month or station. Only those months and stations where a species was 

 'present' by the above criteria were used in the AOV. The stations and 

 months selected for each species are listed in Table 6. Even with these 

 restrictions, the frequency distributions of the catches were highly 

 skewed toward small catches and were non-normal, typical of biological 

 data (Poole 1964) (see Fig. 2-3). 



Highly skewed distributions frequently approximate a negative 

 binomial distribution (Poole 1974), and this possibility was investigated, 

 A goodness-of-f it (chi-square) test of the selected data to negative 

 binomial distributions predicted from the maximum likelihood estimators 

 of k, indicated that eight of the 12 selected species' catch data did 

 come from this type of distribution (Table 7). Thus, after transforming 

 the data (In (catch + k/2)), they could be used in normal theory based 

 statistical tests. 



The autocorrelation structure of the data can be determined by 

 plotting the autocorrelation function (ACF) which represents the pattern 

 of correlations between data points at various lags. The ACF plots for 

 selected species catch at JC were representative of all stations and are 

 presented in Figures 7 and 8. It was apparent that significant serial 

 dependence existed among the data points. Glass et al. (1975) have 

 concluded that the effect of autocorrelations on probability statements 

 cannot be designated. 



The last aspect of the data to be examined was the ability of the 

 current program to detect changes and the number of samples required to 

 improve the detectability. Based on the analysis of variance model 

 presented earlier, estimates of the variances associated with various 

 sources are presented in Appendix 1. The variances in each cell were 



22 



