The Central Idea
Ethical risk communication aims to help people understand uncertainty well enough to make informed decisions.
A responsible presentation should make clear:
- what event is being measured;
- what probability or rate is being reported;
- what population or time period it applies to;
- how uncertain the estimate is;
- what assumptions the model uses;
- what consequences are associated with the event.
Risk Is More Than a Number
A probability such as
\[ 0.02 \]is incomplete without context.
We need to know:
- 2% of what?
- over what time period?
- under what conditions?
- estimated from which data?
- with what uncertainty?
State the Event Precisely
A risk statement should define the event being measured.
Compare:
“The risk is 5%.”
with:
“The estimated probability of system failure during the next year under current operating conditions is 5%.”
The second statement is much more informative.
Use a Clear Reference Class
A probability applies to some group of cases or repeated situations.
That group is the reference class.
For example:
\[ 3\text{ failures per }1000\text{ units per year} \]specifies both the population and the time interval.
Natural Frequencies
Probabilities are often easier to understand when expressed as natural frequencies.
For example,
\[ 0.01 \]can be written as
\[ 1\text{ in }100. \]Likewise,
\[ 0.002 \]can be written as
\[ 2\text{ in }1000. \]Use Consistent Denominators
Comparing risks is easier when they use the same denominator.
Compare:
- 1 in 100;
- 8 in 1000.
Convert the first:
\[ 1\text{ in }100 = 10\text{ in }1000. \]Now the comparison is direct:
\[ 10\text{ in }1000 \quad\text{versus}\quad 8\text{ in }1000. \]Absolute Risk
Absolute risk reports the probability or rate of an event in a group.
Suppose an event occurs in
\[ 4\% \]of Group A and
\[ 2\% \]of Group B.
Absolute Risk Difference
The absolute risk difference is
\[ \boxed{ 4\%-2\%=2\text{ percentage points} }. \]This describes the direct difference in event probability between the two groups.
Relative Risk
The relative risk compares two probabilities by ratio.
\[ \boxed{ RR=\frac{p_1}{p_0} }. \]Using \(4\%\) and \(2\%\),
\[ RR=\frac{0.04}{0.02}=2. \]The first risk is twice the second.
Relative Risk Increase
Relative change can be written
\[ \boxed{ \frac{p_1-p_0}{p_0} }. \]For \(4\%\) versus \(2\%\),
\[ \frac{0.04-0.02}{0.02}=1. \]This is a
\[ \boxed{100\%\text{ relative increase}}. \]Absolute and Relative Risk Should Often Be Shown Together
Saying only
“Risk doubled.”
can sound dramatic.
Saying only
“Risk increased by 2 percentage points.”
can sound modest.
Both describe the same change from \(2\%\) to \(4\%\). Presenting both gives a more complete picture.
Small Baselines Can Create Large Relative Changes
Suppose risk rises from
\[ 0.01\% \]to
\[ 0.02\%. \]The relative increase is
\[ 100\%. \]But the absolute increase is only
\[ 0.01\text{ percentage points}. \]Ethical communication should not use relative change alone when the baseline is important to interpretation.
Probability and Odds Are Different
If an event has probability
\[ p, \]its odds are
\[ \boxed{ \frac{p}{1-p} }. \]A probability of \(0.20\) corresponds to odds
\[ \frac{0.20}{0.80} = \frac14. \]That is odds of \(1\) to \(4\), not a probability of \(1/4\).
Do Not Switch Between Odds and Probabilities Without Saying So
Odds ratios can differ numerically from risk ratios.
A technical report may appropriately use odds, but a public-facing explanation should clearly identify the quantity and avoid implying that odds and probability are interchangeable.
Time Horizon Matters
Risk over one day, one year, and one lifetime can be very different.
A statement such as
\[ P(\text{failure})=0.05 \]should identify the relevant period.
Otherwise the number can be badly misinterpreted.
Conditional Risk
Some probabilities apply only under specified conditions.
For example,
\[ P(F\mid A)=0.08 \]is not the same quantity as
\[ P(F)=0.08. \]The conditioning event \(A\) should be stated clearly.
