r estimate with Burnham model

[knewcomb]

Hello Forum Users:

To deal with my previous issue of telemetry subsamples, I changed my encounter history file so that 1 occasion represented 3 days instead of 1 day. (It took me 3 days to record locations on my entire sample.) My recapture estimate had been very low with the 1 day setup because of the subsamples. I setup uneven time intervals so that they are in terms of a 3 day period (i.e., 1 for 3 day interval, 1.3 for 4 day interval, etc). Additionally, it was suggested that I try the Burnham model instead of Barker's because it had fewer parameters.

So, my data now has 30 occasions and 62 individuals. I ran a Burnham model with all parameters held constant, but it was not able to estimate r. I believe this is because the estimate is very close to 1. (I tried using the sin link, which did a much better job than the logit link, but I've got individual covariates to include.) It did estimate r when I tried running the basic model with a hunt covariate included for S.

I am wondering if I should a) fix r=1, b) find recovery estimates for r in the literature and use them, c) do nothing, or d) abandon ship and try a known fate model.

Thoughts?

Thank you!!

[bacollier]

Hello Forum Users:

To deal with my previous issue of telemetry subsamples, I changed my encounter history file so that 1 occasion represented 3 days instead of 1 day. (It took me 3 days to record locations on my entire sample.) My recapture estimate had been very low with the 1 day setup because of the subsamples. I setup uneven time intervals so that they are in terms of a 3 day period (i.e., 1 for 3 day interval, 1.3 for 4 day interval, etc). Additionally, it was suggested that I try the Burnham model instead of Barker's because it had fewer parameters.

Why, with telemetry, are you estimating a recapture parameter at all? Its assumed 1 last I checked about the 4th line of the known-fate chapter? Why did you go to 3 days instead of 1, with telemetry/known fate data (which is what you seem to be dealing with) if you checked every day you would be fine to estimate daily survival (although a full time*g model might have a lot of dsr estimates of 1 at that time frame).

So, my data now has 30 occasions and 62 individuals. I ran a Burnham model with all parameters held constant, but it was not able to estimate r. I believe this is because the estimate is very close to 1. (I tried using the sin link, which did a much better job than the logit link, but I've got individual covariates to include.) It did estimate r when I tried running the basic model with a hunt covariate included for S.

I am wondering if I should a) fix r=1, b) find recovery estimates for r in the literature and use them, c) do nothing, or d) abandon ship and try a known fate model.

Thoughts?

Thank you!!

If the 'r' you mean is that the animal is found dead and reported, then why would you need to estimate that or 'find a value in the literature' when you have telemetry data so you know when they died and that all are reported? Do you have harvest you are trying to account for

In general, I am unclear on why you would even be using a Burnham/Barker model on the dataset you seem to be describing? What is your logic behind choosing that structure for your data, how is it going to better answer your question (which you have not explained?). Is it a fidelity question?

It seems like you are dealing with telemetry data, yet you are trying to apply the (moderately) complex live-dead modeling approach to and as such are fixing parameters to make it work? If it is survival, as a function of environmental covariates as you described in your other post that your interested in, then why are you not using a standard known fate analysis/ragged telemetry? If not, how are you dealing with the "00" values since you are using a LDLD format (as described in the known fate chapter), was this part of your 3 days of recoding? That information would perhaps allow for more responses from the list.

Bret

[knewcomb]

Bret,

Thanks for your response!

I apologize for being so vague in my postings. I’ll try to clarify things.

I opted for the Burnham/Barker’s model instead of the known-fate model for several reasons. I was not able to keep track of all radiomarked ducks and so did not know the fate of all individuals. I thought that I would have to do a lot of censoring, and the MARK book suggests estimating p and r parameters if that is the case. The study area was defined by the birds and changed somewhat throughout the season, so recapture could not be assumed to equal 1. I worked with black ducks during the winter, and I thought that it would be important to estimate recovery since they could be harvested. There were a few that were harvested and reported outside the general study region. Perhaps some of the missing birds were harvested and not reported or misidentified. Permanent and temporary emigration were certainly occurring. Though my original question focused on how hunting, weather, and body weight affect winter survival, I became interested in how these factors affected fidelity as well.

Unfortunately, I did not check the status of every bird every day in my first year. I only checked on the birds included in the random subsample drawn every day. It took 3 days for me to check/record locations on all radiomarked birds. The recapture probability was very low when I originally ran the model for daily survival, which biased my survival rate a little low. So, I combined 3 daily occasions into one 3 day occasion. If a bird was observed on any day in the 3 day period, then it was marked as “10”. If it was observed and then recovered dead, then it was marked as “11”. If it was not observed but recovered dead, then it was marked as "01". If it was not observed or recovered on any day, then it was marked as “00”.
Running the model again with the 3 day period and no parameters fixed except for the last recovery interval, survival (.99), recapture (0.82), and fidelity (0.95) did just fine. However, the real parameter estimate for recovery is 0.999984 with a confidence interval of practically 0 to 1. I'm wondering how or if to address this issue.

I hope that helps!
Kira