Nest survival covariate-specific estimates

[jbruggin]

In Chapter 14 Jay Rotella provided a nice example of how to get covariate-specific estimates (habitat in this instance) using “User-Specified Covariate Values” (e.g., Native = 1 when the habitat type is native grassland). Are there any inherent problems in specifying more than one covariate simultaneously (e.g., field season and habitat)? I have a treatment covariate and covariates for multiple field seasons. If I use an input file containing data from a single year I get different estimates than if I use an input file containing data from all years, even though I specified treatment and year to correspond to the simpler file. I am trying to figure out why this is the case.

Thanks.

John

[Rotella]

John,

As Gary has stated in posts here and MARK help file, and as is mentioned in the Nest-Survival chapter of Cooch & White's MARK book, it is important to be cautious with standardizing individual covariates for this data type. In some situations, e.g., where nest age on each day of the nesting season is entered as a covariate in the rows of the design matrix, DON'T standardize. In other situations, you could standardize, but ... if you do, you need to be careful in how you go about constructing/interpreting real parameter estimates.

Jay

[Rotella]

John,
A bit more on your questions:
1) There aren't any inherent problmes in specifying more than one covariate at a time in a model. In fact, the desire to do just that was a large part of the motivation for developing the more modern methods of nest-survival analysis.

2) You should be able to obtain the same estimates of daily survival rate for a given year regardless of whether or not the data for that year are contained in a file containing data from other years or just the data for the year of interest. I'm not sure why you might be running into problems here. But, one reason could be if the multi-year model contains parameters that share data from multiple years, e.g., a treatment effect held in common across years.

Jay

[jbruggin]

Jay,

Thanks for your responses. Standardization does not seem to be the issue because I get identical results whether or not I standadize the covariates. I think you may be correct about the problem being related to parameters that share data from multiple years being present in the multi-year model because if I remove "treatment" from the model, I do get identical estimates. If that is the problem, is there a way around it other than carving up the data set? Treatment and year both appear to be important variables, so obviously I am interested in generating year-specific estimates by treatment.

John

[egc]

John,

As Gary has stated in posts here and MARK help file, and as is mentioned in the Nest-Survival chapter of Cooch & White's MARK book, it is important to be cautious with standardizing individual covariates for this data type...

Actually, its important to be cautious - period (independent of data type). This is covered in some detail in the individual covariates chapter (Chapter 12).

[Rotella]

John,
If you want year-specific estimates and you want to include treatment effects, then you need to think about whether or not you want/need to estimate a common treatment effect across all years or to let the treatment effect vary by years. There is lots of good information on how to do this in MARK in other chapters, i.e., text on building competing models.
Jay