Meckel et aL: Evasive behavior of Stenella atlcnuata and S. longirostns 



695 



the northeastern offshore spotted and the eastern spin- 

 ner dolphin (Sti'iu'lla longifostris orientalis). A subset of 

 the PNAAPD database ( 1992-95) was extracted from sec- 

 tions of the DAR (those referring to date and set position) 

 and the RMMOSD (obsei-ver's best estimation of herd 

 size and composition by species and stock, number and 

 identification [also by species and stock] of evading and 

 escaping dolphins, as described previously), and grouping 

 codes. Data from sets that were interrupted (due to loss 

 of tuna catch because all dolphins escaped from the chase 

 or mechanical problems of the vessel or net performance, 

 etc. ) were excluded from the analyses. 



Because data collected by the Mexican observer pro- 

 gram (PNAAPD) represented 50% of the Mexican fleet's 

 effort in the EPO, we assumed that the data set in our 

 study was sufficiently large to represent the dolphins' eva- 

 sive behavior in relation to Mexican tuna fishing boats 

 during the sampled period. 



Data analysis 



Evasion index by set Evasion index by set was calculated 

 in order to search for spatial patterns in evasive behav- 

 ior of the northeastern offshore spotted dolphin and to 

 analyze differences between northeastern offshore spot- 

 ted and eastern spinner dolphins. This index is defined as 

 the estimated percentage of dolphins that evaded capture 

 during each set in relation to the herd's estimated initial 

 size: 



100, 



where /, = estimated evasion index during set / in quad- 

 rant /; 



E = number of dolphins that evaded capture dur- 

 ing set ; in quadrant 7, i.e. the sum of escap- 

 ing dolphins estimated by the observer before 

 the chase, during the chase, and during the 

 encirclement as recorded in the RMMOSD; 



H = herd size before chase started during set / in 

 quadrant J. i.e. the observer's best estimate as 

 recorded in the RMMOSD. 



Evasion index by set was stratified by stock by select- 

 ing from the database only sets on pure herds (i.e. those 

 herds composed lOO'/r of a species) of northeastern off- 

 shore spotted dolphins and sets on pure herds of eastern 

 spinner dolphins. Because only these two stocks were 

 studied, results should not be considered representative 

 of the corresponding species (pantropical spotted dol- 

 phin, Stenella attenuata. and spinner dolphin, Stenella 

 longirostris). 



Sets in which no dolphins escaped and therefore the cal- 

 culated evasion index was zero were included in all analy- 

 ses because they indicated that evasive behavior did not 

 occur or failed. This action is contrary to common practice 

 where "zeros" are often eliminated because they represent 

 missing data that tend to bias calculations. 



Spatial patterns in evasive behavior of northeastern off- 

 shore spotted dolphins To evaluate if there were spatial 

 patterns in evasive behavior for the data in our study a 

 computer program based on Matlab version 4.2c was used 

 to plot the evasion index by set of the northeastern offshore 

 spotted dolphin on a map of the EPO. The progi'am calcu- 

 lated average evasion index by set in 2 x 2 quadrants to 

 smooth the data which were then used to draw a contour 

 map with the commercial surface mapping program Surfer 

 version 6.01 (Smith et al., 1995). This software interpolated 

 the average evasion index by set in 2 x 2 quadrants to 

 form a regular rectangular array of grid values. This pro- 

 cedure was chosen because the smoothness of contours on 

 a contour map is partially a function of the number of X 

 and Y lines in the grid. When a grid is created, reducing 

 the number of lines in the X and Y directions can result 

 in more angular contours on the contour map. Most of the 

 gridding methods in Surfer use a weighted average inter- 

 polation algorithm. The gridding method called "Kriging" 

 with a linear variogram was chosen because it incorporates 

 anisotropy and underlying trends in an efficient and natu- 

 ral manner and has been proven to be quite effective for 

 many data sets in different fields (Smith et al., 1995). 



A geographic difference in evasive behavior was observed 

 in the contour map (see "Results" section. Fig. 1), and the 

 60%, 50%, and 40% contours were considered the limits 

 between three areas with different evasive behavior (as 

 defined by the evasion index) during the study period and 

 for the Mexican fleet. In addition, the following statistical 

 procedures were applied to test for significant differences 

 in mean evasion indices of the three areas. 



Analysis of variance (ANOVA) can be used to compare 

 the mean of three groups of proportions (i.e. for each area) 

 (Zar, 1999). Because proportions (like the evasion index) 

 have a binomial distribution, the individual data should 

 be transformed as follows in order to meet normality and 

 homoscedasticity assumptions (Zar. 1999): 



I' =0.5 



arcsm 



E 



H + 1 



E+1 

 H + l 



where /,' = estimated transformed evasion index during 



set ; in quadrant j; 

 E = estimated number of dolphins that evaded 



capture during set / in quadrant j; and 

 H = estimated herd size before chase during set i 



in quadrantj. 



After transformation, the data still did not have a nor- 

 mal distribution ( Kolmogorov-Smirnov goodness-of-fit test, 

 D=0.0704, P<0.01, /7=808; Zar, 1999). Therefore, distribu- 

 tion-free tests were used to search for significant differ- 

 ences between the three evasion areas (Conover, 1980; 

 Neave and Worthington, 1988). 



To search for significant differences between the medi- 

 ans (usual gi'oup average measure in nonparametric sta- 

 tistics) of three gi-oups (evasion areas), the nonparametric 

 Kruskal-Wallis multisample test seemed to be the most 

 appropriate for the data in our study. The reasons for this 



