WAHLEN AND SMITH: OBSERVER EFFECT ON INCIDENTAL DOLPHIN MORTALITY 



cent zero-kill sets (conditional on period of the year 

 and geographic area) by summing chi-square values 

 and degrees of freedom from each stratum. When 

 observations were not available in one of the strata, 

 or when the alternative was one-sided, an overall test 

 based on the chi-square statistic was not possible 

 In those cases, results of the tests within each 

 stratum were considered separately. 



RESULTS 



The first two comparisons of frequency distribu- 

 tions test for differences in the percent of zero-kill 

 sets in data collected by observers on trips begun 

 before compared with trips begun after the change 

 in NMFS observer data collection purposes in March 

 1981 (Comparisons la and 2a, Tkble 2). The last two 

 comparisons test for differences in the percent of 

 zero-kill sets in data collected by observers on trips 

 begun during the same time period (Comparisons 

 lb and 2b, Table 2). 



Before versus After 



The percent of zero-kill sets for NMFS-observed 

 trips was higher before March 1981 than after that 

 date (Fig. 1), and within all area-period strata with 

 complete data, the percent of zero-kill sets was larger 

 before March 1981 (Tkble 3). The one-sided test of 

 this difference (Comparison la, Tkble 2) was signi- 

 ficant within four of the five area-period strata which 

 had complete data, and was very nearly significant 

 within the fifth (Tkble 3). Thus, the percent of zero- 

 kill sets recorded by NMFS observers was signifi- 

 cantly larger before March 1981. 



The significant difference in percent of zero-kill 

 sets for NMFS observers before compared with after 

 March 1981 could be due to the change in data collec- 



tion purposes of NMFS observers which occurred 

 then. Alternatively, the difference could be due to 

 a temporal decline begun before that date. 



Allowing for period, the data prior to March 1981 

 do not show a pronounced trend for any of the three 

 areas (Fig. 3). Although there appears to be a decline 

 in the South for Period 2, this is unreliable as it 

 depends entirely on the 1980 and 1981 data points 

 representing a total of only 17 sets. Similarly, there 

 seems to be a declining trend for the two northern 

 areas. However, for the North Inside area the Period 

 1 points show no decline, and the possible decline 

 of Period 2 points depends on the 1980 Period 2 

 point. A 95% confidence interval about this point 

 (observed percent +2 x standard error), however, 

 is large relative to the difference between it and the 

 Period 2 point of 1979. Further, any such declining 

 trend in Period 2 points for the North Inside area 

 is not reflected in the low 1978 point. A similar argu- 

 ment can be made for North Outside area data to 

 reject the alternative explanation of the difference 

 in percent zero-kill sets before and after March 1981 

 being the result of a temporal trend begun prior to 

 March 1981. 



That the differences in percent of zero-kill sets for 

 NMFS-observed trips was not due to a temporal 

 trend was also tested by comparing the percent of 

 such sets for lATTC-observed trips before and after 

 March 1981. The percent of zero-kill sets for lATTC- 

 observed trips was higher before March 1981, but 

 within the six area-period strata the differences were 

 not consistent (Tkble 4). The two-sided test (Com- 

 parison 2a, Tkble 2) was significant within only one 

 of the six area-period strata (Period 1, South), and 

 the sample size within that stratum was very small 

 (Tkble 4). The overall conditional test given area and 

 period was not significant. Thus, I ATTC -observed 

 trips with dolphin sets from 1979 through 1982 did 



Table 3.— Numbers of dolphin sets (n) made during NMFS-observed trips, 1978 through 1982. Sets are classified by trip departure date 

 relative to March 1981 (before or after) and to period (1 - Jan.-Mar, 2 = Apr-Dec), by area of set (North Inside, North Outside, South), 

 and by numbers of dolphins killed (0, X)). Percents of column totals (%), expected frequencies (e), and the statistic Z are also tabulated. 

 Values of Z > 1.64 are significant, as indicated by an asterisk. 



'Computation of overall test statistic not possible because of one-sided alternative, and because of lack of data in one stratum (Period 1, North Outside). 



525 



