Caillouet et al.: Relationship between sea turtle stranding rates and shrimp fishing 



intensities 



24 



Because some observations of S and E were equal 

 to and the data required logarithmic transforma- 

 tion to assure bivariate normality for correlation 

 analyses, 1 was added to all observations of S and E. 

 Tables 3 and 4 show that ln(S + 1 ) and ln<£ + 1 ) had 

 lower coefficients of skewness and kurtosis (both 

 approaching zero) than did S and E, indicating that 

 the log-transformed variables approached normal- 

 ity, whereas the nontransformed variables were not 

 normally distributed. 



Forty product-moment correlations (2 periods x 2 

 geographic zones x 10 depth intervals) between the 

 paired variables ln(S + 1) and \n(E + 1) were calcu- 

 lated, each based on 48 observations ( 12 months x 4 yr) 

 within a depth interval. These correlations were not 

 statistically independent of each other, because each 

 was based on one set of 48 observations of ln(S + 1) 

 correlated with each often sets of 48 observations of 



\n(E +1). Significant, positive correlations detected 

 between ln(S + 1) and \n(E + 1) were tested for ho- 

 mogeneity by using chi-square. Finally, 140 product- 

 moment correlations ( 2 periods x 2 geographic zones 

 x 35 correlations) between pairs of ln(£ + 1) for the 

 ten depth intervals were calculated, each based on 

 48 observations. These correlations were statistically 

 independent of each other. 



Results 



Ten significant positive correlations (i.e. correlation 

 coefficients, r, greater than at P <0.05) were de- 

 tected between the ln(S + 1) and lnlE +1), half of 

 them in 1986-89 (Fig. 2) and the other half in 1990- 

 93 (Fig. 3). These ten correlations involved ln(£ + 1) 

 for some of the depth intervals between and 20 fm 



