WAHLEN AND SMITH: OBSERVER EFFECT ON INCIDENTAL DOLPHIN MORTALITY 



pliance with dolphin-release regulations than an 

 observer who is collecting data only for estimating 

 dolphin kill. That this is the case is implied by the 

 constraint in the sampling procedure that any vessel 

 sampled twice within a calendar year may be accom- 

 panied by an NMFS observer on at most one trip. 



METHODS 



We tested for the existence of an observer effect 

 on dolphin kills by comparing the number of killed 

 dolphins recorded by more obtrusive observers with 

 the number recorded by less obtrusive observers. We 

 considered observers who collected data both for 

 estimating dolphin kill and for monitoring com- 

 pliance with dolphin-release regulations to be more 

 obtrusive to vessel operators than observers who col- 

 lected data only for estimating dolphin kill. Thus, 

 we compared kills recorded by (la) NMFS observers 

 before and after March 1981, and (lb) NMFS and 

 lATTC observers before March 1981. As a control, 

 we compared the number of killed dolphins record- 



ed by observers of equal obtrusiveness. That is, we 

 compared kills recorded by (2a) lATTC observers 

 before and after March 1981, and (2b) NMFS and 

 lATTC observers after March 1981. 



The frequency distributions of numbers of dolphins 

 killed were extremely skewed, with very long right 

 tails (Fig . 1). Normality assumptions were violated 

 so strongly by these skewed distributions that 

 ANOVA tests for differences in means, particularly 

 one-sided tests, would be difficult to interpret (Glass 

 et al. 1972). Therefore, we tested for differences in 

 the percent of dolphin sets in which no dolphins were 

 killed (zero-kill sets). This percent relates directly to 

 the regulation requiring release of all live dolphins, 

 and is a dominant feature of the dolphin kill 

 distributions. 



When comparing frequency distributions, we 

 entertained the null hypothesis of equality of per- 

 cent zero-kill sets. When comparing observers of dif- 

 ferent obtrusiveness levels, we tested this hypothesis 

 against a one-sided alternative that distributions 

 from more obtrusive observers had a higher percent 



Before 



After 



>■ 

 o 

 z 

 111 



3 

 O 

 Ui 



oc 

 u. 



lU 



> 



< 



lit 



flC 



^ 



^  — ^ — p^^ . — p 



i—V/Z/i 



1-9 10-19 20-29 30-39 40-49 50-99 ^100 



NUMBER OF DOLPHINS KILLED 



Figure L— Relative frequency distributions of number of dolphins killed incidentally during sets 

 of NMFS-observed trips, 1978 through 1982, by trip departure data 



523 



