Use Of Bayesian Priors Research Articles

The use of Bayesian priors in Ecology: The good, the bad and the not great

Abstract Bayesian data analysis (BDA) is a powerful tool for making inference from ecological data, but its full potential has yet to be realized. Despite a generally positive trajectory in research surrounding model development and assessment, far too little attention has been given to prior specification. Default priors, a sub‐class of non‐informative prior distributions that are often chosen without critical thought or evaluation, are commonly used in practice. We believe the fear of being too ‘subjective’ has prevented many researchers from using any prior information in their analyses despite the fact that defending prior choice (informative or not) promotes good statistical practice. In this commentary, we provide an overview of how BDA is currently being used in a random sample of articles, discuss implications for inference if current bad practices continue, and highlight sub‐fields where knowledge about the system has improved inference and promoted good statistical practices through the careful and justified use of informative priors. We hope to inspire a renewed discussion about the use of Bayesian priors in Ecology with particular attention paid to specification and justification. We also emphasize that all priors are the result of a subjective choice, and should be discussed in that way.

Lumpers vs. splitters: Intelligence in children with specific learning disorders

A multigroup bifactor model was used to compare the explained variance and the reliability of the general vs. specific factors of the Wechsler Intelligence Scale for Children-4th edition (WISC-IV) in 1617 Italian children diagnosed with specific learning disorder (SLD) and an Italian normative sample of typically-developing children (with the exclusion of IQ < 70). Results suggested that more than half of the common variance, 56.1%, was accounted for by the domain-specific factors in SLD, against only 39.5% in typical development. The reliability of both general and specific factors was rather limited in SLD, whereas the reliability of the g-factor was good in typical development. An additional analysis using previous information from American data showed very similar results. Our results suggest that the role of the specific factors, of VCI and PSI in particular and WMI to a lesser extent, should be considered as probably largely distinct from the g-factor in children with SLD. Results also seem to indicate that the PRI is a less distinctive factor, which is, in the SLD group, hardly distinguishable from the g-factor. The use of Bayesian priors from American data indicated that results on Italian and American samples of children with SLD were similar, and different from those on the normative samples of both countries, suggesting remarkable cross-cultural and cross-linguistic similarity of the structure of intelligence in children with SLD.

Joint modeling of reaction times and choice improves parameter identifiability in reinforcement learning models

BackgroundReinforcement learning models provide excellent descriptions of learning in multiple species across a variety of tasks. Many researchers are interested in relating parameters of reinforcement learning models to neural measures, psychological variables or experimental manipulations. We demonstrate that parameter identification is difficult because a range of parameter values provide approximately equal quality fits to data. This identification problem has a large impact on power: we show that a researcher who wants to detect a medium sized correlation (r = .3) with 80% power between a variable and learning rate must collect 60% more subjects than specified by a typical power analysis in order to account for the noise introduced by model fitting. New methodWe derive a Bayesian optimal model fitting technique that takes advantage of information contained in choices and reaction times to constrain parameter estimates. ResultsWe show using simulation and empirical data that this method substantially improves the ability to recover learning rates. Comparison with existing methodsWe compare this method against the use of Bayesian priors. We show in simulations that the combined use of Bayesian priors and reaction times confers the highest parameter identifiability. However, in real data where the priors may have been misspecified, the use of Bayesian priors interferes with the ability of reaction time data to improve parameter identifiability. ConclusionsWe present a simple technique that takes advantage of readily available data to substantially improve the quality of inferences that can be drawn from parameters of reinforcement learning models.

Uniform, uninformed or misinformed?: The lingering challenge of minimally informative priors in data-limited Bayesian stock assessments

A Bayesian approach to parameter estimation in fisheries stock assessment is often preferred over maximum likelihood estimates, and fisheries management guidelines also sometimes specify that one or the other paradigm be used. However, important issues remain unresolved for the Bayesian approach to stock assessment despite over 25 years of research, development, and application. Here, we explore the consequence of a common practice in Bayesian assessment models: assigning a uniform prior to the logarithm of the parameter representing population scale (log-carrying capacity for biomass-dynamics models, or log-unfished recruits for age-structured models). First, we explain why the value chosen for the upper bound of this prior will affect parameter estimates and fisheries management advice given two properties that are met for many data-poor stock assessment models. Next, we use three case studies and a simulation experiment to show a substantial impact of this decision for data-limited assessments off the US West Coast. We end by discussing four methods for generating an informative prior on the population scale parameter, but conclude that these will not be suitable for many assessments. In these cases, we advocate that maximum likelihood estimation is a simple way to avoid the use of Bayesian priors that are excessively informative.

Assessing geographic origins of Teal (Anas crecca) through stable-hydrogen (δ 2H) isotope analyses of feathers and ring-recoveries

Management of migratory game birds must be considered at the flyway scale. Flyway delineation has generally relied on ring recoveries, yet migratory connectivity between breeding and wintering grounds is often difficult to demonstrate due to low ring-recovery rates. To overcome the limitations of ring recoveries, we here relied on stable-hydrogen isotope ratios (δ2H) from feathers of 939 hunted Teal (Anas crecca) in France in the winters of 2002–2003 to 2005–2006 to determine the geographic origin of these individuals. Likelihood-based assignment analyses, including the use of Bayesian priors based on ring recoveries, suggest that Teal wintering in France mostly originate from Russia, up to the Ural Mountains, and to a lesser extent from Finland. This suggests that more collaboration with Russia for the management of Teal in Europe should be encouraged. Our study identifies Teal breeding hotspots, which should receive greater research and monitoring attention. Our isotope assignment results can be used as a reference value in time against which future similar analyses could be compared, potentially to track geographic shifts linked with habitat and climate change.

ESTIMATING LINEAR PRODUCTION COEFFICIENTS FROM FARM BUSINESS SURVEY DATA: A NOTE

This paper presents the results of three approaches to estimating linear production coefficients from sample farm data by apportioning total input usage between individual enterprises. It is suggested that the use of Bayesian priors to conduct constrained estimation is preferable to the use of Inequality Restricted Least Squares.