Stoner: Discard mortality for Alaskan crabs after exposure to freezing temperatures 
455 
monitored for nine days. This monitoring period was 
determined by the duration of the vessel charter, and 
justified by Carl and O’Clair’s (1995) observation that 
“almost all mortality occurred 1-2 days after exposure 
to freezing temperatures.” 
To ensure that holding conditions did not affect the 
condition of crabs during posttreatment monitoring, 
additional crabs collected in the trawls were set aside 
for routine handling and monitoring in the same man- 
ner as that for the test subjects. Males (n- 20 C. bairdi 
and n = 34 C. opilio ) with perfect reflex scores and in 
size ranges equivalent to the experimental crabs were 
removed from the holding tanks daily and monitored for 
mortality. None of these crabs died and all had perfect 
reflex scores at the end of the experimental period, 
indicating that the holding conditions were adequate 
despite water temperature higher than where the crabs 
were collected. 
Reflex impairment indices and statistical procedures 
Scores for reflex actions were combined into a composite 
impairment index. This index, simply the sum of reflex 
actions lost, ranges from 0 to 6. Composites provide 
robust indices of overall condition for the animal and 
have the advantage of reducing the weight of any one 
reflex (Davis, 2007). Analyses described below were con- 
ducted with two different impairment indices, one cal- 
culated for reflexes assessed immediately after removal 
from the freezer (index A) and the other after two min- 
utes in a water bath (index B)(see above). 
As with the statistical approach of Stoner et al. 
(2008), a logistic regression was used to model mortal- 
ity with potential predictors and mediators, namely 
reflex impairment, experimental temperature, and crab 
size. Models were fitted by the method of maximum 
likelihood for binary data (i.e., dead or alive) by using 
the regression module of Systat 12 (SYSTAT Software, 
Inc., San Jose, CA) (Peduzzi et al., 1980). A backward 
stepwise approach was used to determine the most 
parsimonious model for mortality, with an alpha value 
of 0.15 to remove a variable from the full model. This 
model for mortality was described by the following 
equation: 
Log e (p / (1- p)) = a + P'x, (1) 
where p = proportion ofy=l; 
y = 1 if dead and 0 if alive; 
a - intercept; 
/3'= model coefficients; and 
x = the model matrix of explanatory variables. 
The maximum likelihood estimates of mortality (p) were 
calculated as 
p = e fol+ P' x) / 1 + gla+P'x). 
Initially, the data for each species were split randomly 
into equal halves, one representing a learning set and 
Exposure (°h) 
Figure 1 
Numbers of limbs autotomized by Chionoecetes bairdi 
(Tanner crab) and C. opilio (snow crab) during and imme- 
diately after freezing (circles) and at the end of a 9-day 
holding period (triangles). Exposure treatments were 
made at two temperatures and are plotted in units of 
degree-hours (°h, the product of temperature [in degrees 
Celsius] and time [in hours]). 
the other a test set. The most parsimonious logistic 
model was developed by using the learning set that was 
then validated with the test set. After cross-validation, 
a final model was fitted to the entire data set. Finally, 
the logistic model for each species was used to develop 
a response curve showing the probability of mortality 
based upon fixed values for the key observations of crab 
condition. 
Results 
Autotomy 
Autotomy during and immediately after freezing condi- 
tions increased in direct proportion to cold exposure 
and was generally higher in C. opilio than C. bairdi 
(Fig. 1). Chionoecetes opilio lost two legs on average in 
exposures ranging from -4 to -10°h, and none at -2°h. 
Average losses in C. bairdi were less than one across the 
