458 
Fishery Bulletin 107(4) 
Table 4 
Percentages of reflex actions lost in Chionoecetes bairdi 
(Tanner crab) and C. opilio (snow crab). When only one 
reflex was absent in a crab it was considered the 1 st reflex 
lost. In all cases where only one reflex remained that reflex 
was mouth closure. The right column represents the per- 
centage of specific reflexes that were lost among all of the 
crabs where between one and five reflexes were lost. 
Reflex 
1 st reflex lost 
% of total losses 
C. bairdi 
(ra= 20) 
(71=227) 
Chela closure 
40.0 
25.6 
Kick 
35.0 
24.7 
Eye retraction 
20.0 
18.9 
Leg flare 
5.0 
14.1 
Leg retraction 
0 
16.3 
Mouth closure 
0 
0.4 
C. opilio 
(71 = 14) 
(71=201) 
Chela closure 
42.9 
27.9 
Kick 
14.3 
20.9 
Eye retraction 
21.4 
16.4 
Leg flare 
14.3 
16.9 
Leg retraction 
7.1 
16.9 
Mouth closure 
0 
1.0 
Table 5 
Results of logistic modeling for mortality in Chionoecetes 
bairdi (Tanner crab). A backward stepwise approach was 
used to determine the most parsimonious model for mor- 
tality, with an alpha value of 0.15 to remove a variable 
from the full model. 
Parameter 
Estimate 
Z 
P value 
Full model 
constant 
-7.466 
-1.789 
0.074 
temperature 
0.008 
0.013 
0.990 
crab size 
0.058 
1.321 
0.187 
reflex impairment 
0.877 
5.900 
<0.001 
Most parsimonious model 
constant 
-1.953 
-4.427 
<0.001 
reflex impairment 
0.833 
5.568 
<0.001 
Prediction matrix for the most parsimonious model 
Dead 
Live 
Actual 
predicted predicted 
total 
Die 
42 
11 
53 
Live 
9 
38 
47 
Total no. predicted 
51 
49 
100 
Correct (%) 
79.2 
80.9 
False (%) 
26.2 
33.9 
Total correct (%) 
80.0 
predicted 81.0% correctly. The model for a random 
learning set (one half of the data) incorporated the 
same statistically significant variables (constant and 
reflex impairment) and correctly predicted 75.5% of the 
results for an independent test set. A similar analysis 
for C. opilio resulted in a full model that predicted 
80.4% of mortality and survival correctly, with only 
reflex impairment having a significant effect (Table 
6). Restricted to a constant and reflex impairment, the 
model predicted 79.4% of the results correctly. Cross- 
validation for C. opilio with a random learning set re- 
sulting in 80.4% correct predictions for the test set. 
Consequently, models for both species are robust for the 
experiments. Curves of the calculated probabilities of 
mortality based upon the logistic models with specific 
reflex impairments revealed very similar patterns for 
the two crab species (Fig. 6). Both plots rise in smooth 
sigmoid form from 13% to 17% mortality in crabs with 
impairments equal to zero to near total mortality with 
impairments equal to six. 
