Search the Community
Showing results for tags 'Bayesian'.
Found 3 results
Hi All, I'm currently in the process of running ZIP models in rjags to test some hypotheses about duck pair abundance. The zero-inflated models include only an intercept parameter and then 3 different dummy parameters to represent the wetland cover class variables 2, 3, and 4. In theory, one would expect the non-transformed value of the parameter estimates to decrease in value from cover class 1 to 4 sequentially. I'm seeing this trend in the MLE models we ran first actually: (Intercept) WCCone WCCthree WCCtwo -6.4537 6.0062 0.8016 3.5753 Unfortunately, I'm running up against a bit of a snag in rjags. The parameter estimates for the wetland cover classes in the zero-inflated portion of the model all end up being relatively the same (~-7 - -8) and are truncated at -10 (see the not-so-nice looking trace plots in attached word file: a2,a3,a4). I've checked for correlations among the covariates, mistakes in the data, and at this point, I am concerned that this might be a coding problem. I realize this is somewhat of an inane question to ask, however, I'm unfortunately not in an academic lab where I can easily find a fellow JAGS coder, so usually end up having to turn to sources like these if I want a second set of eyes. So, I'm hoping there's someone out there who is willing to look through the attached code and let me know if they see an error that might be causing these weird results. I've literally spent the last 6 months digging through these data and this code so I think my eyes cross every time I look at it. Any help would be great. Thanks, KC Example.docx KCScript.R
Bayesian Workshop for Ecologists and Wildlife Biologists Texas State University, San Marcos, Texas JUNE 1-3, 2016 Instructors Dr. William A. Link USGS Patuxent Wildlife Research Center, USA Dr. Richard J. Barker University of Otago, New Zealand Cost: Students - $299.00, Non-students - $499.00 Dorms available: $45/night, Linen charge $70 (optional) Registration web page and link to tentative outline of topics http://www.txstate.edu/continuinged/Events/Bayesian-Workshop.html Questions: Dr. Butch Weckerly email@example.com or Dr. Jeff Hatfield firstname.lastname@example.org William Link received his Ph.D from the University of Massachusetts, Amherst in 1986. After a year on the faculty of Towson University, Link was hired as Mathematical Statistician at the Patuxent Wildlife Research Center (PWRC) in Laurel, Maryland, where he has collaborated on analyses of count surveys, demographic analyses, mark-recapture, contaminant studies and many other aspects of wildlife statistics. In the mid-1990s, he dabbled with Bayesian methods, and became hooked. After early experience as a fish and game officer in New Zealand, Richard Barker spent a year at PWRC. Link and Barker’s early acquaintance led to a collaboration that is in its third decade, with important contributions as early advocates of Bayesian methods for wildlife statistics. Their recent work has focused on Bayesian multimodel inference, and lead to a book “Bayesian inference, with ecological applications” published in 2010. After Barker’s stint at PWRC, he returned to New Zealand, earning his Ph.D at Massey University. Barker is now Professor and Chair of Statistics at the University of Otago in Dunedin, New Zealand. Workshop participants will receive a free copy of Link and Barker’s book.
Analyzing Ecological Data with Hierarchical Models Five-day Workshop: 15-18 Mar 2016 Time: 9 am – 5 pm Location: Rosenstiel School of Marine & Atmospheric Science – University of Miami, Miami, Florida Instructor: Dr. Robert M. Dorazio Wetland and Aquatic Research Center U.S. Geological Survey Gainesville, Florida 32653 Email: email@example.com Course Description: This workshop is designed to provide wildlife scientists with the training needed to formulate and fit hierarchical models of animal abundance and occurrence to actual data sets. Most of these models can be fitted using either frequentist or Bayesian methods of analysis. However, Bayesian methods will be emphasized in the workshop because scientific interest is often focused on a model’s latent state variables or predictions (such as the abundance or occurrence of animals at sampled or unsampled locations) and computing inferences for these quantities is often difficult with frequentist methods. The workshop begins with a description of hierarchical modeling and its use in the analysis of ecological data. This is followed by a conceptual review of the frequentist system of inference and by an introduction to Bayesian methods of analysis. Next is a description of a generic set of algorithms (Markov chain Monte Carlo) that can be used to fit all of the hierarchical models described in the workshop – and many more! Armed with these tools, participants of the workshop will learn to derive and implement simulation-based algorithms for fitting hierarchical models to different types of data (as opposed to relying on software such as BUGS or JAGS). The workshop is intended to provide substantial hands-on training. Considerable time will be devoted to the technical details of fitting hierarchical models to data and to summarizing and interpreting the results of each analysis. As time permits, data brought to the workshop by participants will be presented and analyzed as class exercises. Prerequisites: (1) working knowledge of the R software program (R Core Team, 2015), (2) laptop computer with R installed, (3) familiarity with the frequentist system of statistical inference (including maximum likelihood estimation), and (4) familiarity with the analytical evaluation of joint, marginal, and conditional probability density functions, particularly as they are used in specifying mixtures of distributions. RSMASworkshopOutline2016.pdf