Geoscience Reference
In-Depth Information
TABLE 7.4
Estimated Size of the Hypothetical Population Summarized in Table 7.3
Under Different Logistic Models Fitted by Huggins's Method
Model
Estimate
SE
AIC
95% CI
M*
0
30.00 1.10 201.15 (29.18, 34.72)
M*
t
29.96 1.09 207.16 (29.16, 34.73)
M*
b
30.82 2.08 202.46 (29.30, 39.89)
M*
h
30.00 1.08 203.14 (29.18, 34.63)
M*
tb
30.14 1.96 209.14 (29.12, 40.37)
M*
th
29.96 1.09 209.15 (29.16, 34.73)
M*
bh
30.82 2.08 204.45 (29.30, 39.19)
M*
tbh
30.15 1.89 211.13 (29.12, 39.85)
Note:
With the exception of
M
*
0
,
M
*
t
,
M
*
b,
and
M
*
tb
, fitted models included the
categorical covariate “sex.” SE, standard error; AIC, Akaike's information
criterion; CI, confidence interval.
of the output of the program CARE-2 (Chao and Yang, 2003). Sex was
also recorded as a dummy variable for each animal and included in the
analysis as an individual covariate.
The application of the Huggins method to these data leads to different
estimates of
N
for the different logistic models, as shown in Table 7.4. The
smallest AIC was attained for the simplest model M*
0
, indicating that
no temporal, behavioral, or individual heterogeneity effects were pres-
ent in the capture histories. The population size estimate under model
M*
0
, obtained from Equation (7.6), is 30 with a 95% confidence interval of
(29.18, 34.72).
7.4 Recent Advances for Closed-Population Models
New methodologies for closed-population models were discussed by Chao
and Huggins (2005a, 2005b), with detailed descriptions of methods illustrated
using the two examples described previously in this chapter. This includes
estimation involving bootstrapping, improved confidence intervals, sample
coverage methods, estimating equation methods, and generalized linear mod-
els. Table 4.1 in the work of Chao and Huggins (2005b) gives a complete list of
more recent methods, including Bayesian methods, parametric modeling of
heterogeneity, latent class models, mixture models, and the use of nonpara-
metric ML. The introduction of covariates to account for variation in probabili-
ties of capture has been an important avenue of research based on conditional
likelihood theory and generalized linear models (Huggins and Hwang, 2011).
Appendix A in Amstrup et al
.
(2005) provides a list of software programs
for mark-recapture analysis with closed-population models, namely, CARE-2,
