Maximum number of covariates before overparametrization ?

[Sarah74]

Hi everyone,

I'm using an occupancy model, and I want to know the maximum number of covariates that I can include in my model before I riks overparameterizing it. Is there any "rule of thumb" to decide how many covariates we can legitimately include based on the effective sample size (1 additional covariate per X sites that were found to be occupied)? By the way how do I compute the effective sample size with single season occupancy models?

Thanks a lot.

[bacollier]

Hi everyone,

I'm using an occupancy model, and I want to know the maximum number of covariates that I can include in my model before I riks overparameterizing it. Is there any "rule of thumb" to decide how many covariates we can legitimately include based on the effective sample size (1 additional covariate per X sites that were found to be occupied)? By the way how do I compute the effective sample size with single season occupancy models?

Thanks a lot.

Sarah74,
I have not seen anyone reply to this so here is my $0.02. The general 'rule' is that you want 10-20 'rows' of data for every parameter you are interested in estimating. I guess it all depends, if you have a dataset with 20K records over 15 occasions, then your risk of overparameterization is slight, if you have 20 records, then its much higher. I would actually advise some thought on your paper on what is actually 'relevant' to be included in a single predictive model, and if you have 35 covariates for 100 rows of data, you probably would need to rethink what is really relevant? Maybe more details would help with a better response?

Bret

[Sarah74]

Thanks a lot Bret.
I did some simulations to see if I obtain good estimations and it looks ok but I will also look at it with your 'rule'.

Sarah