Hampton and Kirkwood: Tag shedding by Thunnus maccoyii 



317 



Results 



Discussion 



Maximum likelihood estimates of the tag-shedding 

 parameters and of their standard errors are given in 

 Table 3. Curves, based on these parameter estimates, 

 that describe the time-dependent probability of tag 

 shedding, together with estimates of the proportion of 

 tags lost (with 95% confidence intervals) based on 

 grouped data, are shown in Figure 1. 



In experiment 1, it proved impossible to distinguish 

 either statistically or visually between a constant-rate 

 model with immediate tag shedding (model 2) and a 

 decreasing-rate model with no immediate tag shedding 

 (model 3). Although the plots of [1 - Q(0] against t (Fig. 

 la) are strikingly different for longer recapture times, 

 apparently both can be accommodated by the grouped 

 recovery data because the number of long-term recov- 

 eries is small, resulting in wide confidence intervals for 

 proportions shed. 



In all other experiments, it was possible to select a 

 single most appropriate model on statistical grounds. 

 However, wide confidence intervals for the estimates 

 of proportions of tags shed were still evident in most 

 experiments for the longer-term recovery periods. In 

 experiment 2 and the medium-term experiments (4 and 

 6), a constant-rate model with some immediate tag 

 shedding was selected (Fig. lb,d,f). In the other long- 

 term experiments (3 and 5), model (3) provided the best 

 fit, and the shedding rate showed a marked tendency 

 to decrease with time (Fig. lc,e). 



The most striking difference among the experiments 

 was the very low shedding rates observed in experi- 

 ments 7 and 8 (Fig. lg,h) compared with the other 

 experiments. After 4 years, the probability of shedding, 

 as predicted by the fitted models, was approximately 

 0.2 for experiments 7 and 8, whereas it was 0.5-0.7 

 for the earlier experiments. 



The tag-shedding rates estimated for experiments 1-6 

 are of a similar order to those obtained for southern 

 bluefin tuna by Kirkwood (1981), who used a restricted 

 subset of the data pooled across these experiments. 

 Constant rates of tag shedding have been estimated 

 by Hynd (1969) for southern bluefin tuna (0.26/yr), by 

 Bayliff and Mobrand (1972) for yellowfin tuna (0.278/ 

 yr), and by Lenarz et al. (1973) and Baglin et al. (1980) 

 for Atlantic bluefin tuna (0.210/yr and 0.205/yr, respec- 

 tively). The essentially constant rates estimated in this 

 paper for experiments 2, 4, and 6 are generally slight- 

 ly lower then these, but still comparable. However, the 

 shedding rates estimated for experiments 7 and 8 are 

 very much lower than for any previously reported tuna 

 double-tagging experiment with substantial numbers 

 of recoveries. The low estimates of shedding rates ob- 

 tained by Laurs et al. (1976) for north Pacific albacore 

 and Lewis (1981) for south Pacific skipjack were based 

 on very small numbers of tag recoveries. 



As mentioned earlier, most previous estimates of tag- 

 shedding rates have been calculated using recovery 

 data grouped by period at liberty. For this method to 

 be effective, the number of observations in each group- 

 ing interval must be above a certain minimum. Where 

 the data are sparse, the grouping intervals must cover 

 longer periods. This can result in loss of information, 

 as is seen, for instance, in Table 1 for the longer periods 

 at liberty. Worse, as Kirkwood and Walker (1984) 

 found for a double-tagging data set with very few 

 recoveries, the grouped data estimates may be highly 

 dependent on the grouping intervals used. Here, build- 

 ing on Kirkwood and Walker (1984), we have adopted 

 an estimation procedure that uses exact periods at 

 liberty, which— at least in principle— gets around this 

 problem. However, this is not achieved without a price: 



