22 



HQI model. Further, when the HQI cover variable was separated into a solid-overhead- 

 cover component and a pool-and-turbulence-cover component and these new components 

 entered as independent variables along with the rest, much stronger correlation 

 yet was obtained . 



In a set of stepwise multiple regressions involving (1) all 30 study stations, 

 (2) control stations only, and (3) urban stations only, mean channel width was 

 the first variable entered, based on its high initial influence relative to that 

 of other independent variables. However, as other variables were entered, mean 

 width rapidly lost significance, its F-value falling below the predetermined 

 rejection level of 1.0. It is inferred that channel width was meaningless in 

 describing the effect of habitat on trout abundance. Another stepwise multiple 

 regression was run with the channel-width variable omitted, and the correlation 

 improved (Table 4). 



Number of over-20-cm trout/km was consistently the dependent variable having 

 tighest correlation (Table 4). We infer that the habitat variables are related 

 to trout abundance in a linear, rather than areal fashion. Trout abundance per 

 unit stream surface area is likely to be poorly correlated with habitat quality 

 because trout orient strongly to instream cover (Hunt 1971, Wesche 1976, 

 Devore and White 1978, Enk 1977) which tends to be concentrated along channel 

 margins, hence is a rather linear variable. 



With the channel-width variable omitted from multiple regression, the 

 dependent trout abundance variable most highly correlated with habitat variables 



was still the number of over-20-cm trout/km (Table 4). For the model involving 



2 

 all 30 stations, the adjusted r values* indicated that variation in habitat 



2 2 2 



♦Adjusted r = r - [ (K-1) / (N-k) ][l-r ], where K = number of independent variables 



and N - number of cases. 