Distinguish Risk from Uncertainty about Risk
A model may estimate
\[ p=0.12. \]But the estimate itself may be uncertain.
Ethical communication separates:
- uncertainty about which outcome will occur;
- uncertainty about the probability assigned to that outcome.
Point Estimates Can Hide Uncertainty
Reporting
\[ \hat p=0.30 \]alone may suggest more precision than the evidence supports.
Where appropriate, uncertainty may be represented using:
- confidence intervals;
- credible intervals;
- ranges;
- scenario bands;
- sensitivity analysis.
Example: Estimate with an Interval
Instead of reporting only
\[ \hat p=0.30, \]a report might state that the estimate is \(0.30\) with an uncertainty interval from \(0.24\) to \(0.36\), if that interval is justified by the statistical method being used.
The interval communicates that the point estimate is not exact.
Do Not Invent Precision
A probability such as
\[ 0.637418 \]may look authoritative.
But if the underlying data and assumptions only justify a rough estimate, reporting six decimal places creates false precision.
Round to Meaningful Precision
The number of reported digits should reflect the quality of the evidence and the needs of the decision.
For many decisions,
\[ 0.64 \]may be more honest and useful than
\[ 0.637418. \]Distinguish Frequency from Severity
A rare event can still matter if its consequence is severe.
Risk communication should often report both:
- how likely the event is;
- what happens if it occurs.
Probability alone does not describe consequence.
Expected Loss Can Help Combine Probability and Consequence
If an event has probability \(p\) and loss \(L\),
\[ \boxed{ E[\text{loss}]=pL }. \]This can be useful, but expected loss should not conceal rare catastrophic outcomes when their severity matters independently.
Expected Values Are Not Guarantees
If expected loss is
\[ \$5{,}000, \]that does not mean the actual loss will be exactly $5,000.
Expected value summarizes a distribution; realized outcomes may differ greatly.
Communicate the Distribution When It Matters
Two alternatives can have the same expected value and very different downside risks.
When variability, tails, or thresholds matter, show more than the mean.
Useful additions may include:
- standard deviation;
- quantiles;
- probability of shortfall;
- worst plausible scenarios;
- tail-loss measures.
Communicate Model Assumptions
A probability model depends on assumptions.
Important assumptions may include:
- independence;
- stationarity;
- representative sampling;
- constant failure rate;
- normality;
- stable behavior over time.
If the conclusion depends strongly on an assumption, that dependence should be visible.
Sensitivity Analysis Supports Honest Communication
A report can show how conclusions change when uncertain inputs vary.
For example:
\[ p\in[0.3,0.5] \]may produce different recommended actions on different sides of a threshold.
Showing this is more informative than presenting only the base-case result.
Communicate Thresholds Explicitly
If the decision changes when
\[ p=0.42, \]and the current estimate is
\[ 0.43, \]the recommendation is fragile.
That fact should not be hidden behind a single categorical recommendation.
Distinguish Data from Model Output
Observed data and modeled predictions are not the same thing.
A report should distinguish:
- measured values;
- estimated parameters;
- forecast values;
- scenario assumptions;
- simulated outcomes.
Do Not Present Scenarios as Forecasts
A scenario asks:
What would happen if these assumptions held?
A forecast asks:
What do we expect to happen?
Confusing the two can exaggerate the evidential status of a hypothetical case.
Visual Communication Can Distort Risk
Graphs and charts can clarify probabilities, but design choices can also mislead.
Risk graphics should avoid:
- truncated axes that exaggerate small differences;
- inconsistent denominators;
- decorative area or volume effects that distort magnitude;
- missing uncertainty ranges;
- selective time windows.
Example: Truncated Axis
Suppose two rates are
\[ 48\% \]and
\[ 52\%. \]A bar chart beginning at \(47\%\) can make the second bar appear many times taller even though the absolute difference is only
\[ 4\text{ percentage points}. \]Use Visual Scales Appropriate to the Question
A zero baseline is often appropriate for bar charts because bar length encodes magnitude.
Line charts may use narrower ranges when the purpose is to show variation, but the scale should be clearly labeled and should not imply a larger effect than the data support.
