Environmental Evidence Synthesis

I have a postdoc position open for someone interested in the assessment of overall uncertainty in risk assessment. It is part of a H2020 project about new methodologies for human risk assessment (Riskhun3r). The candidate will be working with uncertainty analyses for chemical and human health risk assessment, but I don’t require prior knowledge about these types of assessments. We want someone with a degree in mathematics or statistics or similar, with knowledge about Bayesian inference and an interest in subjective probability as a measure for epistemic uncertainty.

Read more and apply before Jan 13 2022

Note, I get a lot of applications from candidates doing fuzzy numbers, but this is not a measure we will look at in this postdoc. This research is about the use of subjective probability to quantify uncertainty.

Contact me ullrika.sahlin [at] cec.lu.se if you want to know more.

2021 is coming to an end and it is with deligth I conclude that two PhD students in the Uncertainty and Evidence Lab research group has finalised and successfully defended their theses. Maria Blasi Romero’s thesis Wild bees in agricultural landscapes: Modelling land use and climate effects across space and time and Ivette Raices Cruz’s thesis on Robust analysis of uncertainty in scientific assessments wraps up some of our research from the last four years.

Now we turn the page and go into 2022 with new adventures.

https://www.cec.lu.se/calendar/public-defence-doctoral-dissertation-environmental-science-ivette-raices-cruz

The application for the graduate course Bayesian Analysis and Decision Theory (NAMV005, 5 credits) is now open!

This year we do 5 meetings in January-February (four of them online and the last one on campus in Lund). PhD students, who are part of the ClimBECO research school will be given priority, but we welcome students from other research schools, as well. The course introduces Bayesian analysis using Stan (for MCMC sampling) and R (as an interface to Stan and a data analysis platform). The major part of the course covers Bayesian data analysis and statistical inference. We also put Bayesian analysis into the context related to subjective probability to quantify uncertainty and Bayesian decision theory. Application is open until 24th of December or when the spots are filled (whichever comes first). More information about the course and a link to the registration form can be found at https://canvas.education.lu.se/courses/5146

Welcome

Ullrika Sahlin and Dmytro Perepolkin

Centre for Environmental and Climate Sciences, Lund University

We encourage you to submit an abstract to and/or spread the word about the session Expert Knowledge Elicitation in environmental assessments – subjective, but scientific (Track 4-Ecological Risk Assessment and Human Health Risk Assessment of chemicals, other stressors and mixtures) at the forthcoming Society of Environmental Toxicology and Chemistry (SETAC) meeting in Copenhagen, in May 2022 (https://europe2022.setac.org ).

This will be an opportunity to share and discuss recent advances in expert knowledge elicitations and how such elicitations can support risk assessment. It will also be an opportunity to introduce expert elicitation as a scientific method to the wider community of SETAC.

The abstract submission deadline is 1 December 2021. Abstracts should be submitted via the online submission page https://europe2022.setac.org

We look forward to receiving your contribution!

Ullrika Sahlin, Anca Hanea and Dmytro Perepolkin (Session chairs)

Session abstract

Expert Knowledge Elicitation (a.k.a. Structured Expert Judgement) is a formal process for eliciting and combining judgments from subject-matter experts about the relevant quantities in scientific assessments. Expert opinion and judgment are used in statistical modelling and inference (e.g., as parameter estimates or as priors in Bayesian inference) and decision-making.  Elicited judgments about an uncertain quantity often take the form of a subjective probability distribution for that quantity. If judgments are obtained from several experts, they may need to be combined, unless the experts reach consensus.

The quality of the elicited data may be hampered by the experts’ limited experience with making probabilistic judgements and/or the poor use of elicitation protocols in practical applications. Because humans are prone to cognitive biases, logical errors and social influences (e.g., peer pressure) when making (probabilistic) judgements, formal Expert Knowledge Elicitation protocols have been designed to address and minimize these effects. Expert Knowledge Elicitation has been used for a long time, and is increasingly being used to fill data gaps and to quantify uncertainty. Disciplined and consistent elicitation based on scientific insight from cognitive psychology and probability theory, combined with the domain knowledge and practical experience in environmental assessment can turn subjective judgements into a solid scientific basis for making policy recommendations and assessing relevant risks.

The purpose of this session is to revisit the methodological framework and to bring up the recent advances within the Expert Knowledge Elicitation, particularly focusing on its usefulness for environmental assessment. We welcome contributions on methodological topics such as :

1. applied examples demonstrating the protocols for elicitation,
2. approaches to combining or conferencing the expert judgements,
3. methods of integrating expert judgements with empirical evidence (e.g., Bayesian inference with informed priors),
4. elicitation of variable quantities; functions, or uncertainty in conclusions from scientific assessments, etc.,
5. elicitation using precise or bounded probabilities, novel approaches to fit probability distributions to expert judgements,
6. methods to extract and use experts’ rationales behind their quantitative assessments.

We also welcome contributions about any other recent developments to strengthen the use and scientific rigour of Expert Knowledge Elicitation when used for assessment.

Join the Swedish Society for Risk Sciences and the Nordic chapter of the Society for Risk Analysis’ upcoming webinar entitled RISKUT2021 on October 21, 2021. 

