Meckel et a\: Evasive behavior of Stenella attenuata and S longirostns 



697 



cautious approach would be to apply the nonparametric 

 Kolmogorov-Smirnov two-sample test because differences 

 in location (average) and spread are tested simultane- 

 ously (Conover, 1980; Neave and Worthington, 1988). The 

 Kolmogorov-Smirnov test compares the cumulative distri- 

 bution function of two samples and uses the maximum 

 vertical difference between them as the test statistic D 

 (Neave and Worthington, 1988). 



Eastern spinner and northeastern offshore spotted dol- 

 phins were also compared with respect to the dolphins" 

 evasive strategies, such as evasion under and evasion over 

 the net. The Kolmogorov-Smirnov test for two indepen- 

 dent samples was used to evaluate estimated evasion indi- 

 ces by set between the stocks when they evaded capture 

 by swimming under the net. The one-sided test was used 

 to confirm if the evasion indices by set in one stock were 

 larger than in the other. Differences in estimated evasion 

 indices by set. when dophins evaded capture by swimming 

 over the net, were not tested owing to very low sample 

 sizes (n=2> in the northeastern offshore spotted, 7;=5 in the 

 eastern spinner dolphin). 



Furthermore, differences between the two stocks were 

 described with respect to the dolphins" dispersion, i.e. the 

 ability of the herds to "explode" (separate suddenly) (Allen 

 et al.. 1980) into subgroups in relation to the herds' config- 

 uration before the chase. For this description, the group- 

 ing codes applied by the observers in paragraph 3 of the 

 RMMOSD were used. The codes used were the following; 

 1, herd is in one group; 2. herd has divided into two or 

 three groups; 3, herd consists of more than three groups. 

 These codes were recorded during three set stages: before 

 chase, during chase, and during encirclement. For each 

 stock, the number of sets where the specified grouping 



code occurred was counted during each set stage. If codes 

 were found in ascending order, this was interpreted as 

 herd dispersion during the fishing operation. 



To test for significant differences in dispersion behavior 

 between northeastern offshore spotted and eastern spin- 

 ner dolphins, only data for evasion area 3 (Fig. 1) were 

 used because sample sizes for both stocks were largest 

 there. A multiway frequency table seemed to be the appro- 

 priate statistical tool, since counts of grouping codes was 

 the response variable during each set stage and for each 

 stock. However, one of the most important assumptions of 

 multiway frequency analysis, independence, was not met. 

 Only designs for comparisons between subjects may be 

 analyzed with this analysis, so that the frequency in each 

 cell is independent of the frequencies in all other cells. If 

 the same case contributes values to more than one cell, 

 those cells are not independent (Tabachnik and Fidell, 

 1996). In our study, each case (i.e. each set) contributed to 

 three different cells (the three set stages). 



Consequently, to test for differences in dispersion behav- 

 ior between the stocks, data were rearranged by using the 

 grouping code as the response variable and set stages as 

 the repeated measures in each set (Table 1). Therefore, 

 this design resembled a repeated-measures analysis with 

 one among-subjects factor (stocks) and one within-sub- 

 jects factor (set stage). If the response variable were in 

 interval scale, a repeated-measures analysis of variance 

 (AN OVA) would have been appropriate to test for differ- 

 ences between stocks (Zar, 1999). However, the response 

 variable in our study was the grouping code, a categorical 

 variable in ordinal scale. 



The plausible alternative to apply was logistic regres- 

 sion, often referred to as linear probability models (Tabach- 



