304 



Fishery Bulletin 101(2) 



indices. Thus these indices may mask the synchronous fluc- 

 tuations in the other species, so that omitting them would 

 produce more extreme years. In some cases it did make the 

 most extreme rank deviations more extreme. However, it 

 also had the effect of increasing the threshold (because the 

 number of species decreased). The net effect was to produce 

 slightly fewer extreme years. There was one additional ex- 

 treme year — 1993 for series 7. However, the following years 

 were no longer deemed extreme: 1980 for series 1; 1994 and 

 1996 for series 5; and 1988 for series 15. 



Our second alternative was based on the idea that envi- 

 ronmental changes may affect different species differently. 

 That is, an environmental extreme produces extreme catch- 

 ability, but this may be high for some species and low for 

 others. To test this we calculated rank deviations for each 

 species and then averaged the rank deviations (rather than 

 averaging the ranks and then calculating deviations). This 

 method identified only four extreme years — three were as 

 in Figure 5 ( 1984 and 1989 for series 9 and 1995 for series 

 10) and one was new (1979 for series 1). 



The third alternative was a variant on the second. We as- 

 sumed that the species for each series fall into two groups: 

 one group whose catchabilities are all affected in the same 

 way by environmental changes, and a second group for which 

 the effect is opposite. That is, when catchability is high for 

 the first group it will be low for the second, and vice versa. We 

 calculated the mean ranks as above and then determined, 

 for each species, the Euclidean distance between these mean 

 ranks and 1) the species ranks, and 2) the "inverse" of the 

 species ranks (if a species ranks are, say 1, 4, 2, 3, then the 

 inverse ranks are 4, 1, 3, 2). When the latter distance was 

 smaller, the species was said to fall into the second group. 

 The ranks for all group-two species were replaced by their in- 

 verse ranks and the mean ranks (and thus rank deviations) 

 were recalculated. With this method only two extreme years 

 were found, both of which are extreme in Figure 5 (1984 

 and 1989 for series 9). Often there was no clear separation 

 between gi'oups one and two. Sometimes (but only when 

 there were few species ) group two was empty. We also tried 

 a cluster analysis approach to the identification of groups one 

 and two but this produced no better results. 



