Overall GOF non-significant, but trans&trap tests signficant

[louise.fairless]

Dear all,

I have a small query regarding GOF testing in U-care. I have searched the U-CARE and the MARK manuals and this forum for answers, but have not found an answer to my question.

A similar question was asked by another forum member, but was entered into the MARK statistics help section
and there doesn't seem to be an answer. [GOF in UCARE - do the directional tests override the chi2?
by cathrumm » Fri Oct 24, 2003 4:39 am]

My question is:

I have run a 'sum of all tests' on data comprising of 12 groups (3xcolony, 2xsex and 2xage of marked [3x3x2=12 groups]).
I am therefore testing the GOF of the global model phi(c*s*m*t)p(c*s*m*t) c=colony, s=sex, m=age at marking, t=time.

When I run the sum of all tests (Test 3.SR+Test 3.SM+Test 2.Ct+Test 2.Cl). The test is highly non-significant (results below),
suggesting that this is an appropriate starting model for these data. However, the transience and trap dependence tests outputted with the overall test show significance.

My results are below:

Global TEST, number of groups =12
df =359
Quadratic Chi2 =309.6936
->P-level=0.97172

N(0,1) statistic for transient(>0) =4.3118
->P-level, two-sided test =1.6192e-005
->P-level, one-sided test for transience =8.0958e-006
N(0,1) signed statistic for trap-dependence =-7.3558
->P-level, two-sided test =1.8985e-013

Therefore, do these transience and trap dependence results mean that I cannot use the above model as a starting
model. Or, does the overall chi squared test override these and can I therefore use phi(c*s*m*t)p(c*s*m*t) as a
starting model for further survival and recapture analysis.

This may be a trivial question to experts, but please do not think that I have not tried to find the answer myself!
All examples I have found in the literature have shown that both the overall chi-squared test and the transience
and trap dependence tests are non-significant.

Your help with this would be much appreciated.

Best wishes,

Louise

[CHOQUET]

Hello,
I have no reply to this problem.
My recommendation is to look carefully each group separately, and to consider
for each group the four available tests. Results are perhaps different for each group.
Sincerely,
Rémi choquet.

[louise.fairless]

Thank you for your reply. I will closely investigate each group and see if I can identify the problem.

Best wishes,

Louise