multivariate techniques, some interpretations may be explored even when the 

 assumptions are not met. Generalized distances are determined in discriminant 

 analysis and the maximum distances for the microcosms and the bay for the 

 first axis in Figure 24-8 and Table 24-6 are shovm in Table 24-7. These distances 

 are much smaller among the replicated microcosms than among bay stations. It 

 seems feasible to consider that standard generalized distances exist for natural 

 systems, for specific variable sets, which might be compared to generalized 

 distances which result during experiments on perturbation and subsequent 

 recovery in microcosms. 



Table 24-7. Maximum Generalized Distances and 



Normalized Distances Among the Microcosms and Among 



Bay Stations from August to December 1976 and 1972 



Respectively, from the Discriminant Analysis Shown in 



Figure 24-7 and Table 24-6. 



D^ max 



Normalized 

 D^ max 



Microcosms: 

 2vs3 

 2vs5 

 2vs9 

 3vs9 



Bay Stations: 



Prov. River vs mouth E. Passage 

 Ohio Ledge vs mouth E. Passage 

 Ohio Ledge vs mouth W. Passage 



CONCLUSIONS 



The nine MERL microcosms operated during the 4-month replicabiUty 

 study were generally as similar to each other as they were to adjacent areas of 

 Narragansett Bay using nutrient and phytoplankton data sets for comparison. 



Zooplankton abundance m the MERL microcosms was somewhat low but 

 this was probably caused by an artifact that can be removed. 



Multivariate statistical techniques seem essential to the comparison of the 

 large and heterogeneous data sets generated in such studies as this. 



381 