Icon Arrays and Frequency Grids
For small probabilities, icon arrays can make frequencies concrete.
For example, a risk of
\[ 3\% \]can be represented as
\[ 3\text{ affected cases out of }100. \]This can be easier to interpret than an isolated decimal.
Use Comparable Frames
If one risk is shown as
\[ 2\text{ in }100 \]and another as
\[ 30\text{ in }10{,}000, \]the comparison becomes harder.
Convert both to a common frame where possible.
Positive and Negative Framing
The same outcome can be framed positively or negatively.
For example:
\[ 90\%\text{ survival} \]and
\[ 10\%\text{ mortality} \]describe complementary probabilities.
Because framing can influence perception, important decisions may benefit from presenting both forms.
Do Not Use Framing to Manipulate
Choosing only the formulation that makes an option sound safer, more dangerous, more successful, or more attractive can bias the audience.
Ethical communication aims for understanding rather than persuasion by selective arithmetic.
Base Rates Matter
Diagnostic and classification probabilities can be badly misunderstood when base rates are omitted.
A test can be highly accurate and still produce many false positives when the underlying condition is very rare.
Relevant prior probabilities should therefore be communicated when they materially affect interpretation.
Conditional Probabilities Should Be Named Carefully
These two quantities are generally different:
\[ P(\text{positive test}\mid\text{condition}) \]and
\[ P(\text{condition}\mid\text{positive test}). \]Reversing them is a common and serious communication error.
Communicate Missing Information
Sometimes the correct statement is not a precise number.
It may be:
- the probability is poorly known;
- the data are sparse;
- the model has not been validated in this population;
- the relevant outcome has not been measured;
- several scenarios remain plausible.
Admitting uncertainty is often more informative than presenting a false estimate.
Separate Known, Estimated, and Assumed Quantities
A clear risk report can distinguish three categories:
- Observed: directly measured from data;
- Estimated: inferred statistically;
- Assumed: specified for modeling or scenario analysis.
This makes the evidential status of each number easier to audit.
Communicate Who Bears the Risk
An average risk can hide unequal distribution.
Two policies may have the same average expected loss while assigning losses to different groups.
When distribution matters, communicate:
- who benefits;
- who bears downside risk;
- how concentrated the consequences are;
- whether vulnerable subgroups face different exposure.
Average Risk Can Hide Heterogeneity
Suppose the overall event probability is
\[ 10\%. \]That average could arise from:
- every subgroup having roughly \(10\%\) risk; or
- one subgroup having \(2\%\) risk and another having \(30\%\) risk.
The average alone cannot distinguish these situations.
Communicate Conflicts of Interest
If the person presenting risk has a financial, organizational, or strategic interest in a particular conclusion, that context can matter to the audience.
Transparency about incentives supports more trustworthy interpretation.
Do Not Cherry-Pick Time Periods
Selecting only the time window that supports a desired conclusion can distort apparent risk.
A responsible analysis explains why the chosen period is relevant and, where useful, shows how the result changes under alternate reasonable windows.
Do Not Cherry-Pick Outcomes
Reporting only favorable outcomes while omitting relevant adverse outcomes can mislead even when every reported number is technically correct.
The set of outcomes shown should match the decision being evaluated.
Communicate Both Benefits and Harms
Many decisions involve tradeoffs.
A balanced risk communication may need to show:
- probability of benefit;
- size of benefit;
- probability of harm;
- size of harm;
- uncertainty in both.
Tail Risks Deserve Explicit Treatment
A low-probability catastrophic event can disappear visually or numerically when only averages are reported.
If tail outcomes matter to the decision, communicate them separately rather than assuming the mean fully represents the distribution.
Distinguish Hazard from Risk
A hazard is a source of potential harm.
Risk usually incorporates the likelihood and consequence of harm under some exposure.
A severe hazard may produce low risk when exposure is nearly absent, while a moderate hazard may create substantial risk under frequent exposure.
Audience Matters
The same mathematical result may need different explanations for:
- specialists;
- executives;
- students;
- the public;
- operators making immediate decisions.
Simplification is appropriate when it improves understanding, provided it does not distort the underlying result.
