Overall Analysis Questions: Modeling behavior, clusters& sex
Posted: Mon Apr 13, 2015 6:05 pmI’ve been analyzing my capture data for the past couple of weeks and I have a couple of analysis questions that I’m hoping to get some insight on. To start with, I’ll describe the set-up of my system and then how I set things up in SECR.
I have 7 years of mark-recapture data for foxes. The trapping array consists of 216 traps, configured into 18 “mini-grids” that consist of 12 traps each. The trapping array is consistent across years and each mini-grid is trapped for 6 consecutive days. My data consists of both male and female adult foxes and my overall goal is to estimate density separately for each of the 7 years.
To begin with I set up my capture file with year and sex as sessions (i.e. 14 sessions total: 2008 male, 2008 female, 2009 male, etc.) and only 6 total occasions (not 42 occasions: 6 annual occasions x 7 years).
1. My first question is about modeling behavior (b or B). Since I have set up my capture file with only 6 occasions, it would only make sense to model behavior (b or B) when “year” is also modeled (i.e. D(yr) g(yr+b) s(yr+b)). If I understand how SECR models behavior it would not make any sense to model behavior across all years (D(yr)g(b)s(b)). Since SECR cannot tell between years in this case it would model behavior as the difference between the first capture and all subsequent captures, as determined by the capture occasion. So if I modeled behavior across years it would be possible for SECR to determine an initial capture as occasion 1 in the year 2010 and the subsequent capture as occasion 4 in year 2008, which would not accurately reflect “initial” and “subsequent” captures. Currently I only modeled behavior in conjunction with year.
2. My next question is about the use of clusters. I have dabbled with this a bit but not had much success. My data seems set up for clusters, seeing as there are actually 18 different mini-grids within the larger trapping array. In my case foxes are only trapped on a single mini-grid within each year but there are some foxes that are trapped on 2 or more grids but never within the same year, always between years. Since I am modeling density by year do you feel the estimates of density with and without the “cluster” analysis would be different at all? Since there is no movement of foxes between grids within each year and density is modeled annually I’m under the impression that the annual estimates would not differ much at all weather the “cluster” command was or was not used. When I include clusters in my analysis, the analysis time is prohibitively long.
3. My final question is about modeling differences in sex and it may be an extremely simple answer. To begin with I set both year and sex as sessions and then created a covariates dataframe:
covariates=data.frame(constant=c("a","a","a","a","a","a","a","a","a","a","a","a","a","a"),
year = c("a","a","b","b","c","c","d","d","e","e","f","f","g","g"),
sex = c("a","b","a","b","a","b","a","b","a","b","a","b","a","b"))
I then entered this dataframe in the secr.fit command (sessioncov=covariates). The top model in my model set was D(yr)g(yr+sex+b)s(yr+b) with 85% of the weight. The density estimate output for this top model is separated by all 14 sessions, giving me a density estimate for males and for females separately for each year, but the estimate for males and females is the same since the model did not vary density by sex. This is an incredibly elementary question but would “adult” fox density be equal to both the male and female density estimates or do I need to double the estimates to account for both males and females?
As a test I wanted to run the “same” model but set it up in a different way. I used the exact same data and set up my capture file to have 7 sessions that correspond to each year and I modeled sex as an individual covariate. I then ran the exact same model D(yr)(g(yr+sex+b)s(yr+b) but now year = session and sex is an individual covariate and I must use the conditional likelihood. When I derive density estimates for each year, the output only gives me the density of females for each year, not males. How would I derive estimates for the males? Since density doesn’t vary by sex I assume that male density is the same as female density. The odd thing is that these density estimates are much higher than in my initial analysis, and in fact they are almost doubled! Again my main goal is to estimate overall adult fox density annually and I’m wondering if the density estimates it outputs for females is the same as adult density? I’m just a bit confused as to why the my initial estimates, which are the same for both male and female, are half the value of the estimates derived from the individual sex covariate analysis.
