A stepwise regression procedure that maximized R values was used to select the best combination 

 of the above variables for the deterministic portion of each ARIMA model. Next, appropriate stochastic 

 terms were added to the deterministic model whenever the residuals were found autocorrelated. The 

 methods of Brocklebank and Dickey (1986) were used to identify the stochastic structure of the residuals 

 from the deterministic function. 



Interpretation of time-series model statistics. Because the time-series models presented in this report 

 characterized the abundance levels and natural fluctuations of local fish populations during the two-unit 

 operation period (1976-1985), the model forecasts consitute the two-unit standard- or reference-series 

 against which future three-unit operational monitoring data will be compared. Several summary statistics 

 reported with the results of the time-series models will be used for impact assessment purposes in the 

 future. ITie "errors" or deviations of each data point from the model forecast were separated into "above " 

 (positive deviations) and "below" (negative deviations) and added up by years. These sums of deviations 

 indicated whether the year being examined was above or below the two-unit reference series. When the 

 deviations are squared, the annual sums become the annual components of the model sum of squares 

 error (SSE). The mean squared errors (MSB's) were obtained by dividing the SSE's by the number of 

 degrees of freedom associated with each year and model. These MSB" values provide a basis for impact 

 assessment because any annual MSB can be statistically compared to the reference series model MSB by 

 means of a simple F-test obtained as the ratio of the two MSB's. It is expected that the annual fluctuations 

 of MSB during the three-unit operation period, will be within the range of MSB values reported for this 

 preoperational period. Note that the periods modeled at BN and NB are not the same (1976-1985 versus 

 1979-1985). This may cause the summary statistics to show different patterns of deviations from the 

 models that characterize temporal fluctuations at the two stations (Appendices XIV through XXVHI). 



Table 7. The "species and program combinations for which the mulliplier variable for season (S) in their lime-sencs models wns 

 set equal to one. 



Species Program Season 



AmmodytP.^ amerkanus Larval November - July 



Anchoa spp. ■ Larval May - November 



Egg April - August 



Menidia spp. Seine May - December 



Mynxocephaliis arnaeus Larval January - June 



Taulogolahrus adspprms Larval May - October 



Hgg April - August 



Tautnga onitis Larval May - October 



Egg April - August 



19 



