FISHERY BULLETIN: VOL. 83, NO. 4 



of zero-kill sets than distributions from less obtrusive 

 observers. When comparing distributions from 

 observers of equal obtrusiveness, we tested the null 

 hypothesis against the two-sided alternate of in- 

 equality (Tkble 2). Results from all of our tests are 

 reported at the 0.05 significance level. 



Table 2. — Alternate hypotheses to the null hypothesis of 

 equality of percent zero-kill sets for each of tour comparisons, 

 where Before and After refer to before or after March 1981. 

 See text for details. 



Comparison 



Alternate hypothesis 



1a. NMFS before vs NMFS after Before > After 



2a. lATTC before vs lATTC after Before # After 



lb. NMFS before vs lATTC before NMFS > lATTC 



2b. NMFS after vs lATTC after NMFS # lATTC 



For two-sided tests of differences in percents, we 

 used the standard chi-square (x^) statistic with 

 one degree of freedom (df). For one-sided tests, 

 we used the square root of the chi-square statistic 

 (Z), which is approximately normal (Snedecor and 

 Cochran 1980, p. 126-127). In some instances, the 

 expected cell frequencies were less than the tradi- 

 tionally accepted minimum of five. However, 

 recent Monte Carlo results (Fienberg 1980, p. 172) 



suggest that the chi-squared distribution is an ade- 

 quate approximation at the 0.05 significance level 

 even when minimum expected values are as low as 

 one. 



While the sampling of vessels was nearly random, 

 the actual sample obtained may not have been 

 representative of factors affecting dolphin kills. It 

 has been demonstrated that within the ETP, dolphin 

 kills vary among three geographic areas^ (Fig. 2) and 

 by periods within the year (Lo et al. 1982). We 

 divided the year into two periods: January-March 

 and April-December. This division corresponds to 

 the date of the change in data collection purposes 

 of NMFS observers in 1981, and also tends to 

 equalize sample size since vessels in this fishery are 

 more active in the early part of the year. 



We stratified the data by area and period of the 

 year to account for biases due to possible non- 

 representativeness of the sample with respect to 

 these two factors. When data on numbers of dolphins 

 killed were available in all six area-period strata, we 

 made overall two-sided tests for differences in per- 



^K-T. T^ai, Inter-American IVopical Thna Commission, c/o Scripps 

 Institute of Oceanography, La Jolla, CA 92093, pers. commun. 

 December 1983. 



40°N 



20" - 



20° - 



ACS 



lecw 



140° 



120" 



100° 



80° 



Figure 2.— The three areas of the eastern tropical Pacific used to stratify the data, bounded 

 by lat. 40°N., long. 160°W., lat. 40°S., and the western coastline of the North and South 

 American Continents. 



524 