RISKUT2021 is a half-day online conference about teaching risk. This webinar will offer group discussions on the theme “the risk curriculum.” The aim of the conference is to create a meeting place for sharing ideas and experiences on how to educate about risk. Individuals from academia, the public and private sector and from all levels of education are invited.

The conference language will be English and this event is free of charge. 

More information and how to sign up

https://sites.google.com/view/riskut/home

This year’s version of Bayes@Lund will be a series of online talks.

Welcome to join

Organisers

Dmytro Perepolkin, Ullrika Sahlin, Rasums Bååth

Matthew Kay, Assistant professor Computer Science and Communication Studies Northwester University- September 15th at 17 CET

Uncertainty visualization with tidybayes and ggdist

When communicating statistical results, we must consider carefully how to communicate uncertainty in ways that people can actually use to make informed decisions. I will describe some useful strategies, tools, and techniques for doing so, based on my (and others’) research and my own practice. I will also showcase recent updates to the tidybayes and ggdist R packages designed to make it easy to quickly create a variety of custom uncertainty visualizations from Bayesian models.

Sign up for the talk on SkåneR meetup page

Thank you, Matt for the awesome presentation! The recording of the talk can be seen here

The slides for today’s talk are available at https://www.mjskay.com/presentations/lund2021-uncertainty.pdf

Will Landau, Statistician and software developer, Eli Lilly and Company – October 20th at 16.00 CET

stantargets and Target Markdown for Bayesian model validation pipelines

The targets R package enhances the reproducibility, scale, and maintainability of data science projects in computationally intense fields such as machine learning, Bayesian Statistics, and statistical genomics. Recent breakthroughs in the targets ecosystem make it easy to create ambitious, domain-specific, reproducible data analysis pipelines. Two highlights include stantargets, a new rOpenSci package that generates specialized workflows for Stan models while reducing the required volume of user-side R code, and Target Markdown, an R Markdown interface to transparently communicate the entire process of pipeline construction and prototyping. The example Target Markdown report at https://wlandau.github.io/rmedicine2021-pipeline (source: https://github.com/wlandau/rmedicine2021-pipeline) demonstrates both capabilities in a simulation-based workflow to validate a Bayesian longitudinal linear model common in clinical trial data analysis.

Will’s talk can be found on our Youtube channel link

Aki Vehtari, Associate professor in computational probabilistic modeling at Aalto University –  November 10th

On Bayesian Workflow

Abstract: I discuss some parts of Bayesian workflow with a focus on the need and justification for iterative workflow. The talk is partly based on a review paper (Gelman et al., 2020) with the following abstract: “The Bayesian approach to data analysis provides a powerful way to handle uncertainty in all observations, model parameters, and model structure using probability theory. Probabilistic programming languages make it easier to specify and fit Bayesian models, but this still leaves us with many options regarding constructing, evaluating, and using these models, along with many remaining challenges in computation.  Using Bayesian inference to solve real-world problems requires not only statistical skills, subject matter knowledge, and programming, but also awareness of the decisions made in the process of data analysis. All of these aspects can be understood as part of a tangled workflow of applied Bayesian statistics. Beyond inference, the workflow also includes iterative model building, model checking, validation and troubleshooting of computational problems, model understanding, and model comparison. We review all these aspects of workflow in the context of several examples, keeping in mind that applied research can involve fitting many models for any given problem, even if only a subset of them are relevant once the analysis is over.” The pre-print is available at https://arxiv.org/abs/2011.01808.

About the speaker: My research interests are Bayesian probability theory and methodology, especially Bayesian workflow, diagnostics, probabilistic programming, inference methods such as Laplace, EP, VI, MC, model assessment and selection,  non-parametric models such as Gaussian processes, dynamic models, and hierarchical models. I’m a member of Stan and ArviZ development teams. I’m the co-author of the textbooks Bayesian Data Analysis, 3rd ed, and Regression and Other Stories.

References:

Gelman A, Vehtari A, Simpson D, Margossian CC, Carpenter B, Yao Y, Kennedy L, Gabry J, Bürkner P-C, Modrák M. 2020. Bayesian workflow [Arxiv]. https://arxiv.org/abs/2011.01808

The Video for Aki’s talk has now been published:

https://www.youtube.com/watch?v=lKRRyrPxxeU

The slides have been uploaded to our drive and shared in the video notes:

https://drive.google.com/file/d/1vb8R3iMmSL1tz61B6T6K7ur9Px972qAu/view

Congratulations to Hazel for the successful publication of the perspective “Bayesian Network Applications for Sustainable Holistic Water Resources Management: Modeling Opportunities for South Africa” in the Journal of Risk Analysis. DOI: 10.1111/risa.13798

Hazel Indrani Govender is a PhD student at University of KwaZulu-Natal, Durban, South Africa, that spent 6 months in Lund on an Erasmus Mundus Scolarship 2016-2017.

Bayesian decision analysis is a useful method for risk management decisions, but is limited in its ability to consider severe uncertainty in knowledge, and value ambiguity in management objectives. We study the use of robust Bayesian decision analysis to handle problems where one or both of these issues arise. The robust Bayesian approach models severe uncertainty through bounds on probability distributions, and value ambiguity through bounds on utility functions.

https://onlinelibrary.wiley.com/doi/10.1111/risa.13722