

is a quite incorrect statement.