Unequal nesting seasons between years for nest survival?

[zugunruhe1]

Hi Everyone,

I have a question about how to set up nest survival data for 2 years, but with the length of the nesting season varying across years. In the first year, the nesting season (the interval I collected data over) was 69 days long, and in the second year, 108 days long. I want to have a file with both years in the same model, so that I can test if year had an effect on DSR and whether certain variables were stronger in both years combined rather than in analysis of separate years.

The way I approached this is for the number of encounters, I added 69 and 108 for a total of 177 encounters over both years. I used 2 lines of dummy variables for the years, as I’ve seen in the AA file for good vs. poor swift colonies and the European dipper file for males vs females. I used year as a group variable and have 8 other covariates besides the “standard” nest info (the input file below is truncated just to demonstrate how I coded year with a space between the 2 years):

/* Data for MODO nests

Covariates are:
1) day of nesting season nest found,
2) last day nest checked when alive,
3) last day nest checked,
4) fate of nest (0=successful),
5) number of nests with that encounter history,

6) age of nest on first day of nesting season
7) VOR
VOR at nest
9) %grass
10) Distance to solar facility
11) Distance to highway
12) Distance to two-track
13) Distance to gravel road
14) Year (dummy variable;1 if in 2014)
15) Year (dummy variable; 1 if in 2015)*/

Nest Survival Group=1;

/*1*/ 1 12 15 1 1 12 5 10 52.22 780 594 174 1442 1 0;
/*5*/ 8 10 13 1 1 -6 10.71 20 44.45 1343 333 870 615 1 0;
/*6*/ 14 14 15 1 1 2 4.29 5 36.67 1422 573 987 1551 1 0;
/*8*/ 15 19 23 1 1 -1 3.57 15 35.56 696 2255 601 863 1 0;

/*2.1*/ 1 3 6 1 1 14 7.86 15 41.07 159 774 159 491 0 1;
/*2.2*/ 3 5 6 1 1 -1 10.71 15 39.96 403 750 114 258 0 1;
/*2.3*/ 7 9 10 1 1 -6 7.14 10 47.73 488 997 328 1349 0 1;
/*2.4*/ 11 29 30 1 1 -10 7.86 5 41.07 1175 1672 76 970 0 1;

The problem is when I open the PIMs for the different years, they both go all the way to 177, whereas I’m trying to get 2014 to just go to 69. All the examples worked had the same number of encounter histories for both groups so I’m not sure how to go about this. I would greatly appreciate any advice!

[Rotella]

One way to handle your situation is to have the number of occasions be equal to the number of days of the year for which you have nest-survival data over the 2 years: start with your earliest calendar date in either year as day 1 and end with the latest calendar date in either year as the last date. If you do this, the resulting PIM cells will correspond to days of the year, which can be convenient for setting up various models of interest. Imagine that you have 110 total occasions (days of the year) that are used in at least 1 year across the 2 years. If you enter the data as being from 2 groups (1 for each year), you will have 2 PIMs (1 for each year). If you leave the PIMs numbered so that each day for each group is numbered uniquely (time*group), you can then use the design matrix to evaluate various models of interest quite readily using the various linear modeling tools described throughout Cooch & White chapters. The PIM for year 1 (and perhaps year 2 depending on your circumstances) will have occasions represented for which you have no observations. This is fine, but if you run a model that tries to estimate daily survival uniquely for each day in year 1, you won't be able to estimate the rate for those days for which you have no observations. If you run a model that pools across years and estimates daily survival uniquely for each day (without regard to year), you'll get estimates for each day based on whatever data you have across the 2 years. Of course such general models probably won't perform very well or provide very useful results for typical datasets.

[zugunruhe1]

Thank you! I'll give that a shot.

[zugunruhe1]

I realized I have a follow-up question: if I am using the 2-group approach, would I need to adjust the age of nest relative to first day of nesting season for the year in which nesting started later in the season? Or is that not necessary because I will be entering 0's in the DMs for days in the year that do not have observations? I guess more specifically I'm asking how MARK will "know", or how I'll tell it, which days there are observations missing for.

By this statement:

If you run a model that pools across years and estimates daily survival uniquely for each day (without regard to year), you'll get estimates for each day based on whatever data you have across the 2 years. Of course such general models probably won't perform very well or provide very useful results for typical datasets.

are you saying the models won't be very informative and/or accurate unless I include year in each of them?

Thanks again.

[Rotella]

if I am using the 2-group approach, would I need to adjust the age of nest relative to first day of nesting season for the year in which nesting started later in the season? Or is that not necessary because I will be entering 0's in the DMs for days in the year that do not have observations? I guess more specifically I'm asking how MARK will "know", or how I'll tell it, which days there are observations missing for.

No. You would want to use the earliest date from either season as your reference date. You are explicitly telling MARK which days you have observations for (or don't) by providing your encounter histories.

If you run a model that pools across years and estimates daily survival uniquely for each day (without regard to year), you'll get estimates for each day based on whatever data you have across the 2 years. Of course such general models probably won't perform very well or provide very useful results for typical datasets.

are you saying the models won't be very informative and/or accurate unless I include year in each of them?

No. I was trying to point out that such a model would be heavily parameterized, i.e., trying to estimate daily survival rate uniquely for each day of the nesting season with no constraint or pattern imposed. Thus, for most data sets I've seen, it would be unlikely that such a model would provide useful estimates. I think it's more typical to impose constraints and run models that constrain daily survival rate in some way. For example, you might constrain daily survival rate to follow a trend (e.g., linear or quadratic on logit scale) over the season. You could then run competing models that evaluate, for example, whether such patterns appear similar across years, whether models with such trends are more supported than models that hold daily survival rates constant across all days, years, etc. Of course, the set of competing models would depend heavily on the questions you have and the details of your dataset.

[zugunruhe1]

Ok, I *think* that clears things up! Just to be absolutely sure before I run the models, if my 1st day a nest was found for year 2 is day 15 relative to year 1 and its "nest age" was -12 relative to year 1, I should still keep the encounter history for that nest in year 2 as day 1 for day found and 2 for nest age?

Thank you again, this has been very helpful!

[Rotella]

if my 1st day a nest was found for year 2 is day 15 relative to year 1 and its "nest age" was -12 relative to year 1, I should still keep the encounter history for that nest in year 2 as day 1 for day found and 2 for nest age?

I recommend that you number days of the year the same way in each season. To do so, start with your earliest calendar date in either year as day 1 and end with the latest calendar date in either year as the last date. Thus, for your example, the nest in year 2 would be entered as having been found on day 15. And, if I followed your text above correctly, it's age on day 1 (which is the earliest date in either season), would be -12.

[zugunruhe1]

I’m back… I apologize that I keep having more questions! So when I take the approach of doing 2 groups, and uniquely numbered PIMs for each group, I get numbers 1-118 for the group representing year 1 and numbers 119-236 for the group representing year 2. When I am running a model involving year and different survival each day, would I need to handle the cells for which I do not have observations (specifically, days 70-118 for year 1, and days 1-11 for year 2) any differently from just coding them “1” for one year from cells 1-118 and “0” for the other year from cells 119-236? Similarly, if I am running a model involving age, would I leave the age cells blank for days of years that did not have any active nests, or would I fill in what the age would have been if there was an active nest?

[Rotella]

I think that your questions are a bit difficult to fully answer in the forum. If you'd like, you can feel free to contact me at rotella@montana.edu.