Based on this cost, the number of stations and times per year to sample 

 for 1, 2 or 3 replicates to minimize the variance of the annual mean was 

 found to vary with the species under consideration (Table 9). However, 

 it appeared that increasing the times per year and decreasing the number 

 of stations sampled would be effective in reducing the variance. 



While normal theory tests are fairly robust with respect to devia- 

 tions from the assumptions of normality and homogeneous variances, the 

 effect of non- independence or autocorrelation of the data in time on 

 probability statements can be unsatisfactory (Glass et al. 1975). Time- 

 series analysis actually makes use of the internal autocorrelations and 

 has been proposed as a powerful tool for statistical testing in environ- 

 mental monitoring (Saila et al. 1980). The initial application of this 

 analytical method used the autocorrelation structure to forecast fish 

 catches a year in advance. However, techniques exist for removing the 

 autocorre-lations and then testing for differences in the remaining 

 values. Additionally, tests can be constructed to determine if an 

 intervention effect, like the start-up of a power plant, is detectable 

 in the series. This approach, then, would assist in meeting both the 

 second and the third objectives of the seining program, that is to be 

 able to distinguish natural changes in finfish population levels or 

 changes in community composition from those which are power plant induced, 

 Because 50 serial data points are needed to adequately identify a time 

 series model (Glass et al. 1975), and results are best if at least two 

 observations contribute to each data point (Saila, pers. comm.), it is 

 recommended that any changes in the sampling program be considered in 

 terms of the potential for using the extant historical data base in time 

 series analysis. 



71 



