318 



Fishery Bulletin 88(2), 1990 



"D 

 ■D 



o 



n 



"D 

 0) 



o 



(a) Experiment 1 



Model 2 



5 10 15 



Period at liberty (yr) 



(b) Experiment 2 



5 10 15 



Penod at liberty (yr) 



20 





nj 



(c) Experiment 3 



10 



08 



1 On 



Period at liberty (yr) 



(d) Experiment 4 



Model 2 



5 10 



Period at liberty (yr) 



— 1 — 

 15 



20 



Figure 1 



Comparisons of the probabilities of tag shedding by southern bluefin tuna as predicted by the best fitting models, with the proportions of 

 tags shed as calculated from grouped data. Bars represent 95% confidence intervals on the proportions. 



having to condition on the observed times at Hberty, 

 the calculated standard errors of the estimated param- 

 eters are almost certainly biased downwards. As the 

 numbers of recaptures increase and the grouping in- 

 tervals shorten, estimates based on grouped and un- 

 grouped data should converge. However, the rate of 

 convergence may be slower than might be expected. 

 Table 4 gives parameter estimates that were calculated 

 by applying the Kirk wood (1981) method to the grouped 

 data in Table 2. Comparison of estimates using un- 

 grouped data (Table 3) and grouped data (Table 4) 

 reveals substantial differences, not only in terms of the 



parameter estimates themselves (e.g., experiments 

 3 and 5) but even in the model selected (e.g., ex- 

 periments 6 and 7), despite the relatively large numbers 

 of recaptures. 



No formal statistical analysis is needed to conclude 

 that the shedding rates in experiments 7 and 8 are 

 significantly lower that those in experiments 1-6. We 

 believe that the most likely explanation for these re- 

 duced shedding rates lies in differences in tags and 

 tagging methods. Some of the tags used in the early 

 experiments were reported to have inadequately at- 

 tached streamers (Hynd et al. 1967). If so, this would 



