342 
Fishery Bulletin 106(4) 
0 1 2 3 4 5 6 7 
Reflex impairment score 
Figure 1 
Percent mortality of Chionoecetes bairdi (Tanner crab) 
and C. opilio (snow crab) shown as a function of reflex 
impairment indices A and B. Index A was calculated as 
the sum of completely lost reflexes (score = 0) and index 
B was calculated as the sum of reflexes either lost or 
weak (scores = 0 or 1). 
ity was observed for C. opilio. Crabs with high reflex 
impairment (4 to 6) usually died within 1 to 2 days 
and there was little variation in this period of mortal- 
ity. From these results, we concluded that the holding 
period for the experiment (11 days) was sufficiently long 
to test the relationship between reflex impairment and 
mortality. 
Re-assessment of reflex impairment in surviving crabs 
at the end of the experiment revealed that condition of 
the vast majority did not deteriorate during the hold- 
ing period at sea. Only two of the surviving C. bairdi 
(n = 176) had increased reflex impairment. One of these 
had multiple new leg autotomies and the other had 
multiple weak responses upon capture. Improvement 
in the reflex impairment index by one or two points 
occurred in 7.4% of the crabs, whereas the balance 
showed no changes. Only 8.5% of the survivors had 
reflex impairments >0 by the end of the study. Among 
the 295 C. opilio survivors, none had higher or lower 
reflex impairment, although most of the survivors (98%) 
0 2 4 6 8 10 
Days after capture 
Figure 2 
Percent cumulative mortality of Chionoecetes bairdi 
(Tanner crab) and C. opilio (snow crab) over time during 
holding at sea. Total numbers of C. bairdi and C. opilio 
dying over the course of the experiment were 73 and 
96, respectively. 
had no impairment initially. These results indicate that 
the deck tanks provided a suitable means of holding to 
assess delayed mortality in the subject species. 
Mortality and its relationship to reflex impairment 
Logistic regression analysis revealed that mortality in 
C. bairdi was associated primarily with reflex impair- 
ment index and injury score (Table 4). Despite a wide 
range of crab size and shell condition (Table 2), neither 
these variables nor crab gender was a significant vari- 
able in the regression model. In the most parsimonious 
model, containing a constant, reflex impairment and 
injury score correctly predicted 82.7% of the mortality 
and survival (Table 4), whereas the full model (with all 
variables) predicted 82.8% correctly. Cross-validation 
showed that the model was robust; i.e., the random 
learning set produced a model that correctly predicted 
86.7% of the test set for C. bairdi. When the injury 
score was removed from the regression model, 79.5% of 
the observed outcomes were predicted correctly, dem- 
onstrating the value of the assessed reflexes. A similar 
analysis for C. opilio revealed reflex impairment to be 
the only significant predictor of mortality (Table 5). The 
most parsimonious model, containing a constant and the 
reflex impairment index, correctly predicted 91.0% of the 
mortality and survival — a percentage almost identical to 
that of the full model (82.8%). As with C. bairdi, cross- 
validation for C. opilio with a learning set resulted in 
94.0% correct predictions for the test set. Consequently, 
models for both species were robust for the field study. 
Surface plots for probability of mortality with the 
logistic models for each species (Fig. 4) showed the 
strong relationships between reflex impairment and 
mortality. For C. bairdi, mortality increased rapidly 
