U-CARE vs. Bootstrap results

[rbn]

Hello,

First, I'm starting with the analysis of capture-recapture so, I'm afraid I have rather newbie questions.

I have these results from the overall GOF test from U-CARE:

--------------------------------
Global TEST, number of groups =1
df =92
Quadratic Chi2 =56.4755
->P-level=0.99869
N(0,1) statistic for transient(>0) =2.9524
->P-level, two-sided test =0.0031529
->P-level, one-sided test for transience =0.0015765
N(0,1) signed statistic for trap-dependence =-0.85985
->P-level, two-sided test =0.38987
--------------------------------------

Perhaps I'm wrong but I assumed from these results that a 'transience' model (Phi(a2_t/t), p(t)) can be used as starting model for my dataset.

But when I perform a bootstrap GOF (100 simulations) with this model, it has a p<0.01 and all the c-hat values are 0.000.

First, I'm not sure if my interpretation of the U-CARE overall test is correct. If it is, I'm not sure how to continue. Should I search for another starting model that pass the bootstrap GOF?

Thanks in advance.

David

[abreton]

Although your results from UCARE suggest that you're not dealing with sparse data, c-hats of 0.00 from all 100 bootstraps suggests numerical problems. When you imported your inp file into MARK, did you include individual covariates? If so, try stripping them from the inp file and repeating the bootstrap assessment of the transience model.

[rbn]

Thanks very much for the answer,

Following your recommendation I stripped the input file so now it only contains encounters. This is an excerpt:

0000000000000000000000000000000000010000000000000000000000000000000000 1 ;
0000000000000000000000000000000001001000000000000000000000000000000000 1 ;

U-CARE shows these statistics, and seems that a 'transience' model fits well with the data...
--------------------------------
Global TEST, number of groups =1
df =92
Quadratic Chi2 =56.4755
->P-level=0.99869
N(0,1) statistic for transient(>0) =2.9524
->P-level, two-sided test =0.0031529
->P-level, one-sided test for transience =0.0015765
N(0,1) signed statistic for trap-dependence =-0.85985
->P-level, two-sided test =0.38987
--------------------------------


Then, with the same .inp file I run a model Phi(a2-t/t)p(t). This is part of the output:
--------------------------------
Number of Estimated Parameters {Phi(a2-t/t)p(t)} = 160
DEVIANCE {Phi(a2-t/t)p(t)} = 205.47480
DEVIANCE Degrees of Freedom {Phi(a2-t/t)p(t)} = -76
c-hat {Phi(a2-t/t)p(t)} = 0.0000000
AIC {Phi(a2-t/t)p(t)} = 1717.7731
AICc {Phi(a2-t/t)p(t)} = 1732.0882
Pearson Chisquare {Phi(a2-t/t)p(t)} = 1009568.0
--------------------------------

There are 3625 rows in the .inp file.

But bootstrap with 100 simulations still gives ALL the c-hats with 0.000 and the model has a p=0.01


I don't know where is the problem. I thought that U-CARE and Bootstrap should give similar results about the adequacy of the model to the dataset

If it can help, the PIM for the apparent survival is of this type:

1 70 71 72 ...
2 71 72 ...
3 72 ...
4 ...


The PIM for recapture parameter is:

138 139 140 ...
139 140 ...
140 ...




Thanks. I hope someone can help.

David