Prescriptive analysis

Risk Communication:
An Introduction
Baruch Fischhoff
University Professor
Carnegie Mellon
Department of Social & Decision Sciences
Department of Engineering & Public Policy
Adapted for the SuperCourse from a talk to
International Society for Pharmacoepidemiology”
April 19-20, 2004
Overview
Communication research strategies
Two case studies
medical informed consent
public health information management
Six brief examples
Communication research management
Pressure for Lay Participation
in Decisions Involving Risks
health
finance
environment
national security
…
Poor Execution Can…
undermine effective lay decision
making create feelings of
helplessness erode public faith in
authorities erode authorities’ faith in
the public erode the social
coordination produced by
sharing trusted information
sources
Research Strategy Choices
Begin with formal analysis?
Persuasive or non-persuasive?
Optimality or mastery?
Which individual differences?
Research Strategy Choices
Begin with formal analysis?
Persuasive or non-persuasive?
Optimality or mastery?
Which individual differences?
Also: Do you consider affect? social context?
resources? …(potentially important, but
outside present talk)
Behavioral Decision Research
Normative analysis of decision situation
Descriptive behavioral research, in terms
comparable to normative analysis
Prescriptive interventions, closing gaps
between normative ideal and
descriptive reality
Behavioral Decision Research
Strategy Choices
Begin with formal analysis? yes
Persuasive or non-persuasive? Non
(default)
Optimality or mastery? either
Which individual differences?
circumstances, values,
competence
Three Classes of Information
Quantitative
How big are the risks - and benefits?
Qualitative
What determines risks - and benefits?
Communication process
What is social context of message?
Uses of Quantitative Information
calculating expected utility
assessing the value of new information
setting priorities
applying a threshold for concern
Uses of Qualitative
Information
making quantitative information credible
reconciling competing claims
generating options
monitoring environment for changes
feeling realistically empowered
Uses of Process Information
establishing source credibility
(competence, honesty)
assessing source expectations
(audience competence, locus of
control)
identifying conversational context
(interpretation, reading between
the lines)
A Quantitative Information
Example:
Carotid Endarterechtomy
Surgery can extend life, but lots can
go wrong.
Patients face flood of information at
a stressful time.
Physicians face duty to inform.
(with Jon Merz, Paul Fischbeck, Dennis Mazur)
Normative Analysis
Value-of-information analysis, using
risk analyses of patient - relevant
outcomes, formalizing the materiality
standard
Many Possible Side Effects
death
stroke
facial paralysis
myocardial infarction
lung damage
headache
resurgery
tracheostomy
gastrointestinal upset
broken teeth
…….
But knowledge of only a few would
affect many patients’ choices
death
stroke
facial paralysis
myocardial infarction
lung damage
headache
resurgery
tracheostomy
gastrointestinal upset
broken teeth
15.0%
5.0
3.0
1.1
0.9
0.8
0.4
0.2
0.09
0.01
(% that would decline, if they knew of each risk)
Descriptive Research
(anticipated)
Patients don’t know probabilities of top
risks: death, stroke, facial nerve paralysis
Patients don’t know event meaning for
facial paralysis
Prescriptive Design Issues (1)
Identify Communication Challenges
verbal quantifiers (e.g., “likely” threat)
experientially unfamiliar events
value uncertainty (what do I really want?)
but not
very low probabilities
cumulative risk (from repeated exposure)
Prescriptive Design Issues (2)
Choose Performance Measures
knowledge
inferential ability
appropriateness of confidence
appropriateness of self-efficacy
personally rational choices
satisfaction (?)
Conclusions
Systematically using clinical trial
results can: drastically simplify task
(for patient & physician)
formalize materiality standard
set research priorities
The adequacy of non-persuasive
communication
is plausible, but not proven
Research Strategy
Begin with formal analysis?
yes
Persuasive or non-persuasive? non
Optimality or mastery?
ptimality
Which individual differences?
health status, values
(assume competence)
A Qualitative Information
(and Communication Process)
Example: Emergency Notification
for Waterborne Pathogens
Cryptosporidium intrusion in domestic
water supplies. Special vulnerability
for immunocompromised individuals.
(Liz Casman, Felicia Wu, Claire Palmgren, Mitch
Small, Joan Rose, Hadi Dowlatabadi)
Normative Analysis
Behaviorally realistic model of
system performance, including
detection, organizational
coordination, and consumer
response
Descriptive Research
“Mental models” interviews
(open-ended, structured by model)
with:
HIV+ individuals:
knew a lot, could do little
residents of communities with past
intrusions:
knew little, mildly curious
Conclusions
Abandon emergency communication
for crypto
(perhaps OK for other
contaminants, E. coli, bioterror)
Provide bottled water to those at risk
Reevaluate communications for other
immunocompromised individuals
Research Strategy
Begin with formal analysis? yes
Persuasive or non-persuasive?
persuasive
Optimality or mastery? mastery
Which individual differences?
health status, values, competence
Six Brief Examples
Sexual assault prevention advice
Teen STI risk
Vaccines (anthrax, MMR)
Dietary supplements
Hazardous chemicals
Homeland security
Sexual Assault
Problem: confident, universal, contradictory
advice fosters confusion and guilt
Normative: inventory of strategies, metaanalysis of effectiveness studies,
identification of values
Descriptive: nuanced belief structure,
exaggerated effectiveness
Prescriptive: create realistic expectations,
promote research evaluating effectiveness
STIs (teens)
Problem: flood of repetitious information
creates illusion of understanding.
