Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Scientific methods for integrating expert knowledge in Bayesian models

Perepolkin, Dmytro LU orcid (2023)
Abstract
Generating scientific advice to environmental management involves assessments with complex models, sparse data, and challenging empirical experiments, necessitating the integration of expert judgment with data into scientific models. To integrate expert judgement, assessors might elicit judgement by experts as quantiles, find a probability distribution that matches the quantiles, and add this information to the model. Data is then integrated into the model by Bayesian inference to learn parameters or make predictions. This thesis aims to simplify such
integration of expert judgment, and introduce the use of Quantile-Parameterized Distributions (QPDs) into Bayesian models. Key questions addressed include identifying suitable QPDs for... (More)
Generating scientific advice to environmental management involves assessments with complex models, sparse data, and challenging empirical experiments, necessitating the integration of expert judgment with data into scientific models. To integrate expert judgement, assessors might elicit judgement by experts as quantiles, find a probability distribution that matches the quantiles, and add this information to the model. Data is then integrated into the model by Bayesian inference to learn parameters or make predictions. This thesis aims to simplify such
integration of expert judgment, and introduce the use of Quantile-Parameterized Distributions (QPDs) into Bayesian models. Key questions addressed include identifying suitable QPDs for encoding expert judgment, and conditions for using QPDs as priors or likelihoods in Bayesian inference. The creation of new QPDs through quantile function transformation is explored, providing a methodological advancement. The use of the proposed methodology is demonstrated on expert-informed bias-adjustment of citizen science data in a Species Distribution
Model for conservation assessment. (Less)
Please use this url to cite or link to this publication:
author
supervisor
opponent
  • Professor Quigley, John, University of Strathclyde
organization
publishing date
type
Thesis
publication status
published
subject
keywords
Bayesian inference, Expert judgement, Quantile-parameterized distributions, Quantile functions
pages
131 pages
publisher
Lund University
defense location
Blå hallen, Ekologihuset.
defense date
2024-01-23 13:00:00
ISBN
978-91-8039-914-2
978-91-8039-915-9
project
Expert Knowledge
language
English
LU publication?
yes
id
f4bf6ffc-b476-4289-8aee-42a6b5d74c00
date added to LUP
2023-12-12 12:24:28
date last changed
2024-03-13 12:57:02
@phdthesis{f4bf6ffc-b476-4289-8aee-42a6b5d74c00,
  abstract     = {{Generating scientific advice to environmental management involves assessments with complex models, sparse data, and challenging empirical experiments, necessitating the integration of expert judgment with data into scientific models. To integrate expert judgement, assessors might elicit judgement by experts as quantiles, find a probability distribution that matches the quantiles, and add this information to the model. Data is then integrated into the model by Bayesian inference to learn parameters or make predictions. This thesis aims to simplify such<br/>integration of expert judgment, and introduce the use of Quantile-Parameterized Distributions (QPDs) into Bayesian models. Key questions addressed include identifying suitable QPDs for encoding expert judgment, and conditions for using QPDs as priors or likelihoods in Bayesian inference. The creation of new QPDs through quantile function transformation is explored, providing a methodological advancement. The use of the proposed methodology is demonstrated on expert-informed bias-adjustment of citizen science data in a Species Distribution<br/>Model for conservation assessment.}},
  author       = {{Perepolkin, Dmytro}},
  isbn         = {{978-91-8039-914-2}},
  keywords     = {{Bayesian inference; Expert judgement; Quantile-parameterized distributions; Quantile functions}},
  language     = {{eng}},
  month        = {{12}},
  publisher    = {{Lund University}},
  school       = {{Lund University}},
  title        = {{Scientific methods for integrating expert knowledge in Bayesian models}},
  url          = {{https://lup.lub.lu.se/search/files/166665609/DP_Thesis-5-unsigned.pdf}},
  year         = {{2023}},
}