Geoscience Reference
In-Depth Information
8.8.2 Allowing for the Effect of a Flood
Suppose now that there is some interest in comparing the model just fitted
with a simpler one that says that the probability of recapture was the same
for all birds in all years but the probability of survival was one of two values.
For the periods 1982-1983 and 1983-1984, there was one survival probabil-
ity because of the effects of the flood, and a different survival probability
applied for the other years.
This model can be fitted using MRA-LGE when a
Y
variable is included in
the input data, which is 1 for the years affected by the flood and otherwise
0. Fitting the three-parameter model with a constant probability of capture
and two survival rates leads to a maximized log-likelihood of −330.05. The
log-likelihood test to compare the fit of this model with the one previously
fitted then gives
D
= 2{(−329.87) - (−330.05)} = 0.36, from Equation (8.24), with
7 - 3 = 4 df. On this basis, there is no evidence that the extra complexity of
the first model is needed.
8.8.3 Model Selection Using the QAIC
The computer program MRA36 can be used to fit the 36 models that were
defined in Section 8.6 to the data on the European dipper. It was found that
c
equaled 1.18 and the best model according to the QAIC is (
p
constant, ϕ con-
stant), that is, the probability of capture and the probability of survival were
the same for all birds at all times. Because the model allowing for a flood
effect is not one of the ones considered by MRA36, there is now some interest
in comparing the (constant, constant) model with the flood effect model. It is
then found that the QAIC suggests that the model with a flood effect allowed
is the most appropriate of the two considered.
8.9 Recent Advances with Open-Population Models
The topic
Handbook of Capture-Recapture Analysis
(Amstrup et al
.,
2005)
discusses many recent advances concerning the analysis of capture-recap-
ture data for open and closed populations, tag-recovery models, the joint
analysis of tag-recovery and live-resighting data, and multistate models
(in which animals move between several states while capture-recapture
sampling takes place). It also includes the discussion of the analysis of eight
data sets.
