Assessing model fit valid when convergence/var-covar issues?

[Alfricho]

Hello all,

I'm analysing some data for Bolivian mammals using a few covariates. When running the "global model" to assess the the model fit, I get both the Convergence and Variance/covariance warnings, probably because there are just too many covariates in that model, as I don't get the warnings with simpler models. The question is if the results from the assessment of model fit are still valid despite these warnings (according to those results, the model has a good fit).

Many thanks in advance,
Alfredo

[darryl]

As far as I'm aware the method has never been assessed in those circumstances, though I'd certainly be cautious as the warnings are suggesting that perhaps the estimated parameters (which get used during the bootstrap) may not be reasonable. I'd also be concerned about potential over-fitting of the data if you have a lot of covariates. With no claims that it's the 'right' thing to do, one thing you could try is assess the fit of 2 or 3 simpler, but still relatively complex models, that do not give you any warnings and see if you get consistent results from the model fit assessment.
Cheers
Darryl

[Alfricho]

Many thanks Darryl for your fast and helpful response.

I've followed your suggestion and the simpler models have still a good fit, and no warnings.

All the best,
Alfredo