Solutions to pressing environmental problems require understanding connections between human and natural systems. Analysis of these systems requires a model that can deal with complexity, is able to exploit data from multiple sources, and is honest about the uncertainty from multiple sources. Synthesis of results from multiple studies is often required. Bayesian hierarchical models provide a powerful approach to analysis of socio-environmental problems.
Past participants of this short course have worked on research questions including the use of network analyses to understand measurement uncertainly in the context of extreme weather events, the study of governance effectiveness and fisheries biomass, the effect of changing climate on population dynamics of polar bears, and the relationship between advocacy group compositions and estuarine quality.
The National Socio-Environmental Synthesis Center (SESYNC) will host a nine-day short course May 29 - June 8, 2018 covering basic principles of using Bayesian models to gain insight from data. The goals of the course are to:
- Provide a principles-based understanding of Bayesian methods needed to train students, evaluate papers and proposals, and solve research problems.
- Communicate the statistical concepts and vocabulary needed to foster collaboration between ecologists, social scientists, and statisticians.
- Provide the conceptual foundations and quantitative confidence needed for self-teaching modern analytic. methods.
The course will enable participants to:
- Explain key principles of Bayesian statistics, including the concepts of joint, conditional, and marginal probabilities; posterior and prior distributions; likelihood; conjugacy; and the relationship between Bayesian and maximum likelihood approaches to inference.
- Use basic statistical distributions (e.g., binomial, Poisson, normal, log normal, multinomial, beta, Dirichlet, gamma, multivariate normal) to write joint and conditional posterior distributions for hierarchical Bayesian models that couple models of socio-ecological processes, models of data, and random effects.
- Explain how Markov chain Monte Carlo (MCMC) methods can be used to estimate the posterior distributions of parameters.
- Write algorithms and computer code in R implementing MCMC methods to estimate parameters in simple models.
- Use JAGS software to implement MCMC methods for estimating posterior distributions of parameters, latent states, and derived quantities.
- Evaluate model convergence and assess goodness of fit of models to data.
- Develop and implement hierarchical models that explicitly partition uncertainties.
- Understand the basis for statistical inference from single and multiple Bayesian models.
- Use Bayesian methods to synthesize results from multiple scientific studies.
- Understand Bayesian methods for modeling spatially structured data.
Short Course Format
The course will include lectures and laboratory exercises. Labs will emphasize problem solving requiring programming in R and JAGS.
Short Course Details
- The course will be held May 29 - June 8, 2018 at SESYNC in Annapolis, Maryland, and will meet daily from 9 am – 5 pm. We may meet at 8 a.m. the first day.
- There will be no meeting on Sunday, June 3.
- The course is aimed at postdoc, researcher, and faculty participants. Grad students may also be considered.
- There is no fee to attend, but participants are responsible for most of their own travel and accommodations. For those willing to share a room with another participant, SESYNC will provide local hotel accommodations. If a private hotel room is desired, SESYNC will cover 50% of hotel costs.
- Coffee/tea and lunch will be provided daily.
- Complete the webform here. In addition to your CV, you will need to submit a paragraph for each of following topics:
- Describe your background in mathematics and statistics. How competent are you with the programming language R and how often do you typically use R?
- Describe how you plan to apply what you learn in this course to a research project of your own. Please be as specific as possible.
Dr. Tom Hobbs has taught ecological modeling at Colorado State University for 15 years. His course has evolved over time; during the last eight years, it has emphasized Bayesian methods for gaining insight from models and data. He has also taught short courses for the U.S. Geological Survey, Conservation Science Partners, the Woods Hole Research Center, the Grimsö Wildlife Research Institute, and the Department of Ecology, Swedish Agricultural University. He is the author, with Mevin Hooten, of Bayesian models: A statistical primer for ecologists from Princeton University Press. Dr. Hobbs takes special pride in making challenging, quantitative concepts clear and accessible to students who never considered themselves to be particularly adept with mathematics and statistics.
Dr. Mary Collins is an environmental social scientist and Assistant Professor of Environmental Health at the College of Environmental Science and Forestry at the State University of New York (SUNY-ESF). Dr. Collins uses hierarchical Bayesian models to assess inequalities in pollution generation between US-based industrial facilities and potential human health impacts. Currently, she is working on the temporal dimensions of hazardous waste generation as it relates to links between specific chemical exposures and rare cancers in New York State. Since participating in an earlier version of this course, she has been a co-instructor since 2015 and is specifically interested in the translation of concepts across disciplinary boundaries.
Dr. Christian Che-Castaldo is an ecologist and postdoctoral associate with Dr. Heather Lynch at Stony Brook University. Dr. Che-Castaldo uses hierarchical Bayesian models to explain the occupancy and population dynamics of Antarctic wildlife. He helped create the online application for penguin population data that is used for Southern Ocean feedback management (www.penguinmap.com). Currently, he is developing a modeling approach to incorporate the size of penguin guano deposits captured with high-resolution satellite imagery directly into Bayesian models of penguin distribution and abundance in Antarctica. Also a participant in one of Dr. Hobbs’ earlier workshops, he has been a co-instructor since 2015 and served as an external reviewer for Bayesian models: A statistical primer for ecologists.
The University of Maryland is an Equal Opportunity Employer
Minorities and Women Are Encouraged to Apply