analysis with LOTS of individual covariates

[dtempel]

I have a 21-year data set that I need to analyze for individual survival. I have a lot of individual time-varying covariates (annual amount of different habitat types present at an individual's location, annual amount of timber harvested at an individual's location, etc.). Given that MARK has a 240-character limit for each record in the encounter histories file, this will be very difficult to do in MARK. Essentially, I would need to create multiple data input files, run separate analyses for each input file, and then combine the model results in Excel.

I've never used RMark, but I'm wondering if it has the capability to handle data with many covariates? Does anyone have any recommendations for an alternative software to use for my analyses?

Thanks!

[ganghis]

Hi,
RMARK basically just uses R to simplify model construction (well if you know what you're doing) and then passes everything to R. So I don't think that will get you around the character limit. Gary might be able to increase it for you, but another option you might look into is marked, which was formulated in part to deal with large numbers of time-varying individual covariates. Currently only a few MR models have been implemented, but CJS is one of them. Here's a link to the accepted article:

http://onlinelibrary.wiley.com/doi/10.1 ... 5/abstract

Models are formulated in a similar manner to RMARK, so you'll probably want to read help materials for RMARK in addition to the the article and vignette for marked. It's up on CRAN and ready to be downloaded.

-Paul

p.s. if you need help, ask Jeff

[bacollier]

Hi,
RMARK basically just uses R to simplify model construction (well if you know what you're doing) and then passes everything to R. So I don't think that will get you around the character limit. Gary might be able to increase it for you, but another option you might look into is marked, which was formulated in part to deal with large numbers of time-varying individual covariates. Currently only a few MR models have been implemented, but CJS is one of them. Here's a link to the accepted article:

http://onlinelibrary.wiley.com/doi/10.1 ... 5/abstract

Models are formulated in a similar manner to RMARK, so you'll probably want to read help materials for RMARK in addition to the the article and vignette for marked. It's up on CRAN and ready to be downloaded.

-Paul

p.s. if you need help, ask Jeff

Paul beat me to it. I think the you would most likely want to use the crm() wrapper function in marked that is quite similar to how the mark() call works in RMark. Best I can tell it handles time-dependent covariates the same way at RMark/MARK does.

FYI, Jeff is not going to be to near email for a few weeks, so if you need some help or get stuck I will give you a hand, just let me know.

\bret

[dtempel]

Thanks a lot for your help. I have a couple more questions. I have capture histories that I have to truncate when the individual moves to a new territory (i.e., -1 in the frequency column if using MARK). Can 'marked' handle this situation?

Also, my age modeling is a little more complicated than just the age of the individual. I'd like to model 1- and 2-year-old individuals with a different survival rate than individuals that are 3 or more years old. Thus, age would be a categorical, time-varying covariate. I can do that in MARK by restructuring the PIMs, but in 'marked' I guess I'll have to create an individual covariate where 0 = 1 or 2 years old, 1 = 3+ years old. The values for each individual will depend on the age at initial capture (most were already 3+ years old at initial capture). Is that the simplest approach?

[dtempel]

I'll add that with individual covariates, I wouldn't necessarily need to truncate capture histories, but one benefit to doing so is that I don't have to make any assumptions when an individual changes location after one or more 0's in the capture history.