Personal paper: Risk language and dialectsBMJ 1997; 315 doi: https://doi.org/10.1136/bmj.315.7113.939 (Published 11 October 1997) Cite this as: BMJ 1997;315:939
- Correspondence to: Dr Royston
- Accepted 20 May 1997
For something which matters so much to us all and is such an important consideration in medicine it is odd that we have no common language for discussing the hazards of life.1 2 An earlier article contained some suggestions for clarifying our language for describing risk.3 This paper extends those ideas setting out several ways in which the magnitude of risks might more easily be presented understood and discussed.
Risk, or the chances that a hazard will give rise to harm,4 is generally couched in terms of numerical odds or probabilities (see table 1) yet research has shown that people find it difficult to digest such measures.5 One difficulty is that the range of risks is so wide—from, say, the greater than 1 in 10 risk that cancer will be our eventual cause of death to the less than 1 in 10 million chance per year of being killed by lightning. We all find it hard to grasp such extremes.
Better ways are required for presenting risk magnitudes in a digestible form, and a logarithmic scale provides a basis for a common language for describing a wide range of risks
Various “dialects” of this language—visual, analogue, and verbal scales—could help with grasping different risk magnitudes
Combining the above ideas with the idea of anchoring risk magnitudes to the classification by size of human communities produces a “community risk scale”
Factors other than magnitude are important in considering risk, but an appreciation of magnitude is a crucial first step
The proposed risk scales need to be tested to see if and how they improve people's ability to understand and communicate about risks
A logarithmic scale for risk
Risk is not the only area that presents a wide range of size. Other examples include earthquakes, sound, and acidity-basicity. In all these the range is spanned by using a logarithmic scale—the Richter scale for earthquakes, the decibel scale for sound, and the pH scale for acidity and basicity. It is noteworthy that human responses to many sensory stimuli follow a non-linear relation between perceived and actual magnitude,6 and something similar if more complex seems to be true for perception of both the magnitude and the importance of risks.7 8 9
It has been suggested that risk (or its opposite, safety) should be measured on a logarithmic scale (table 2).10 11 12 A safety scale can be easily turned into a risk scale by subtraction of the magnitudes from 10 (see table 3). A justification for adopting a 0-10 scale is given later.
Measurements of risk are often accurate only to within an order of magnitude, so an integer log scale is sufficient; indeed it can help avoid spurious precision. Where the data allow, however, the basic risk scale could clearly be augmented with finer detail. Decimal points could be added—for example, the risk of death per year from cancer would be about magnitude 7.5 and that from influenza about 6.3.
Here a logarithmic scale is taken to provide the basis for a common language of risk. The rest of the paper is about some possible dialects of this language.
A visual scale for risk
A logarithmic numerical scale helps in the presentation of risks of different magnitudes but it may not help in appreciating just how different these magnitudes are. A visual illustration such as that given in figure 1 can often help. A risk scale spanning more magnitudes could be shown in this manner. This could be achieved without having to resort to multidimensional “hypercubes” by using the final large cube as the “starting” cube for the next three risk magnitudes, and so on. However, this is probably an overly complex approach; the next section describes a simpler method.
A distance analogue scale for risk
An alternative to direct visualisation of risk magnitudes would be to use analogy. One possible analogue scale would be based on distance. For this, the certain occurrence of an adverse event could be represented by a marked stick one metre long. A risk of 1 in 10 could then be represented by the chance of finding such a stick by selecting one at random from a line of one metre sticks stretching for 10 metres, a risk of 1 in 100 by the chance of similarly finding the stick from a line stretching a distance of 100 metres, and so on. Table 4 presents such a distance analogue risk scale.Thus in thinking about a 1 in 1000 risk you would have to imagine searching for a one metre “risk stick” over a distance of a kilometre, for a one in a million risk you would have to consider the task of searching for it from London to John O'Groats, and for a one in a billion risk you would have to imagine searching 25 times round the earth's equator or more than all the way to the moon and back.
A verbal scale for risk up arrow
The log scale and its visual and distance analogue expressions help provide a language of risk but these are all essentially mathematical constructs. A risk classification based on translating risk probabilities from numbers into words (see table 5) has been suggested.6 If the difficulties of getting general agreement about what such words mean could be surmounted13 such a scale could be of considerable help in communicating risk.
A community risk scale
The approach of scaling risk logarithmically and using verbal descriptors could be further developed. The scale could be anchored to something in everyday life which shows large variations in size but can nevertheless be discussed quite easily. We are all interested in what risks mean for us, our families, and our communities. A natural anchor for a risk scale might therefore be provided by the classification by size of human communities. We group communities in roughly logarithmic clusters, from a street of around 100 through a small town of 10 000 to a large country of 100 000 000. Table 6 shows a complete classification of this type. It is, of course, only an approximation as families vary in size; cities often have around a million inhabitants but can be much bigger or smaller and world population is not 10 billion—yet. Nevertheless, it arguably provides a “rule of thumb,” and for our purposes this should be enough.
