2 

 a . The sample size, N, required to detect a specified difference, d, at 



various a and ftlevels was given by Snedecor and Cochran (1967): 



N = (Z a + Z3) * 2a 2 / d 2 (5) 



where Z = normal deviate such that the area from -00 to Z 

 a is (1 - a) 

 Zg = normal deviate such that the area from -°° to Z 



is (1 - 3) 

 d = log ? R 

 a 2 = variance estimated from ANOVA 



The In transformed catch data were analysed as time series. The 

 autocorrelation structure of the data was investigated by visually 

 evaluating the autocorrelation function (ACF) generated by the PROC 

 ARIMA procedure of SAS. The general hypotheses being considered were 

 that the catch did not change over time (i.e., no power plant effect). 

 A "change" could be an increase or decrease in the average values (trend) , 

 or an alteration of the frequency or amplitude of the fluctuations 

 (cyclical components) of the values over time. While these specific 

 changes can be tested with time series methodology (Box and Jenkins, 

 1976), as a first step an alternate approach was taken. The monitoring 

 data collected prior to the current year were used to construct auto- 

 regressive models. Because long term constant values were hypothesized, 

 the data were not detrended (long term trend removed from the data by 

 regression) . Detrending data that are primarily the results of stochastic 

 processes is a procedure that may give misleading results in forecasting 

 models (Box and Jenkins, 1976). Autocorrelations between the catch at 

 time 'i' and 'n' time units away (up to 26 in this case) were calculated. 

 If any correlations were significantly different from zero at a =0.05, 

 autoregressive terms of "An" were included in the model. For example, 



10 



