638 



Fishery Bulletin 91(4). 1993 



narrow limits seen for offshore northern spotted dol- 

 phins at high coverage. Confidence intervals for com- 

 mon dolphins illustrate the problems inherent in rela- 

 tively heterogeneous data, where mortality is 

 sometimes quite high, usually relatively low, and data 

 overall are relatively few. For this species, confidence 

 intervals were extremely variable even with relatively 

 high coverage (Fig. 12) and were very narrow (or non- 

 existent) for replicates which fortuitously included only 

 low-kill data, but very wide for replicates including a 

 few very high kills. 



Although these analytic confidence intervals are 

 clearly inappropriate measures for precise estimation 

 of variance characteristics of the estimates (note nega- 

 tive intervals in some cases; Figs. 10, 11, and 12), they 

 are presented here because they provide an effective 

 illustration of the effects of varying conditions on the 

 variability of the mortality estimates. Despite the 

 inter-group differences, the general response to increas- 

 ing coverage and increasing fleet size is similar in all 

 three dolphin group types. Confidence intervals be- 

 come narrower and more stable as more data become 

 available. 



Discussion 



Dolphin group type 



Of the three factors investigated here (dolphin group 

 type, observer coverage level, and fleet size), dolphin 

 group type had the greatest effect on dolphin mortal- 

 ity estimates, followed by percent coverage and fleet 

 size. This hierarchy of effects is controlled by two char- 

 acteristics of the kill data for each dolphin group type — 

 frequency (the number of times that mortality occurs) 

 and variability (differences between times in the num- 

 ber of dolphins killed). The total number killed can 

 have relatively little influence on the quality of the 

 estimate. This is illustrated by comparing results for 

 whitebelly spinner dolphin and common dolphin. Al- 

 though total kill was comparable for both dolphin group 

 types (981 deaths of whitebelly spinner dolphin, 882 

 deaths of common dolphin; Fig. 3) the data sets dif- 

 fered markedly both in number of trips incurring kill 

 (62 for whitebelly spinner dolphin, 12 for common dol- 

 phin) and in the distribution of kill per day among 

 those trips (Fig. 3). 



The data set for common dolphins exhibits the worst 

 of both characteristics; frequency of kill was low (few 

 trips killed common dolphins) and variability between 

 trips in kill per day was high. These problems exem- 

 plify an unfortunate interaction between data collec- 

 tion problems and the ecology of the dolphins them- 

 selves. Mortality of common dolphins due to the U.S. 



fleet during 1987 was infrequent because the geographic 

 range of this species is relatively limited and occurs 

 primarily within Mexico's Exclusive Economic Zone. 

 U.S. vessels rarely fish in this area, therefore common 

 dolphins rarely die in U.S. tuna nets in this area. Mor- 

 tality was variable at least in part because common 

 dolphins have an unfortunate habit (in this context) of 

 forming very large schools, pre-disposing them to the 

 possibility of very large-kill "disaster" sets. 



The data set for whitebelly spinner dolphin exhibits 

 a problem with only one of the characteristics; data 

 are relatively infrequent. Unlike the case for common 

 dolphins, kill per day was not extremely variable. This 

 similarity in kill per day generates statistics for 

 whitebelly spinner dolphin that are much less biased 

 and variable than for common dolphin. 



Observer coverage level 



The effect of observer coverage level is influenced both 

 by fleet size and dolphin group type. To achieve a de- 

 sired level of precision and accuracy in mortality esti- 

 mates, observer coverage levels will have to be higher 

 in smaller fleets because the available data will be 

 fewer, and higher in dolphin groups with "messy" data, 

 because observer coverage levels affect the probability 

 of encountering an unusually large kill 5 . 



Observer coverage will need to be relatively high 

 even for large fleets, when estimating mortality of dol- 

 phins with sparse and heterogeneous data. For ex- 

 ample, with kill data as sparse and variable as was 

 the case for common dolphins in 1987, nearly 100% 

 coverage would be required to generate CVs lower than 

 20% regardless of fleet size (Fig. 6). With more fre- 

 quent and less variable kill data, such as for northern 

 spotted dolphin in 1987, CVs lower than 20% can be 

 achieved with 50% coverage of 10-boat fleets (Fig. 6). 

 When fleet size drops to 5 boats, even this relatively 

 well-behaved data set requires coverage at about 75% 

 to achieve CVs less than about 20 f £ . With coverage as 

 low as 25%, even a fleet size of 20 boats was insuffi- 

 cient to meet the management objective of 20% CV. 



Fleet size 



In smaller fleets, each data point comprises a larger 

 fraction of the available mortality data. In particular, 

 the influence of unusually large mortalities (e.g., the 



'The possibility of high kills is more important than the possibility 

 of low kills; this is because all dolphin group types experience zero 

 kill frequently (therefore it is not unusual or unexpected I. Also, zero 

 kill is a definitive lower bound, while the upper bound on kill i^ 

 limited only by the potential school size of the dolphin group. 



