374 



Fishery Bulletin 91(2), 1993 



& Robson 1966). This assumes that the length:girth 

 ratio is the same for all fish. Each fork-length interval 

 was divided by the appropriate mesh size of capture, 

 and then multiplied by the appropriate observed fre- 

 quency. This produced "generic" size-frequencies by 

 mesh size. Each resulting size-frequency interval, by 

 mesh size, was normalized to the greatest frequency. 

 The resulting frequencies, one for each mesh size, were 

 then scaled to an assumed 100% efficiency at the peak 

 frequency sampled (the mid-60 cm mode); thus, differ- 

 ential sample sizes by mesh are taken into account. If 

 all mesh sizes have the same selectivity curve, and no 

 sampling variance, they would all be the same. 



Rather than using the method of Ishida to describe 

 a common selectivity curve (Regier & Robson 1966), 

 the "mean" shape of the selectivity curve was esti- 

 mated from the selectivity of the different mesh sizes 

 for the largest (78cmFL) and smallest (53 cmFL) 

 modes captured by all mesh sizes. The curve obtained 

 for the 53 cm mode described the left or ascending limb 

 of the selectivity curve, and the curve for the 78 cm 

 mode described the right or descending limb. Lengths 

 between these two limbs were assumed to be sampled 

 at 100%. This included some lengths with very low 

 percent-frequencies (i.e., we assumed few fish were 

 available) as well as the size mode near 62 cm. There 

 was no assumption that the left and right limbs have 

 the same shape. 



For the 78 cm mode, the cumulative percent- 

 frequency of capture vs. length of fish caught for all 

 mesh sizes are shown in Fig. 3. When plotted on nor- 

 mal probability scales, cumulative percent-frequencies 



48 50 52 



FORK LENGTH (CM) 



Figure 5 



Mean selectivity-curve left limb for albacore Thunnus alalunga 

 from the 53cmFL mode sampled in the 1980 JAMARC sur- 

 vey (JAMARC 1983). 



formed straight lines with virtually the same slopes. 

 An average cumulative percent-frequency (subtracted 

 from 100) provided the shape of the right limb (Fig. 4). 

 This introduced the assumption of a constant variance 

 and normal distribution (Hamley 1975). The same pro- 

 cedure provided the left limb of the selectivity curve 

 which increased more rapidly or had a greater slope 

 (Fig. 5). 



These two limbs must now be placed appropriately 

 on the X axis for each mesh size. This is accomplished 

 separately for each limb by mesh size. The sampling 

 efficiency at 53.5 cmFL for each mesh size was esti- 

 mated by linear regression. The observed maximum 



