Kimura and Lyons: Between-reader bias and variability in age-determinations 



57 



Because percentage agreement decreases with the age 

 of fish (Tables 1-6), and age distributions vary greatly 

 among different species and among samples of the 

 same species, percentage agreement lends itself only 

 to age-specific comparisons. This is illustrated above 

 by Pacific ocean perch and sablefish which show a 

 similar percentage agreement; however, Pacific ocean 

 perch is much more "ageable" than sablefish, as re- 

 flected in the corresponding coefficients of variation. 



Although percentage agreement and coefficient of 

 variation both reflect the relative difficulty of ageing 

 each species, only the coefficient of variation adjusts 

 for the absolute age of the fish. Therefore, one might 

 conclude that the easiest species to age are Pacific 

 whiting (Table 1) and yellowfin sole (Table 2); and the 

 medium-difficult species are Pacific ocean perch (Table 

 3) and walleye pollock (Table 4). 



The most difficult species to age— species with un- 

 resolved criteria, or species for which readers needed 

 further training— were Atka mackerel (Table 5) and 

 sablefish (Table 6). In fact, in this study the age-reader 

 for Atka mackerel was inexperienced with the species, 

 and there were unresolved criteria for sablefish. 



The most important usage of reader/tester data is 

 in maintaining quality control. Unlike the data pre- 

 sented in Tables 1-6, for quality-control purposes we 

 need to compare only one tester with one reader, with 



the nominal age being the tester age. For pollock (Table 

 7) there were significant between-reader biases at older 

 ages in the case of inexperienced reader A. There were 

 no such between-reader biases for the experienced 

 reader B. To ensure data quality, these types of be- 

 tween-reader biases are constantly reviewed and the 

 samples partially re-aged, before the data are released 

 for use. 



Sablefish is an especially difficult species to age 

 (Table 8). There were so many problems with age deter- 

 mination that we suspended ageing, reviewed criteria 

 with other ageing labs, and re-aged several large sam- 

 ples. It is probable that between-reader bias for this 

 species can be reduced, but it is doubtful that the coef- 

 ficient of variation for this species can be substantial- 

 ly reduced. Nevertheless, the availability of reader/ 

 tester data was useful in revealing problems. Also, data 

 users deserve a quantitative presentation of variabil- 

 ity in age determinations and may be able to use these 