Normative: influence diagram summarizing
decision-relevant science
Descriptive: broad knowledge of HIV/AIDS,
with critical “bugs”; little knowledge of other
STIs, limited feeling of control
Prescriptive: targeted HIV/AIDS messages;
DVD on sexual decision making
Vaccination (anthrax, MMR)
Problem: mistrust of officials affects trust
in medicine
Normative: comprehensive model of
factors determining health, trust
Descriptive: critical issues outside official
problem space, terminological confusion
Prescriptive: provide context, including
“irrelevant concerns”; research as
communication
Dietary supplements
Problem: commercial-freedom-of-speech
policies expanding direct-to-consumer
communication
Normative analysis: model sensitive to
burden of responsibility for consumer/market
failure
Descriptive analysis: court-mandated labels
may erode consumer rationality
Prescriptive analysis: standardized,
validated labels; legal standard for adequacy
Hazardous Chemicals
Problem: Can voluntary controls
substitute for regulatory controls?
Normative: diffusion/uptake model,
including effects of behavior
Descriptive: ineffective measures
intuitively appealing (e.g., gloves, inroom breaks)
Prescriptive: standardized, validated
labels; legal standard for adequacy
Homeland Security
Problem: Plans without behavioral input
Normative: Behaviorally realistic emergency
response model
Descriptive: Simple facts not transmitted;
unfounded belief in panic; specific fears
decline, general anxiety stable
Prescriptive: Communication to experts,
interdisciplinary teams, preemptive messages
Also…
BSE
SARS
HRT
GMOs
Xenotransplantation
smallpox (vaccination)
…
General Conclusions
Effective risk communication requires
continuing collaboration among
domain specialists, risk analysts, risk
communicators, practitioners, and
patients (or their representatives) -with each willing to have its
assumptions challenged.
Why doesn’t everyone adopt
behavioral decision research?
Possible Professional
Reasons
need for specialization
separation of research worlds
isolation of researchers from
practitioners
predisposition to persuasion
predisposition to sweeping
generalizations
…
Possible Management Reasons
focus on single outcome
narrow reviewing of proposals and
products
favor method application over
development
mandate to persuade
difficulty of evaluating non-persuasive
communications
Persuasive communication
can inappropriately
undermine credibility of public and
experts
alienate audience
overlook critical audience goals &
constraints
impose bureaucratic priorities
Non-persuasive communication
can inappropriately
assume nonexistent abilities
deny people needed protections
muddle value issues
leave wrong priorities unchallenged
Non-persuasive communication
is needed when:
No single, clear optimal choice
multiple values, situations
large uncertainties
No clear advisory role
experts distrusted
shared decision-making
option desired
The Empirical Question
How competent are lay people to
participate?
Many Strong Opinions,
Drawn from
assertion
selected research evidence
anecdotal observation
small, unrepresentative
samples
speculative interpretation
ambiguity regarding decision
frame
Common Rhetorical Answers
Political
predisposition:
Behavioral assumption:
“liberal”
“conservative”
hyper-rational
public
popular
democracy
free markets
paternalistic
regulation
technocratic
control
irrational
public
The Empirical Question
(reformulated)
How competent are lay people to
participate?
-- when facing specific decisions
-- and given half a chance
A (Complex) Working
Hypothesis
People often do sensible things if:
They get relevant information in a
concise, credible form with adequate
context, and without distractions
They have control over their
environment and are judged by their
own goals
They have some minimal decisionmaking competence
Collaborators
Lita Furby, Marcia Morgan, Julie Downs,
Wändi Bruine de Bruin, Sara Eggers, Matt
Dombroski, Donna Riley, Mitch Small, Steve
Risotto, Dan Kovacs, Paul Fischbeck, Claire
Palmgren, Robyn Dawes, Rebecca Parkin,
Gordon Butte, Sarah Thorne, Martha Embrey,
Jenn Lerner, Roxana Gonzalez, Deb Small,
Gabriel Silverman, Lisa Schwartz, Steve
Woloshin…
Sources
Eggers, S.L., & Fischhoff, B. (in press). A defensible
claim? Behaviorally realistic evaluation
standards. Journal of Public Policy and
Marketing.
Fischhoff, B. (1992). Giving advice: Decision theory
perspectives on sexual assault. American
Psychologist, 47, 577-588.
Fischhoff, B. (1994). What forecasts (seem to)
mean. Intl Journal of Forecasting, 10, 387-403.
Fischhoff, B. (1995). Risk perception and
communication unplugged: Twenty years of
process. Risk Analysis, 15, 137-145.
Fischhoff, B. (1999). Why (cancer) risk
communication can be hard. Journal of the
National Cancer Institute Monographs, 25, 7-13.
Fischhoff, B. (2000). Scientific management of