The other issue is that I feel strongly that my initial estimates are correct, and the estimates derived from the sex as a covariate analysis are twice as high as they should be for “adult” fox density. I feel like I’m missing something quite simply but would appreciate a bit of input.
Thanks again,
Adam
I have 7 years of mark-recapture data for foxes. The trapping array consists of 216 traps, configured into 18 “mini-grids” that consist of 12 traps each. The trapping array is consistent across years and each mini-grid is trapped for 6 consecutive days. My data consists of both male and female adult foxes and my overall goal is to estimate density separately for each of the 7 years.
To begin with I set up my capture file with year and sex as sessions (i.e. 14 sessions total: 2008 male, 2008 female, 2009 male, etc.) and only 6 total occasions (not 42 occasions: 6 annual occasions x 7 years).
1. My first question is about modeling behavior (b or B). Since I have set up my capture file with only 6 occasions, it would only make sense to model behavior (b or B) when “year” is also modeled (i.e. D(yr) g(yr+b) s(yr+b)). If I understand how SECR models behavior it would not make any sense to model behavior across all years (D(yr)g(b)s(b)). Since SECR cannot tell between years in this case it would model behavior as the difference between the first capture and all subsequent captures, as determined by the capture occasion. So if I modeled behavior across years it would be possible for SECR to determine an initial capture as occasion 1 in the year 2010 and the subsequent capture as occasion 4 in year 2008, which would not accurately reflect “initial” and “subsequent” captures. Currently I only modeled behavior in conjunction with year.
2. My next question is about the use of clusters. I have dabbled with this a bit but not had much success. My data seems set up for clusters, seeing as there are actually 18 different mini-grids within the larger trapping array. In my case foxes are only trapped on a single mini-grid within each year but there are some foxes that are trapped on 2 or more grids but never within the same year, always between years. Since I am modeling density by year do you feel the estimates of density with and without the “cluster” analysis would be different at all? Since there is no movement of foxes between grids within each year and density is modeled annually I’m under the impression that the annual estimates would not differ much at all weather the “cluster” command was or was not used. When I include clusters in my analysis, the analysis time is prohibitively long.
3. My final question is about modeling differences in sex and it may be an extremely simple answer. To begin with I set both year and sex as sessions and then created a covariates dataframe:
covariates=data.frame(constant=c("a","a","a","a","a","a","a","a","a","a","a","a","a","a"),
year = c("a","a","b","b","c","c","d","d","e","e","f","f","g","g"),
sex = c("a","b","a","b","a","b","a","b","a","b","a","b","a","b"))
I then entered this dataframe in the secr.fit command (sessioncov=covariates). The top model in my model set was D(yr)g(yr+sex+b)s(yr+b) with 85% of the weight. The density estimate output for this top model is separated by all 14 sessions, giving me a density estimate for males and for females separately for each year, but the estimate for males and females is the same since the model did not vary density by sex. This is an incredibly elementary question but would “adult” fox density be equal to both the male and female density estimates or do I need to double the estimates to account for both males and females?
As a test I wanted to run the “same” model but set it up in a different way. I used the exact same data and set up my capture file to have 7 sessions that correspond to each year and I modeled sex as an individual covariate. I then ran the exact same model D(yr)(g(yr+sex+b)s(yr+b) but now year = session and sex is an individual covariate and I must use the conditional likelihood. When I derive density estimates for each year, the output only gives me the density of females for each year, not males. How would I derive estimates for the males? Since density doesn’t vary by sex I assume that male density is the same as female density. The odd thing is that these density estimates are much higher than in my initial analysis, and in fact they are almost doubled! Again my main goal is to estimate overall adult fox density annually and I’m wondering if the density estimates it outputs for females is the same as adult density? I’m just a bit confused as to why the my initial estimates, which are the same for both male and female, are half the value of the estimates derived from the individual sex covariate analysis.
The other issue is that I feel strongly that my initial estimates are correct, and the estimates derived from the sex as a covariate analysis are twice as high as they should be for “adult” fox density. I feel like I’m missing something quite simply but would appreciate a bit of input.
Thanks again,
Adam