How Should Fishery Scientists Calculate Diet Proportions by Weight? A Simulation Study Evaluating Two Common Methods

Stomach content analysis is one of the lowest-technology techniques in fisheries science but diet data are critical to research in aquatic community ecology, conservation biology, invasive species ecology, and multispecies stock assessment.  In studies of predation impacts and flows of energy and nutrients, researchers generally wish to quantify the proportion by weight of the total aggregate diet of a fish population represented by a prey item of interest.  In these studies, fisheries scientists commonly use two distinct methods to estimate diet proportions by weight for a given prey item i: proportion by weight (W_i), in which a single diet proportion is calculated from the pooled contents of all stomachs, and mean proportion by weight (MW_i), in which separate proportions are calculated for each individual stomach and the mean of these values is reported.  These methods can produce very different diet proportions from identical input data, yet they are often used interchangeably in the literature.  I used a simulation study framework to evaluate the accuracy and precision of each method under conditions representing a wide range of predator-prey systems.  To my knowledge, this is the first quantitative study to evaluate the performance of these alternative methods.  I used an operating model to generate a set of pseudodata representing the diets of a population of predators, then sampled a subset of these data, analyzed the subset using both methods, and compared the results to metrics calculated from the full dataset.  This process was repeated across a range of operating model scenarios.  The W_i method was unbiased under all scenarios, approaching the true diet proportion with a large enough sample size.  The MW_i method was biased under most conditions, typically estimating diet proportions for large prey 10-70% below the true population value.  MW_i was most biased when the size difference between prey types was large and when large prey were rare.  However, the MW_i method was more precise than W_i, with estimates stabilizing faster as sample size increased.  The main results were consistent for both generalist and specialist predators.  I encourage caution in the use of the MW_i  method, which is likely to be accurate only under special conditions.  I propose an empirical test of these model results and discuss implications for studies using stomach content or stable isotope data to estimate diet.