330 likes | 514 Views
When building risk models, we can acknowledge uncertainty at different levels: . Specific future eventsQuantities/parameters in a model Assumptions underlying the best' model (both internal and external)Inadequacies of our best' model. . . But what about unacknowledged uncertainties?. . Cl
E N D
1. Risk and uncertainty: a range of approaches David Spiegelhalter
winton professor of the public understanding of risk,
university of cambridge
SAPPUR, Bristol 2009
With thanks to Mike Pearson, Ian Short , Hauke Riesch, Owen Smith, Arciris Garay, etc etc
2. When building risk models, we can acknowledge uncertainty at different levels: Specific future events
Quantities/parameters in a model
Assumptions underlying the ‘best’ model (both internal and external)
Inadequacies of our ‘best’ model
3. But what about unacknowledged uncertainties?
4. Classify into uncertainty about : Specific future events
6. Text? Numbers? Graphics? Animations?
We may be concerned with what people –
like
understand / can reproduce
are influenced by
But
These are not necessarily the same formats!
Formats are influential (framing)
People vary hugely in their preferences and understanding
7. Benefits of Tamiflu
9. Classify into uncertainty about: Specific future events
Quantities in a model
Missing data
Fixed or random effects
Parameters (including systematic biases)
10. UKPDS risk engine
11. Problems with confidence limits Suggestion that all points in interval are equally likely
Media can report “up to X people might have Hepatitis C”
13. Express as a probability distribution, such as Bank of England fan charts for GDP
14. Bank of England ‘fan charts’
15. Hepatitis C prevalence in UK
16. Classify into uncertainty about : Specific future events
Quantities in a model
Assumptions underlying the ‘best’ model (both internal and external)
Which model selection criterion: AIC, BIC, DIC etc etc?
Can we put weights on models?
Does it make sense to talk of probabilities of models?
17. IPCC projections
18. Some epistemic uncertainties about swine flu Infectiousness (R0)
Severity (‘case fatality ratio’)
Risk of recombination with other flu viruses
Pattern of re-emergence
etc
19. Government response? Standard approach in face of major epistemic uncertainties
Play it safe
Worst case planning scenarios
65,000 deaths now down to 19,000 (3rd Sept)
Based on 30% clinical cases, 1/1000 die
20. NICE Complex cost-effectiveness models used to help decide which treatments NHS should fund
Only interested in mean response
Prior distributions to express parameter uncertainty
Alternative competing models
Informal acknowledgement of model inadequacy
21. Classify into uncertainty about : Specific future events
Quantities in a model
The structure of the ‘best’ model
(Recognised) inadequacies of our ‘best’ model
It’s only a ‘guide-book’, not the truth!
Doubt about many assumptions
But can we quantify these limitations?
22. ‘extra-model’ uncertainties So far examined 4 levels of (potentially) quantifiable ‘intra-model’ uncertainties
What about ‘unquantifiable’ extra-model sources?
‘unknown unknowns’: possibilities that have not been thought of
unrecognised major scientific error
unacknowledged cultural assumptions
ambiguities in meaning
unrecognised implicit value judgements as to what is ‘important’
‘indeterminacy’ – human element beyond modelling (Wynne)
Not a clear division with acknowledged inadequacies
Is this the responsibility of the modeller or risk manager?
23. Some responses to ‘deeper’ uncertainties Frank Knight
Donald Rumsfeld (+ Zizek)
Oliver Cromwell
Brian Wynne
Ulrich Beck / Anthony Giddens
RAND
Renn
Etc etc [almost all concerned with environmental risk]
Risk assessment to risk management
24. Frank Knight (1885-1972) Risk, Uncertainty, and Profit (1921)
The essential fact is that 'risk' means in some cases a quantity susceptible of measurement, while at other times it is something distinctly not of this character; and there are far-reaching and crucial differences in the bearings of the phenomena depending on which of the two is really present and operating....
It will appear that a measurable uncertainty, or 'risk' proper, as we shall use the term, is so far different from an unmeasurable one that it is not in effect an uncertainty at all
25. Memorable quote #325
27. Brian Wynne (1992) Risk
know the odds
Uncertainty
parameter/structure
Ignorance
‘don’t know what we don’t know’, scientific errors
Indeterminacy
Causal chains open due to human element
Emphasises that modelling conclusions are contingent upon truth of assumptions
28. Ulrich Beck / Anthony Giddens Risk Society ‘where we increasingly live on a high technological frontier which no one completely understands and which generates a diversity of possible futures
‘Manufactured risk’ is risk created by the very progression of human development .. We often don’t really know what the risks are, let alone how to calculate them accurately in terms of probability tables
‘organised irresponsibility’ – ‘a diversity of humanly created risks for which people and organisations are certainly ‘responsible’ in a sense that they are its authors but where no one is held specifically accountable.
29. RAND (Bob Lempert etc) ‘Robust decision-making’
Neither optimal nor precautionary
Trades some optimal performance for less sensitivity to assumptions
Satisficing over a wide range of futures
Keeping options open
30. Ortwin Renn (2004)
32. Damocles – dams, nuclearCyclops – earthquakes, volcanoesPythia – sudden catastrophic climate events Pandora – unintended man-made effectCassandra – climate changeMedusa - EMR
33. ‘Conclusions’ Statisticians/modellerd tend to have (or at least are taught) a rather narrow view of uncertainty
Different communities approach the hierarchy of uncertainty from opposite ends
Impact of different sources of uncertainty needs to be clearly communicated
Robust use of quantitative methods, with due humility, is of huge value