Bias in Using an Age–Length Key to Estimate Age-Frequency Distributions

Abstract

Consider 2 representative samples of fish [e.g., Sebastes alutus] taken in different years from the same fish population, this being a population in which year-class strength varies. For the parental sample the length and age of the fish are determined and are used to construct an age-length key, the fractions of the fish in each (short) length interval that are of each age. For the filial sample only the length is measured, and the parental age-length key is used to compute the corresponding age distribution. Trials show that the age-length key will reproduce the age-frequency distribution of the filial sample without systematic bias only if there is no overlap in length between successive ages. Where there is much overlap, the age-length key will compute from the filial length-frequency distribution approximately the parental age distribution. Additional bias arises if the rate of growth of a year-class is affected by its abundance, or if the survival rate in the population changes. The length of the fish present in any given part of a population''s range can vary with environmental factors such as depth of the water; nevertheless, a sample taken in any part of the range can be used to compute age from the length distribution of a sample taken at the same time in any other part of the range, without systematic bias. But this of course is not likely to be true of samples taken from different populations of the species.