by mia.w » Tue Feb 21, 2012 4:03 am
Hi,
I am having a similar problem with some of my models.
I work on presence/absence data of lactating fur seal females. When they are on land with their pups and when they are at sea foraging. I have two states (on land and at sea), 1 age class, 18 groups (3 groups per year where the groups are pup sexes, i.e. female in year1, male in year 2, unknown sex year 1; female in year 2, male in year 2, unknown sex in year 2 etc.
The number of observations is 152, which consists of 2 seasons - summer and winter.
Survival is based on seasons. I know from my data that a female survived within a season but not between.
So I programmed it like this
Survival: t(1:59,60,61:151) where t(1:59) is summer and t(61:151) is winter. These I set as constant to 1 with IFVF. t(60) I let the model estimate. As I was explained by someone that uses M-surge regularly, this will allow for females to "die" between seasons.
Based on AIC values and parameter estimates the best explanation for Transition is state.season.year+sex in other words, I coded it like this.
Transition: from.to.t(1:59,60:151).g(1:3,4:6,7:9,10:12,13:15,16:18)+ g(1 4 7 10 13 16, 2 5 8 11 14 17, 3 6 9 12 15 18)
Now I am trying to find the best model using the aforementioned values/parameters for survival and transition to find the best parameters for capture probability by trying different combinations of state, season, sex and year.
Obviously when the female is at sea the capture probability is 0. So my basic 'state' coding is like this:
to(2)+to(1) where I set to(2) in IFVF to zero. But now when I make the capture probability somewhat more complex by adding more parameters I either get a error message "OUT OF MEMORY" or with some other ones "Failling with linesearch" I only manage to include one or two parameters for capture probability. I.e using sex or year, but not both.
I have tried your suggestion by changing my initial values to "Random". I have also installed the latest MCR 7.16 and appropriate M-surge, but still no luck.
Any suggestions? Would running the models on a computer with more memory resolve this problem?
Thank you,
Mia