WAHLEN AND SMITH: OBSERVER EFFECT ON INCIDENTAL DOLPHIN MORTALITY 



Table 6.— Numbers of dolphin sets (n) made from April 1981 through 1982. Sets are classified by observer type (NMFS, lATTC), by area 

 of set (North Inside, North Outside, South), by date of set relative to period (1 = Jan.-f\/lar., 2 = Apr-Dec), and by numbers of dolphins 

 killed (0, >0). Percents of column totals (%), expected frequencies (e), and the statistic x^ with degrees of freedom (df) are also tabulated. 

 Values of x^ > 3.84 (1 df) or 12.59 (6 df) are significant, as Indicated by an asterisk. 



ing the data by period (Fig. 3) and by demonstrating 

 the lack of a corresponding change in data collected 

 by lATTC observers (Tkble 4). 



Second, we demonstrated that before March 1981 

 the percent of sets with no dolphins killed was higher 

 for NMFS observers collecting data both for esti- 

 mating dolphin kill and for monitoring compliance 

 with dolphin-release regulations than for lATTC 

 observers collecting data only for estimating dolphin 

 kill (Tkble 5). Tb validate this comparison we also 

 demonstrated that the difference disappeared, or 

 perhaps was reversed, following March 1981 when 

 monitoring compliance with dolphin-release regula- 

 tions was removed from the NMFS observers' 

 responsibilities (Tkble 6). Following Johnson and 

 Bolstad (1973), these differences in the data collected 

 by observers differing in their purposes of data col- 

 lection, and hence in their obtrusiveness, imply the 

 existence of an observer effect. 



In making these comparisons, we stratified the 

 data to account for possible differences in fishing 

 conditions in different geographic areas and 

 throughout the year because both area and time of 

 year are important determinants of dolphin mor- 

 tality. Thus, the differences in the percent of zero- 

 kill sets which we identified cannot be attributed to 

 nonrepresentativeness of the data with respect to 

 area and time of year. 



We did not attempt to test for other differences 

 in the frequency distributions of kills, such as 

 changes in the percent of moderate or large kill sets. 

 Sets with large numbers of dolphins killed are rare, 

 and are generally associated with unusual circum- 

 stances, such as mechanical failures. The percent of 

 sets v^th 1-9 dolphins killed appears to vary inversely 

 with the percent of sets with zero dolphins killed 

 (Figs. 1, 4, 5). 



Powers et al.^ showed that the use of some dolphin- 

 release procedures significantly reduces dolphin 

 mortality. Thus, more time and effort expended by 

 the operator on release of dolphins could result in 

 an increase in the frequency of sets with no dolphins 

 killed, and a corresponding decrease in the frequency 

 of sets with 1-9 dolphins killed. A greater tendency 

 for vessel operators to take the additional time in 

 the presence of an observer collecting data for 

 monitoring compliance with dolphin-release regula- 

 tions could account for the differences we have 

 demonstrated. 



The significantly different relative frequency of 

 zero-kill sets recorded by NMFS observers after 

 March 1981 (Tkble 6) was not expected under our 

 hypothesis. As noted above, this difference was 

 localized to one area-period stratum, and the other 

 five strata were consistent with the null hypothesis 

 of no difference Either this difference is merely a 

 sampling anomaly, or there are differences between 

 observers in more recent years that we have not 

 taken into account. 



Gulland (1983, p. Ill) described a method of 

 testing for the existence of a tagging effect that is 

 analogous to our indirect method of testing for an 

 observer effect. He suggested comparing the pro- 

 portions of tags returned from fish tagged under 

 poor and good conditions. In both Gulland's and our 

 methods, the absolute magnitude of the effects can- 

 not be estimated. For instance, in Gulland's exam- 

 ple improvement in the conditions under which tags 

 are applied is unlikely to eliminate entirely the tag- 

 ging effect. Similarly, the reduction in observer ob- 



«Powers, J. E., N. C. H. Lo, and B. E. Wahlen. 1979. A statis- 

 tical analysis on effectiveness of porpoise rescue procedures in 

 reducing incidental mortality. Southwest Fish. Cent. La JoUa Lab., 

 Natl. Mar. Fish. Serv., NOAA, Admin. Rep. LJ-79-7, 29 p. 



529 



