David
Spiegelhalter
Winton Professor of the Public Understanding of Risk
Centre for Mathematical Sciences, Cambridge
Statisticians try to face up to uncertainty, but
this term has many subtle shades of meaning. The classical
paradigm deals with variability in observable random
quantities, while a Bayesian approach extends our
range to include formal expressions of epistemic uncertainty
about unknown states of the world. More controversial
is the use of probability statements that measure
what we believe about how the world works, for example
in climate change modelling. It has been argued, however,
that all such quantitative approaches are rather restrictive
and that more informal methods are needed to deal
with the much deeper uncertainties and ambiguities
in human affairs. I shall attempt to examine how well
statistical methods deal with all these demands, with
special emphasis on the Bayesian paradigm.
Honest communication of uncertainty seems an essential
part of any statistical project. We can construct
risk estimates, interval estimates for unknown quantities,
various measures of evidence for and against hypotheses,
and so on, but the way in which these are communicated
can strongly influence the perception of the consumers
of the analysis. I shall look at the different forms
of text, numbers, and graphics that have been used
in a variety of contexts to communicate uncertainty,
whether to individual members of the public or those
with responsibility for policy. I will suggest that
the current possibility for interactive animations
provides a fine opportunity for a more flexible and
multilayered approach, so there will be a lot of
pictures, many of them moving.
