Influence of sample size and number of age classes on characterization of ageing error in paired-age comparisons Fisheries Research

Diagnosis of ageing error is critical to the proper interpretation of age data used in fisheries science and management. However, the influence of sample size and number of age classes on the characterization of ageing error has not been thoroughly evaluated. We conducted a simulation study of ageing error diagnostics for paired-age comparisons across 648 scenarios differing in 1) number of age classes, 2) total number of samples aged, 3) trend in sample size by age, and 4) magnitude and type of imprecision and bias. Imprecision was identified by comparing average coefficient of variation (ACV) with two common thresholds. Bias was evaluated using maximally (McNemar’s), diagonally (Evans & Hoenig), and unpooled tests of symmetry (Bowker’s). Imprecision was identified less frequently at low to moderate (=6% of runs) levels of random vs high (=55% of runs) error, and ACV was artificially inflated in the presence of bias. McNemar’s and Evans & Hoenig bias tests outperformed Bowker’s (=4% vs 29% false positives), particularly at large sample sizes, and its use is strongly discouraged. This study can help guide the interpretation of ageing error studies and their products (e.g., ageing error matrices) used to inform stock assessment and management. Grant no. NA19OAR4320074