54 



Fishery Bulletin 89(1), 199! 



species. For example, a species that can be aged with 

 a larger percentage agreement (percentage of speci- 

 mens aged the same on two occasions by the same or 

 different reader), or smaller coefficient of variation, 

 provides a statistical confirmation of the statement that 

 species "x" is easier to age than species "y." 



Between-reader bias provides a measure of the ade- 

 quacy of criteria for distinguishing ages in a particular 

 species of some nominal age. Presumably, if there is 

 no between-reader bias, ageing criteria are being ap- 

 plied similarly by both readers, and the data only con- 

 tain random measurement error. If between-reader 

 bias and measurement error are independent, at this 

 point between-reader variance would also be mini- 

 mized. Significant between-reader bias may indicate a 

 lack of resolving power in the criteria, insufficient 

 training, or even peculiarities in the structures being 

 aged. Between-reader variance is generally an indi- 

 cator of overall "ageability," but is not as effective as 

 between-reader bias measurements for pointing out 

 between-reader differences in criteria. 



Species often have a characteristic age above which 

 between-reader biases become larger. This age may be 

 interpreted as a line distinguishing which ages are more 

 reliable. For age-readers themselves, between-reader 

 bias is usually of more interest than variability. This 

 is because while measurement error is inherent in the 

 age-determination process, between-reader bias can be 

 controlled to a greater extent. 



In age-determination studies the term "precision" is 

 used to describe "agreement," or variability between 

 readings of the same specimen by the same or different 

 age-reader. The term "accuracy" is reserved to de- 

 scribe a comparison of ages generated by readers with 

 the "true" age for specimens of known age. 



By emphasizing the importance of between-reader 

 bias and variability, we do not mean to denigrate the 

 obvious importance accuracy and age validation play 

 in the age-determination process (Beamish and McFar- 

 lane 1983, 1987). Validation (the comparison of ages 

 determined by counting rings on hard parts with known 

 ages) can be carried out in a variety of ways, all of 

 which are difficult. These include combining an ex- 

 ternal tag with an oxytetracycline (OTC) injection that 

 labels calcium rings with a mark visible under ultra- 

 violet light, following unusually strong year-classes 

 through time, ageing young fish of known ages, and, 

 most recently, measuring the activity of naturally 

 occurring radioisotopes. Scientists at the Pacific Bio- 

 logical Station (Beamish et al. 1983, Cass and Beamish 

 1983, Leaman and Nagtegaal 1987, McFarlane and 

 Beamish 1987) have made wide use of the OTC mark. 

 And, recently, two studies appear to have succeeded 

 in validating longevity in rockfish using radioisotopes 

 (Bennett et al. 1982, Campana et al. 1990). Typically, 



validation can be carried out on only a very few fish. 

 Often, doubts remain concerning criteria for certain 

 age groups, or structures that look different. Never- 

 theless, the validation process is a critical one, and age- 

 readers must constantly strive to improve the accuracy 

 of their age determinations. 



Because we seldom knew the true age of a fish, ab- 

 solute bias and total mean-square error in the ageing 

 process were not known. Therefore, our discussions 

 here will be limited to between-reader bias and vari- 

 ability. These quantities are defined by the between- 

 reader bias and coefficient-of-variation formulas 

 described in the following section. 



Materials and methods 



The Ageing Unit at the Alaska Fisheries Science 

 Center has the broad responsibility of ageing commer- 

 cially important fish species and fish stocks in U.S. 

 waters from California to the eastern Bering Sea. 

 Historically, data have been accumulated from three 

 principal sources: scientific surveys using various 

 fishing gear, and foreign and domestic vessels fishing 

 in U.S. waters. The present data consist of ages read 

 from the otoliths (ear bones) of various groundfish 

 species collected using assorted gear. 



Since 1981, the preferred method of reading ages 

 from these structures has been to either break or saw 

 the otolith cross-wise, burn the exposed surface, and 

 read the cross-section under a microscope (Chilton and 

 Beamish 1982). Only young or unusually clear speci- 

 mens of select species can be read from the intact 

 surface. 



In 1983 a quality-control program was initiated 

 wherein 20% of all routine age readings would be in- 

 dependently re-aged by an age-reader (i.e., the tester) 

 particularly experienced in a species. Statistics were 

 calculated on these reader/tester data (one reading per 

 otolith from each age-reader) in the following manner: 



1 mean (x) = (tester + reader)/2. 



2 standard deviation (SD) = v^[(tester - i) 2 + 

 (reader - x) 2 ] 



3 nominal age (age): x (rounded), or tester age 



4 n (count): sample size (number of specimens aged) 



5 percentage agreement: (n agree/n) x 100 



6 coefficient of variation (CV) = (SD/i) x 100 



7 between-reader bias: reader age - tester age 



8 percentage bias: [(reader age - tester age)/x] 

 x 100 



Elements 5-8 were averaged over the "n" specimens 

 of the same nominal age, and over all ages (weighted 

 by n) for overall statistics. 



