564 



Fishery Bulletin 100(3) 



the direction of sampling was reversed in some years, pro- 

 viding some contrast. 



Species richness, as measured by the number of spe- 

 cies per haul, was initially modeled as a Poisson variable 

 (counts). However, an analysis of the distribution of the 

 observed numbers and of residuals from the fitted models 

 suggested that it could more appropriately be modeled 

 as a normally distributed variable (see below). We fitted 

 models assuming either a Poisson or a normal distribu- 

 tion and found that the estimated patterns were virtually 

 identical and in both cases the same model was selected as 



best model. Only results based on the normal distribution 

 are presented. Species richness was highly variable and 

 the best model explained a small portion of the overall 

 variability (Pseudo-r2=0.25). The average number of spe- 

 cies increased with area swept (Fig. 2), which may be ex- 

 pected if species area relationships established for terres- 

 trial ecosystems also hold in the marine environment. The 

 number of species tended to peak at intermediate depths 

 (200-300 m) and was highest in the eastern GOA (Table 4), 

 decreasing steadily west of Prince William Sound (km 

 1100, approximately 147''W). Species richness appeared 



