Gilpatrick Estimation of dolphin school size with aerial photography 



645 



E 

 s 



o 



u 



200 



400 



School size 



600 



Figure 1 



Comparison of mean school size and precision (using CVl of count estimates for 

 schools (/?=48) photographed during the 1989 ETP dolphin population survey. 



levels of the factors "school size" and "quality rating" 

 (see between-factors; Table 4). This was expected be- 

 cause photographed schools of different sizes were in- 

 tentionally selected for the analysis. Often, with RM- 

 ANOVA, between-factor effects are confounded with 



different levels of the factors (as is true 

 for the two factors above) and it is the 

 temporal trends of the within-factor 

 data that are of primary interest in 

 the analysis (Winer, 1971, p. 299). 

 There were no differences between 

 readers in the overall means of their 

 repeated counts (temporal trends not 

 considered; P=0.7898; Table 4). 



The F value for the within factor 

 "Time" indicated no significant linear 

 trend for repeated-counts (with the 

 mean of the three reader counts). How- 

 ever, a significant interaction for 

 "reader with time" <F=3.503; P=0.0258; 

 df=6, 31) indicated there were differ- 

 ences between readers in the respec- 

 tive temporal trends of their repeated- 

 counts (Fig. 3). To investigate the 

 source of the significant effect, the 

 trend data for each reader were tested 

 a posteriori by contrasts among means 

 with linear coefficients to weigh the 

 log-transformed data ( Sokal and Rohlf, 

 1981; Abacus Concepts, 1990). Results 

 showed the mean of Reader l's counts 

 at time one differed significantly from 

 the means of counts made at times 

 two, three, and four (F=14.977; 

 P=0.0005; df=l). This suggested that 

 Reader l's counts made at time one 

 were important in causing the significant "reader with 

 time" interaction. 



Because the initial counts (at time one) of the ex- 

 periment schools were done just prior to the start of 

 the 12-week counting period, this suggested that 



