investigation was to determine our ability to detect changes in the 

 demersal finfish assemblages using the accumulated data, and to distinguish 

 plant-induced changes from natural changes in the fish populations. 



Hypotheses generation and statistical testing provide the most 

 quantitative way to detect differences or changes. Standard normal 

 theory tests assume, among other things, that the observations come from 

 a probability distribution that is normal, that the variances are homo- 

 geneous, and that the observations are independent. A knowledge of the 

 data base characteristics, including any changes in methodology that 

 might effect nonbiological changes in the data, is desirable for the 

 correct application and interpretation of various tests and results. 



Several changes in the trawl program have occurred since its inception 

 in 1973. These have been described along with the reasons for the 

 changes in a previous section of this report. The way in which the data 

 were manipulated in Annual Reports was one consequence of these changes. 

 Because not all of the currently sampled stations have been sampled 

 since 1973, but year to year comparisons were desired, some stations 

 were grouped together into regions. The data were summed in a region 

 and month and divided by the total tow time and multiplied by 15 to 

 arrive at a CPUE. This allowed comparisons over time, but the information 

 contained in the variance of the individual data points was lost. The 

 other major changes in the trawl program affecting the data was the 

 conversion from a time-based CPUE to a distance-based CPUE. The effect 

 of changing unit-effort bases was not measured experimentally. The 

 results of mathematically manipulating the data then, was to add unknown 

 and unmeasured amounts of variance to the data collected after October 

 1977. Detected differences might be related to environmental or plant- 



33 



