Wood and Cadrin: Mortality and movement of Limanda ferruginea tagged off New England 
281 
balance between the parsimony of the model and the 
goodness-of-fit, where parsimony decreases as the 
number of parameters in the model increases. Model 
fit was judged with the model likelihood (L): 
AIC = 2K - 2 ln(L), 
where K - the number of parameters. 
To determine whether the general model (fully pa- 
rameterized) was a reasonable fit to the tag-recovery 
data, goodness-of-fit was tested. A simulation proce- 
dure was used to calculate an estimate of overdisper- 
sion (c). Data were simulated at varying levels of over- 
dispersion (simulated c), and the deviance of each data 
set was divided by the degrees of freedom to obtain a 
range of values. A logistic regression was used to esti- 
mate the level of c where 50% of the simulated values 
were above and 50% were below the observed deviance 
divided by degrees of freedom of the general model. A 
model solution that perfectly conforms to the assumed 
error distribution would produce an expected vari- 
ance equal to the observed variance and would have 
a c value of 1.0. Deviations of c above or below 1.0 
indicate over- or underdispersion, respectively. Gener- 
ally, a c value >3.0 indicates poor model fit because the 
model deviance is greater than the expected deviance 
(Lebrenton et al., 1992; Burnham and Andersen, 2002). 
To account for c and for differences in effective sam- 
ple size (N), a quasi-likelihood adjusted AIC (QAIC c ) 
was used to adjust fit of the top selected models (An- 
derson et al., 1994; Burnham and Anderson, 2002): 
QAIC c =2K + 
c N-K-l 
The adjusted results from the top ranked models de- 
termined through the model selection criterion were 
then examined for biologically realistic parameter es- 
timates. Models that estimated multiple parameters at 
their boundaries (e.g., S=1.0) were rejected in favor of 
the next ranked model. 
Results 
Researchers worked with commercial fishermen to tag 
44,882 Yellowtail Flounder with conventional disc tags, 
and 3767 of these tags were recovered from the com- 
mercial fishery. Of all the lottery tags and $100 high- 
reward tags, 8% and 14% were returned, respectively. 
The relative return rate of lottery tags to high-value 
tags indicates a 59% reporting rate, assuming that 
100% of the high-reward tags were reported (Table 
1). The results from the analysis of observed recovery 
rates by sex, size, condition code, and damage code in- 
dicate that females had a greater recapture rate than 
males (particularly small males). Fish categorized as 
good had the same recovery rates as fish that were 
excellent. All damage codes had similar recovery rates, 
except for the slightly lower recovery rates for those 
Table 1 
Total releases and recaptures of tagged Yellowtail 
Flounder ( Limanda ferruginea) by tag type from 2003 
to 2006 in a cooperative tagging study off New England. 
The ratio of recovery rate from lottery tags and high- 
value tags indicates a 59% reporting rate, assuming 
100% reporting of high-value tags. 
Percentage 
Tag type 
Releases 
Recovered 
recovered 
Lottery tags 
44,501 
3713 
8.3 
$100 tags 
381 
54 
14.2 
Total 
44,882 
3767 
8.4 
fish with net marks (5% recovered) and those showing 
evidence of lymphocystis (3% recovered). 
Releases occurred in monthly batches over a 
39-month period from June 2003 to August 2006, most- 
ly in summer (Fig. 1). The full recovery matrix (with 
known month of capture) included a total recapture 
rate of 7.9% (Table 2). There was a higher rate of re- 
capture for females (8.4%) than for males (6.5%). 
Mortality 
The release and recovery data were audited, and only 
tags with both location and fish sex information were 
included in the modeling. The final recovery matrix 
used to estimate survival included 43,907 releases and 
3457 recaptures. Several model variations with both 
time-dependent and time-independent parameter es- 
timates and with sex-dependant parameter estimates 
were successfully fitted to the data. 
All of the top-ranked models determined through 
model selection had a time-dependent recovery rate 
parameter iff) with varying levels of sex dependence 
on both recovery rate and survival (S g ). Results from 
procedures for the simulation of goodness of fit indi- 
cate that the general model fitted the tag-recovery data 
well, returning a c estimate of 2.12 (Fig. 2). After the 
goodness-of-fit adjustment to the models, full weight 
was given to the model with a time-dependent survival 
and time-dependent sex-specific recovery rate. The 2 
best models had a time-dependent survival estimate. 
However, many of the estimates were at the upper 
boundary of survival (S=1.0) because of sparseness in 
the data, and therefore the 2 models with time-depen- 
dent survival were not considered. The optimum mod- 
el was a model with constant survival and time- and 
sex-dependent recovery rate (Table 3). On the basis of 
goodness-of-fit and model validation diagnostics, the 
optimum model appears to be a reliable representation 
of the data. 
A constant rate of survival of 0.89/month with a 
standard error of 0.016 was estimated from the best 
model. The annual rate of survival was calculated at 
