Lukacs Young or Known fate? and Inp questions

[kernicholson]

A few questions here for clarification:
Can I use the Lukacs Young model if I have "known" fates? We followed radio-collared female moose and counted their calves but did not mark the calves. If we did not see the calves we would spend longer time searching and then return the next 2 days searching again. Thus the confidence level of having observed the calf if it was there and not getting a false negative is pretty darn high. Considering the close bond behavior of moose mamma and baby.

Also, the frequency count at the end of the capture history - This is simply a count of how many animals had this exact capture history AND the exact same co-variates associated with that capture history? I do not have to compile those do I? Can I choose to leave each individual as its own entry line even if there are more than 1 with that exact same history and co-variates?

How do I indicate groups in the inp file? I have a column immediately after my frequency that is a 0 or 1 to indicate a group, but I am not understanding where MARK recognizes this column as the "grouping" variable. Number of groups - My groups are 2 the period before wolves (approx 5 years of data) and the period after wolves (approx 5 years of data). Does MARK automatically assume that the column immediately after their frequency column (the one with -1 or 1 to indicate censoring) that this column will be the "group" column? And then, its asking how many groups - well, is that asking I have 1 grouping variable "wolf presence" which has 2 options (before/after), or is it asking for the 2 options? Because maybe I have another grouping variable like sex or age class.

Then it asks for co-variates. Right now Ive 3: data regarding year of this survey, year when they first captured that mother, and age class of mother at that survey - I will eventually get other co-variates such as habitat quality and browsing pressure. But for now, this is it. So, when it asks for # of covariates, is it counting the columns over from the end of the capture history where the ";" is or starting from the frequency count/capture history and saying - ok in order of input, 1. capture history, 2. frequency, 3. "groups", anything beyond is covariates and we have 3 of them?

My last question is advise how to code some of these co-variates. Lets use Age Class as an example. There are 6 classes. I could code each class as a binary value which would allow me to test for a particular influence of 1 age class over the others - but I do not want this, what Im really after is just does age have any effect. Likely the middle age classes will have more surviving calves than the younger or older. So, I would just keep this as 1 categorical variable column instead of multiple binary columns? I have the same issue with year.

Ive been reading chapter 6, 4 and 19 in detail as I sit here and walk through each example. But for some reason am getting confused and just need some clarification. If there is some part in the book I should read even more carefully - please suggest it.

[bacollier]

A few questions here for clarification:
Can I use the Lukacs Young model if I have "known" fates? We followed radio-collared female moose and counted their calves but did not mark the calves. If we did not see the calves we would spend longer time searching and then return the next 2 days searching again. Thus the confidence level of having observed the calf if it was there and not getting a false negative is pretty darn high. Considering the close bond behavior of moose mamma and baby.

Not knowing how high "pretty darn high" actually is, I would suggest that you are probably ok not using Lukacs model and just going with known fates. It may be heresy to some to say that, but realistically, I suspect that if they are alive, you are seeing them, so you would not really gain much using a more complex modeling approach. This might introduce some bias, but I cannot see it being very much.

Also, the frequency count at the end of the capture history - This is simply a count of how many animals had this exact capture history AND the exact same co-variates associated with that capture history? I do not have to compile those do I? Can I choose to leave each individual as its own entry line even if there are more than 1 with that exact same history and co-variates?

You can leave them individually, MARK will aggregate them for you.

How do I indicate groups in the inp file? I have a column immediately after my frequency that is a 0 or 1 to indicate a group, but I am not understanding where MARK recognizes this column as the "grouping" variable. Number of groups - My groups are 2 the period before wolves (approx 5 years of data) and the period after wolves (approx 5 years of data). Does MARK automatically assume that the column immediately after their frequency column (the one with -1 or 1 to indicate censoring) that this column will be the "group" column? And then, its asking how many groups - well, is that asking I have 1 grouping variable "wolf presence" which has 2 options (before/after), or is it asking for the 2 options? Because maybe I have another grouping variable like sex or age class.

Yes, MARK recognizes it, most of your above necessitates some understanding of how design matrices work, but the typical 'standard' is to put your grouping values after your frequency columns. Read up on the linear models chapter.

Then it asks for co-variates. Right now Ive 3: data regarding year of this survey, year when they first captured that mother, and age class of mother at that survey - I will eventually get other co-variates such as habitat quality and browsing pressure. But for now, this is it. So, when it asks for # of covariates, is it counting the columns over from the end of the capture history where the ";" is or starting from the frequency count/capture history and saying - ok in order of input, 1. capture history, 2. frequency, 3. "groups", anything beyond is covariates and we have 3 of them?

Its 3 based on your above, but since 1 or 2 of those have some inherent time dependency (age of mom, maybe year of capture) that you will likely want to use when modeling calf survival if you think older mom's have better surviving calves, etc.

My last question is advise how to code some of these co-variates. Lets use Age Class as an example. There are 6 classes. I could code each class as a binary value which would allow me to test for a particular influence of 1 age class over the others - but I do not want this, what Im really after is just does age have any effect. Likely the middle age classes will have more surviving calves than the younger or older. So, I would just keep this as 1 categorical variable column instead of multiple binary columns? I have the same issue with year.

You should probably look into age/cohort (chapter 7) and the tsm models

\bret

[kernicholson]

A few questions here for clarification:
Can I use the Lukacs Young model if I have "known" fates? We followed radio-collared female moose and counted their calves but did not mark the calves. If we did not see the calves we would spend longer time searching and then return the next 2 days searching again. Thus the confidence level of having observed the calf if it was there and not getting a false negative is pretty darn high. Considering the close bond behavior of moose mamma and baby.

Not knowing how high "pretty darn high" actually is, I would suggest that you are probably ok not using Lukacs model and just going with known fates. It may be heresy to some to say that, but realistically, I suspect that if they are alive, you are seeing them, so you would not really gain much using a more complex modeling approach. This might introduce some bias, but I cannot see it being very much.

The concern is (though I can't explain why/how) that Lukacs model assumes <100% detection probability. Here with our "known" fate (and "pretty darn high" is 95% confidence) means we confidently assume 100% detection. The bonus from using Lukacs is that Ive got everything else to meet the Lukacs needs - I have unmarked young, we follow the marked radio-collared mother and Ive got more than 1 individual in my "brood" count. So, is there a way to get a combination of Lukacs and Known fate? If I use the known fate, I feel Ive 2 options:

1. take a capture history that was something like 020201 and split it into 2 capture histories of 010101 and 010100
or
2. make it all or nothing where all calves have to survive to the end. a capture history like 020201 would then become 010100

Neither is correct although option 1 seems "better" than option 2. Which option is better statistically?

[Eric Janney]

Kerry,

Why not use Lukacs model and run a model to see just how close to 1.0 your p's are? If they are really close to 1.0, they will likely be on the boundary and thus not estimated. If this is the case, you can simply fix the p's to 1.0? This basically makes it a "known fate" but still allows you to accomodate the data type you have (i.e., individually marked parent with unmarked young). Also, if your p's are 1.0 you will be able to reasonably assume that the confounded parameter (the phi and p form your last interval) is actually survival.