Clarity Is Not the Same as Oversimplification
A clear explanation can use ordinary language while preserving important mathematical distinctions.
For example:
“About 8 in 1000 comparable systems are expected to fail within one year under these assumptions.”
is both accessible and precise.
Give the Audience a Comparison Point
A risk is easier to interpret when compared with:
- a baseline;
- an alternative decision;
- a previous period;
- a relevant benchmark.
Comparisons should be genuinely relevant rather than selected for emotional effect.
Ethical Communication Supports Autonomy
Good risk communication provides enough information for the audience to make an informed choice consistent with its own objectives and preferences.
It should not hide essential tradeoffs merely because the presenter prefers one decision.
Uncertainty Should Not Be Used to Pretend Nothing Is Known
Ethical communication avoids two opposite errors:
- presenting uncertain evidence as certain;
- using the existence of uncertainty to imply that no conclusion is possible.
Most real evidence lies between perfect certainty and total ignorance.
Calibrated Language
Language should match the strength of the evidence.
Words such as:
- certain;
- likely;
- possible;
- unlikely;
- unknown;
should be used consistently and, in technical settings, preferably linked to explicit probability ranges when appropriate.
Communicating Model Updates
New data can change a risk estimate.
When a model is updated, explain:
- what new information was added;
- which assumptions changed;
- how the estimate changed;
- whether the decision recommendation changed.
Correct Errors Transparently
If a probability, graph, or interpretation was wrong, the correction should clearly identify:
- the original error;
- the corrected value;
- the effect on the conclusion.
Quietly replacing a number can make it difficult for decision makers to understand what changed.
Common Error: Reporting Relative Risk Alone
A statement such as
“Risk increased by 200%.”
is incomplete without the baseline.
A change from \(0.1\%\) to \(0.3\%\) and a change from \(10\%\) to \(30\%\) both represent a \(200\%\) relative increase but have very different absolute scales.
Common Error: Reporting a Percentage without a Denominator
A rate such as
\[ 5\% \]should identify what population, event, and time period the percentage describes.
Common Error: Treating a Confidence Interval as a Probability Statement about a Fixed Parameter
In a standard frequentist interpretation, a confidence interval procedure has a long-run coverage property.
It should not automatically be described as the probability that the fixed parameter lies inside this particular realized interval.
Common Error: Hiding Unfavorable Scenarios
If the model supports materially adverse plausible outcomes, removing them from the presentation can make the decision appear safer than the analysis actually shows.
Common Error: Equating Statistical Significance with Practical Importance
A statistically detectable difference may be too small to matter operationally.
Conversely, a practically important difference may remain statistically uncertain when data are limited.
Communicate both statistical evidence and decision relevance.
Common Error: Using a Single Summary for Everyone
Different audiences may need different levels of technical detail.
A good communication can provide a clear headline result while preserving access to assumptions, data, uncertainty, and methodology.
A Reliable Procedure for Communicating Risk
- Define the event precisely.
- State the relevant population and time horizon.
- Report absolute probability or frequency.
- Use a consistent denominator for comparisons.
- Report relative changes together with absolute changes when both matter.
- Describe the consequence as well as the likelihood.
- Show important uncertainty in the estimate.
- State material model assumptions and limitations.
- Use visual scales that do not exaggerate or hide differences.
- Separate observed data, estimates, forecasts, and scenarios.
- Identify who bears the benefits and risks when distribution matters.
- Use language appropriate to the evidence and audience.
A Reliable Procedure for Reviewing a Risk Claim
- Ask what event the number describes.
- Identify the denominator.
- Identify the time period.
- Check whether the number is absolute risk, relative risk, odds, or another quantity.
- Look for the baseline used in comparisons.
- Check whether uncertainty is shown.
- Check whether important outcomes or subgroups are omitted.
- Inspect graphs for scale or framing distortions.
- Identify assumptions behind the model.
- Ask whether the communication supports understanding or merely persuasion.
Practice 1: Absolute Risk Difference
Risk rises from \(3\%\) to \(5\%\).
Find the absolute risk difference.
