Adding Covariates- Design Matrix or Encounter History?

[ctlamb]

I have an encounter history with the following format-

/*bear_name*/ 00101010001010100010011000100 1 0;

With the last two columns denoting sex, M/F.

My dataset spans 7 years, with 4 sessions/year. I would like to add annual covariates to my analysis to elucidate the effect of bottom-up vs. top-down factors on population growth. For my pilot analysis, I would like to try annual variation in huckleberry abundance (a subjective annual huckleberry measure ranging from 1-5) and total female mortality (also an annual measure).

Do I create a design matrix for these? I have read up to Ch.5 in the "gentle introduction" book, but I have stumbled with Ch.6, it is getting the better of me. So, before I power through it, I just wanted to confirm that I do in fact need to make a D.M..

Thanks everyone!
C

[bacollier]

I have an encounter history with the following format-

/*bear_name*/ 00101010001010100010011000100 1 0;

With the last two columns denoting sex, M/F.

My dataset spans 7 years, with 4 sessions/year. I would like to add annual covariates to my analysis to elucidate the effect of bottom-up vs. top-down factors on population growth. For my pilot analysis, I would like to try annual variation in huckleberry abundance (a subjective annual huckleberry measure ranging from 1-5) and total female mortality (also an annual measure).

Do I create a design matrix for these? I have read up to Ch.5 in the "gentle introduction" book, but I have stumbled with Ch.6, it is getting the better of me. So, before I power through it, I just wanted to confirm that I do in fact need to make a D.M..

Thanks everyone!
C

if your covariates apply to all individual equally, which I think from your description above they do, then you can just put them in the DM as shown in the MARKBOOK in the appropriate place. Personally I prefer things in the .inp just for documentation simplicity, but either way will work.

I do wonder what type of analysis you are running where you are putting a annual estimate of mortality into a estimation approach that looks like it would be able to estimate mortality for you (seems you have a open RD design)? Also, be sure you read up on the potential issues with using categorical values (huckleberry abundance, 1, 2,3...) in a regression format as a continuous, not factor, metric. There are a few posts on the list in that regard, and its in the book.

\bret

[ctlamb]

Thanks for your reply. Is the MARKBOOK you are referring to the one by Cooch and White (http://www.phidot.org/software/mark/docs/book/), this is the one I'm working through.

I do wonder what type of analysis you are running where you are putting a annual estimate of mortality into a estimation approach that looks like it would be able to estimate mortality for you (seems you have a open RD design)? Also, be sure you read up on the potential issues with using categorical values (huckleberry abundance, 1, 2,3...) in a regression format as a continuous, not factor, metric. There are a few posts on the list in that regard, and its in the book.

I am VERY open to suggestions. I am newly exploring MARK so my initial approach may have some flaws. I want to use an open model though for sure.

The mortality data I will be entering is human-caused mortality as this is a hunted population and all harvests require reporting to the gov't, which is where I obtained this data. I'm interested if hunter harvest affects overall population growth.

Thanks for the heads up about the huck data. Can you direct me to the chapter this issue is addressed in?

Cheers,
C

[bacollier]

Thanks for your reply. Is the MARKBOOK you are referring to the one by Cooch and White (http://www.phidot.org/software/mark/docs/book/), this is the one I'm working through.

I do wonder what type of analysis you are running where you are putting a annual estimate of mortality into a estimation approach that looks like it would be able to estimate mortality for you (seems you have a open RD design)? Also, be sure you read up on the potential issues with using categorical values (huckleberry abundance, 1, 2,3...) in a regression format as a continuous, not factor, metric. There are a few posts on the list in that regard, and its in the book.

I am VERY open to suggestions. I am newly exploring MARK so my initial approach may have some flaws. I want to use an open model though for sure.

Cheers,
C

Well, for instance, think about your huck data such as this, lets say that over the 7 years of your study, you have 3 years of huck that is moderate = 1, 2 years of huck that is good = 1, and 2 years of huck that is bad = 3; you really don't need a covariate for that as huck & time are potentially confounded if say good huck years are good rain years (or vice versa, or whatever). So, for instance, thinking about a simple set of comparisons here for huck, you could implicitly evaluate that through time in the PIMs, for instance:

Constant (no time effect) your annual PIMS would be coded something like 1111111, e.g., all annual estimates constrained to be the same

Time effect: 1234567: all annual estimates different

Huck effect: 1331212: annual estimates differ by huck ranking for that year

then the question (in this extremely simple example) would be, which model fits best give the data based on AIC or whatever.

The mortality data I will be entering is human-caused mortality as this is a hunted population and all harvests require reporting to the gov't, which is where I obtained this data. I'm interested if hunter harvest affects overall population growth.

Well, it can't not impact it, right (I may have just tossed a grenade here given the range of folks that follow this list, stand down everyone)...

So, what parameters do you expect are impacted by harvest mortality? I mean, in a open model you are estimating population size/recruitment, survival, recapture prob (and whatever else depending on the model), so what parameter are you thinking about modeling as a function of harvest mortality? Maybe for recruitment modeling in a Pradel model, but I don't think modeling the survival parameter in a open population model as a function of female mortality would work, but other approaches to looking at the survival/recruitment tradeoff are available in MARK that would work better. Maybe someone else has a different opinion...

\bret

[ctlamb]

Agreed on the huck data. Thanks for clearing that up.

So, what parameters do you expect are impacted by harvest mortality? I mean, in a open model you are estimating population size/recruitment, survival, recapture prob (and whatever else depending on the model), so what parameter are you thinking about modeling as a function of harvest mortality? Maybe for recruitment modeling in a Pradel model, but I don't think modeling the survival parameter in a open population model as a function of female mortality would work, but other approaches to looking at the survival/recruitment tradeoff are available in MARK that would work better. Maybe someone else has a different opinion...

Yes, clearly survival will be impacted by harvest (by definition, really). I am moreso interested in modelling population growth rates between years. I have not made it to Ch.6 in the book so far and now understand the design matrix, but have not dug too deep into the various models (the book is an amazing resource, but the brain can only absorb so much info/day!) My end goal would be to model 8-9 different covariates (female, male, and total mortality; huckleberry production and abundance; climate variables) in my general model, with the intention of elucidating their affect on population growth rates. Is this possible?