Modeling Growth: Consequences from Selecting Samples by Size

Simulated size–age observations were collected from hypothetical populations with known growth to quantify potential bias in estimates of growth and mortality that can result from size‐selective sampling. The population lengths at ages were normally distributed around means predicted by published von Bertalanffy ( VB ) equations for the Striped Marlin Kajikia audax or Vermilion Snapper Rhomboplites aurorubens . Parameter estimates for the VB equation were accurate when the data were collected with random and age‐stratified sampling. Size‐stratified sampling caused important bias in estimates of length at age and fitted VB parameters. The bias increased with the coefficient of variation of mean size at age and total mortality. It became more severe when sampling effort was increased to compensate for low sample numbers in bins within the upper tail of the distribution of sizes in the populations. Error occasionally exceeded 100% for asymptotic length L ∞ and 50% for the Brody growth coefficient K . When a random sample was supplemented with ≤10% large fish, the bias was sometimes greater than that observed when a completely size‐stratified design was used. Such error could corrupt the reliability of analyses that depend on VB parameters. For Striped Marlins, natural mortality computed with the Pauly (1980) equation using results from length‐selective sampling strategies often deviated 50% or more from the values predicted using the “true” growth coefficients. Potential for nonrepresentative sampling of lengths at age should be routinely described when VB models are first published and should be considered whenever the growth estimates are later used for other purposes.