Bayesian Meta-Analysis of Literature Information: Estimating Functional Traits for 305 Tree Species
Kiona Ogle, University of Wyoming
Monday, September 28, 2009
4:00 p.m., 223 Weber
We describe a hierarchical Bayesian (HB) approach to conducting meta-analyses of information in the literature. The HB approach overcomes important limitations of classical meta-analysis methods such as incomplete reporting (e.g., missing or unreported samples sizes, error estimates, or important covariates) and non-independence of within study information. Incomplete reporting results in a large quantity of missing covariate information, and we estimate this missing information within the HB framework. In doing so, we explore different models that control the feedback between the response variable of interest, the missing covariate information, and parameters of interest. We illustrate our HB meta-analysis models with literature data obtained for specific leaf area (SLA), an important plant functional trait affecting plant carbon gain. A goal of the analysis is to obtain species-specific SLA estimates for 305 tree species occurring in the United States. We have literature information on measured SLA for about 50% of the species, and we employ a hierarchical parameter model based on taxonomies to obtain estimates of the species-specific SLA parameters for all 305 species. These parameter estimates will ultimately be used in process-based models of tree growth and mortality for understanding forest dynamics.