Search

Ecologists routinely set out to estimate the trophic position of individuals, populations, and species composing food webs, and nitrogen stable isotopes (δ(15)N) are a widely used proxy for trophic position. Although δ(15)N values are often sampled at the level of individuals, estimates and confidence intervals are frequently sought for aggregations of individuals. If individual δ(15)N values are correlated as an artifact of sampling design (e.g., clustering of samples in space or time) or due to intrinsic groupings (e.g., life history stages, social groups, taxonomy), such estimates may be biased and exhibit overly optimistic confidence intervals. However, these issues can be accommodated using hierarchical modeling methods. Here, we demonstrate how hierarchical models offer an additional quantitative tool for investigating δ(15)N variability and we explicitly evaluate how δ(15)N varies with body size at successively higher levels of taxonomic aggregation in a diverse fish assemblage. The models take advantage of all available data, better account for uncertainty in parameters estimates, may improve inferences on coefficients corresponding to groups with small to moderate sample sizes, and partition variation across model levels, which provides convenient summaries of the 'importance' of each level in terms of unexplained heterogeneity in the data. These methods can easily be applied to diet-based studies of trophic position. Although hierarchical models are well-understood and established tools, their benefits have yet to be fully reaped by stable isotope and food web ecologists. We suggest that hierarchical models can provide a robust framework for conceptualizing and statistically modeling trophic position at multiple levels of aggregation.

Three lipid correction models were evaluated for liver and white dorsal muscle from Squalus acanthias. For muscle, all three models performed well, based on the Akaike Information Criterion value corrected for small sample sizes (AIC(c) ), and predicted similar lipid corrections to δ(13) C that were up to 2.8 ‰ higher than those predicted using previously published models based on multispecies data. For liver, which possessed higher bulk C:N values compared to that of white muscle, all three models performed poorly and lipid-corrected δ(13) C values were best approximated by simply adding 5.74 ‰ to bulk δ(13) C values., (© 2011 The Author. Journal of Fish Biology © 2011 The Fisheries Society of the British Isles.)