Practice 2: Relative Risk
Using \(3\%\) and \(5\%\), find the relative risk
\[ \frac{0.05}{0.03}. \]Practice 3: Relative Increase
Risk rises from \(2\%\) to \(3\%\).
Find the relative increase.
Practice 4: Natural Frequency
Express
\[ 0.007 \]as a frequency per \(1000\).
Practice 5: Compare with a Common Denominator
Compare \(1\) in \(200\) with \(7\) in \(1000\) using a denominator of \(1000\).
Practice 6: Probability versus Odds
An event has probability
\[ p=0.25. \]Find its odds.
Practice 7: False Precision
A model is based on sparse data and highly uncertain assumptions, but the result is reported as
\[ 0.483726. \]What communication problem should be considered?
Practice 8: Framing
A system succeeds \(92\%\) of the time.
State the complementary failure framing.
Practice 9: Scenario or Forecast?
An analyst assumes demand falls \(20\%\), costs rise \(15\%\), and production capacity falls \(10\%\), then calculates the resulting loss.
Is this necessarily a forecast?
Practice 10: Relative versus Absolute Change
Risk increases from \(0.1\%\) to \(0.2\%\).
Find:
- the absolute increase in percentage points;
- the relative increase.
Practice Answers
Practice 1
\[ 5\%-3\% = \boxed{2\text{ percentage points}}. \]Practice 2
\[ RR = \frac{0.05}{0.03} \] \[ \boxed{\approx1.67}. \]The \(5\%\) risk is about \(1.67\) times the \(3\%\) risk.
Practice 3
\[ \frac{0.03-0.02}{0.02} = \frac{0.01}{0.02} = 0.5. \]Therefore,
\[ \boxed{50\%\text{ relative increase}}. \]Practice 4
\[ 0.007(1000)=7. \]Therefore,
\[ \boxed{7\text{ in }1000}. \]Practice 5
\[ 1\text{ in }200 = 5\text{ in }1000. \]Therefore the comparison is
\[ \boxed{5\text{ in }1000\text{ versus }7\text{ in }1000}. \]Practice 6
\[ \frac{0.25}{1-0.25} = \frac{0.25}{0.75} = \frac13. \]The odds are
\[ \boxed{1\text{ to }3}. \]Practice 7
The main concern is
\[ \boxed{\text{false precision}}. \]The number of digits may imply more certainty than the model supports.
Practice 8
\[ 100\%-92\%=8\%. \]Therefore the complementary framing is
\[ \boxed{8\%\text{ failure}}. \]Practice 9
No. This is a scenario calculation unless the analyst is claiming those assumptions are expected to occur.
Practice 10
Absolute increase:
\[ 0.2\%-0.1\% = \boxed{0.1\text{ percentage points}}. \]Relative increase:
\[ \frac{0.2\%-0.1\%}{0.1\%}=1. \] \[ \boxed{100\%\text{ relative increase}}. \]What to Remember
Ethical risk communication preserves the mathematics while making the decision context understandable.
- Define the event, population, and time horizon.
- Use natural frequencies and common denominators when they improve comprehension.
- Distinguish absolute risk from relative risk.
- Distinguish probabilities from odds.
- Show uncertainty when the estimate itself is uncertain.
- Avoid false precision.
- Communicate both likelihood and consequence.
- State important model assumptions and limitations.
- Use graphs and framing that do not exaggerate or conceal differences.
- Separate observations, estimates, forecasts, and scenarios.
- Communicate who bears the risk when consequences are unevenly distributed.
- Support informed understanding rather than selective persuasion.
The central ethical principle is: present risk in a way that helps the audience understand the evidence, the uncertainty, and the consequences without manipulating the decision through selective framing.
Section Wrap-Up
This completes 4.18 Decision, Risk, and Information.
Across the section, we moved from basic decision tables and decision trees to expected value, expected utility, loss functions, risk and uncertainty, sensitivity analysis, value of information, entropy, coding, reliability, strategic games, portfolio allocation, and finally the responsible communication of uncertain outcomes.
Together, these tools provide a broad mathematical framework for making, evaluating, improving, and explaining decisions when outcomes are uncertain and information is limited.