Statistical distribution of age readings of known-age sablefish (Anoplopoma fimbria)

A mark-recapture experiment provided a collection of 172 known-age sablefish from Alaska waters. Otoliths from each fish were read by three readers. The readings have a positive bias among young fish and a negative bias among older fish. Among otoliths of the same age, some tended to give consistently high or low annulus counts, so that the variance of age readings at each true age was about half due to variance among otoliths and half due to variance among replicate readings of individual otoliths. The statistical distribution of age reading errors is well described by an asymmetrical two-sided geometric distribution with age-varying parameters. For comparison, the error distribution was estimated with naive methods that do not use the known ages and that assume the readings are unbiased. These estimated distributions do not match the actual error distributions very well, but they do a surprisingly good job of predicting the distribution of age readings from a stock assessment model's internal estimate of a true age composition. They also produce estimates of recruitment and biomass close to those obtained with the actual error distributions when used in the present sablefish stock assessment.

► Among otoliths of a given true age, some tend to give consistently high readings and some low.
► About half the variance of sablefish age readings is due to otoliths and half to readers.
► The empirical distribution of age reading errors is well fitted by an asymmetrical geometric.
► Estimates of error distributions based solely on replicate readings perform surprisingly well.
► Accounting for aging error in stock assessment naively is far superior to ignoring it.