The ability of two age composition error distributions to estimate selectivity and spawning stock biomass in simulated stock assessments

•A two by two cross of true and estimated error distributions for age composition were simulated.•The multinomial approach performed better than the adjusted lognormal approach in all four cases.•Both approaches generally estimated domed selectivity patterns when they did not exist, or stronger domes than were actually present.•Domed selectivity estimates should be treated with caution, and simulations should check for this bias in actual assessments.

A simulation study was conducted that examined two different error distributions for age composition data; the multinomial and the adjusted lognormal. Across 24 different simulation cases, and 4800 total data inputs, the multinomial error distribution consistently outperformed the adjusted lognormal error distribution, even when the true error was generated from the adjusted lognormal distribution. Both error distributions had a tendency to overestimate spawning stock biomass, but the bias was more severe for the adjusted lognormal. The only scenario in which the adjusted lognormal performed better was when selectivity was mis-specified; in this circumstance, the strong positive bias associated with the adjusted lognormal compensated for model runs that forced flat selectivity when selectivity was truly domed. Model selection criteria were overly sensitive when using the adjusted lognormal error. The utility of the adjusted lognormal error distribution in stock assessments is significantly diminished by the biased nature of the estimator, high variability of estimates, and inability to properly identify the correct selectivity pattern. Stock assessments that estimate domed selectivity patterns should have simulations conducted to evaluate if this bias is present. This can be accomplished by simulating data according to the error distributions used within the stock assessment and evaluating biases in estimates of selectivity and spawning stock biomass.