NPI models explained and complained
(2021) In ISIF Perspectives on information fusion 4(1). p.7-14- Abstract
- Numerous modelling efforts have attempted to characterize the effects of different non-pharmaceutical interventions (NPIs) on the Covid-19 spread. Arguably the most famous is one published in Nature by an Imperial College group. A slight variation of it was later published in Science by a group of Oxford researchers. Both publications are based on hierarchical Bayesian modelling that aims to explain observed data by information on enacted NPIs. Due to the Bayesian approach, the models become quite complex and opaque, with many priors that have been assigned more or less ad hoc, and there are even priors on the prior parameters. We show how these models can be recast into the classic linear regression framework. This enables us to... (More)
- Numerous modelling efforts have attempted to characterize the effects of different non-pharmaceutical interventions (NPIs) on the Covid-19 spread. Arguably the most famous is one published in Nature by an Imperial College group. A slight variation of it was later published in Science by a group of Oxford researchers. Both publications are based on hierarchical Bayesian modelling that aims to explain observed data by information on enacted NPIs. Due to the Bayesian approach, the models become quite complex and opaque, with many priors that have been assigned more or less ad hoc, and there are even priors on the prior parameters. We show how these models can be recast into the classic linear regression framework. This enables us to transparently analyze basic concepts such as persistency of excitation, identifiability, and model sensitivity. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/3cd63d10-f80e-492c-9bd8-74d3eaba8298
- author
- Gustafsson, Fredrik
and Soltesz, Kristian
LU
- organization
- publishing date
- 2021
- type
- Contribution to specialist publication or newspaper
- publication status
- published
- subject
- in
- ISIF Perspectives on information fusion
- volume
- 4
- issue
- 1
- pages
- 7 - 14
- publisher
- International Society of Information Fusion (ISIF)
- ISSN
- 2409-4846
- project
- COVID-19: Dynamical modelling for estimation and prediction
- language
- English
- LU publication?
- yes
- id
- 3cd63d10-f80e-492c-9bd8-74d3eaba8298
- alternative location
- https://confcats_isif.s3.amazonaws.com/web-files/perspectives/ipif-04-01-COMBINED_SMALLER.pdf
- date added to LUP
- 2021-11-18 06:05:40
- date last changed
- 2021-11-22 08:12:11
@misc{3cd63d10-f80e-492c-9bd8-74d3eaba8298, abstract = {{Numerous modelling efforts have attempted to characterize the effects of different non-pharmaceutical interventions (NPIs) on the Covid-19 spread. Arguably the most famous is one published in Nature by an Imperial College group. A slight variation of it was later published in Science by a group of Oxford researchers. Both publications are based on hierarchical Bayesian modelling that aims to explain observed data by information on enacted NPIs. Due to the Bayesian approach, the models become quite complex and opaque, with many priors that have been assigned more or less ad hoc, and there are even priors on the prior parameters. We show how these models can be recast into the classic linear regression framework. This enables us to transparently analyze basic concepts such as persistency of excitation, identifiability, and model sensitivity.}}, author = {{Gustafsson, Fredrik and Soltesz, Kristian}}, issn = {{2409-4846}}, language = {{eng}}, number = {{1}}, pages = {{7--14}}, publisher = {{International Society of Information Fusion (ISIF)}}, series = {{ISIF Perspectives on information fusion}}, title = {{NPI models explained and complained}}, url = {{https://lup.lub.lu.se/search/files/109864886/soltesz21j.pdf}}, volume = {{4}}, year = {{2021}}, }