FISHERY BULLETIN: VOL 76, NO. 3 



RESULTS 

 Study A 



Integration of values 



The results of study A are summarized in Table 

 2. The total chlorophyll values derived from the 

 four systematic sample designs were calculated by 

 integration with linear interpolation (i.e., using 

 the arithmetic mean of adjacent samples to repre- 

 sent the average chlorophyll in the stratum be- 

 tween them). Use of the geometric mean ir. this 

 calculation resulted in the true total being under- 

 estimated 27 out of 36 times (P = 0.01). Nor was 

 there any increase in accuracy (the resultant ac- 

 curacies, based on use of the geometric mean, were 

 0.538, 0.987, 0.488, and 0.752 for RSS-1 through 

 RSS-4). The use of the geometric mean in the cal- 

 culation of total chlorophyll does not appear to be 

 justified. 



Bias 



The biases observed in the eight sampling 

 strategies are summarized in Table 3. Of the four 

 restricted systematic designs, only RSS-2 gave no 

 signs of bias. Design RSS-3, the "best informed" 

 design, overestimated the true population total in 

 eight of the nine trials (P<0.05). RSS-1 overesti- 



Table 3. — Bias of systematic and stratified random sampling 

 designs, Study A. 



Date 



Systematic designs 



Stratified random designs 



mated the population only fivir- out of nine times, 

 but the overestimates were clustered toward the 

 end of the series and the underestimates toward 

 the beginning. This temrjral trend lies just out- 

 side the usual level of significance (run test; 

 P<0.10) but it indicates that the time period 

 necessary for the population to provide "random" 

 variability of sufr.cient magnitude to eliminate 

 bias may be of tht- order of several days or 100-200 

 km. The magnitudes of the biases were —4.09^ for 

 the period 19-21 September and +3.77c for the 

 period 24-27 September. Similarly, the bias intro- 

 duced by using RSS-3 to estimate the true popula- 

 tion total for 19-25 September was -3.69c. 



The peculiar periodicity of bias seen in RSS-4 

 also indicates a nonrandom interaction between 

 the sampling design and the sampled population 



Table 2. — Restilts of study A, a computerized simulation sampling study. The estimated parameter ( 6) is total chlorophyll above 180 



m; units are milligrams per square meter; time is local time. 



'Mean values from 10 sets of unreplicated casts. 



622 