Such a classification should be useful in thinking about risk because it allows risk to be expressed in terms of “you would expect this to happen to around one person in a street, or one in a town, or one in the whole country.” Of course the nature of the risk (for example, death or injury), the population being considered (for example, everyone or only those participating in a given hazardous activity), and the time period over which risk is being measured (for example, a lifetime or a year) would always need to be made clear. The first two of these are straightforward enough but the third is sometimes a source of confusion. For instance, the risk of death in a year from regular cigarette smoking is about 1 in 200 (a “one per street” risk); the lifetime risk, however, is nearer one in four (a risk at “one per family” level).
Putting these ideas together yields a community risk scale as illustrated in table 7. The risk magnitudes are now anchored via the community cluster classification. The verbal risk scale—“one per street,” “one per town,” “one per country”—has its numeric equivalent based on the underlying probabilities. Drawing on the scale in table 3 we see that a risk of one per person (that is, certainty that it will happen to everybody) would score 10 and a risk of 1 per 10 billion people (the level at which it would be unlikely that even one person anywhere in the world would be affected) would score 0. (It seems not unreasonable to set the zero of a risk scale at the level at which nobody on the planet is likely to be affected. If necessary the scale could still cater for even smaller risks, by using negative magnitudes; which seems rather appropriate for risks which are astronomically small.) “Normal” risks would score in the range 9 to 5; anything lower would be most unlikely to affect anybody in your locality. The community risk scale shows, for example, that in any year in Britain you can expect that around one person in your street will die, that one person in your nearest large town will be murdered, and that one person in a whole region will be killed by lightning.
Various ways have been suggested for presenting risk magnitudes using visual, analogue, and verbal scales. These could be anchored to the way in which human communities are clustered by size, which also provides an empirical justification for using a 0-10 risk scale. The various presentational approaches amount to dialects in the language of risk.
These approaches are not meant to be mutually exclusive; a risk situation might be clarified by using several in combination. It might be helpful, for instance, to include a paragraph along the lines of the illustrations in the box in a statement about some new or reassessed risk.
Examples of use of risk language and dialects
On the best evidence currently available the chance of someone being affected during a year by this hazard is 1 in 100. This is magnitude 8 on a 0-10 risk scale. This level of risk is analogous to the chance of an individual being selected at random out of a line of people standing one metre apart stretching for 100 metres. In community terms it means that during one year you could expect to find about one person affected in every street. Many people would judge this level of risk to be moderately high compared with other risks of normal living
On the best evidence currently available the chance of someone being affected by this hazard is one in 1000. This is magnitude 7 on the 0-10 risk scale. This level of risk is analogous to the chance of an individual being selected at random out of a line of people standing one metre apart stretching for one kilometre. In community terms it means that you could expect to find about one person affected in every population grouping the size of a rural village or an inner city housing estate. Many people would judge this level of risk to be moderately low compared with other risks of normal living
On the best evidence currently available the chance of someone being affected by this hazard is 1 in 1 000 000. This is magnitude 4 on the 0-10 risk scale. This level of risk is analogous to the chance of an individual being selected at random out of a line of people standing one metre apart stretching from London to John O'Groats. In community terms it means that you could expect to find about one person affected in every population grouping the size of one of the largest cities or average county in Britain. Many people would judge this level of risk to be minimal or even negligible compared with other risks of normal living
(In any specific case these statements would also need to make clear the nature of the risk, the time frame concerned, and the population group being considered.)
Knowing the magnitude of a risk is just the first step in comprehension. A further step might be made by considering how this magnitude compares with that of some other risk. The information in tables 1 and 7 allows examples of such comparisons. For instance, the risk of being killed by lightning is about one thousandth of that of being killed in a road accident. Risk comparison is a somewhat contentious area, particularly when it involves comparing risks with very different features,14 but even this can be useful where the emphasis is on conveying a feeling of the magnitude of a risk, rather than on insisting that a given risk must be acceptable if it is smaller, or unacceptable if it is larger, than some other risk that people already take7. Comparisons of risks with similar features do not present such difficulties but even then when relative risks are stated it is important also to state the risk in absolute terms. People's reactions to being informed that the risk of treatment A is, say, double that of treatment B may be very different depending on the level of absolute risk. It is likely to matter whether it is appreciated, for instance, that although the risk has doubled the rise is from one in a million to two in a million (rather than, say, from one in a hundred to two in a hundred), as shown perhaps by recent experience with publicity about the risks of third generation oral contraceptives.3
As well as the basic probabilistic aspect, risk has many other facets such as the severity of the adverse event in question. Furthermore, people's attitude to risk depends on the context - for instance, whether the risk is voluntary or imposed, whether adverse events are concentrated or dispersed over time or place, and whether the risk is framed in a negative or a positive way.7 8 7 Whether a hazard is seen as “dread” and whether it is regarded as an “unknown” are particularly important factors; hazards which score high on both these aspects generate especially strong concern.9 15 It would in principle be possible to extend the risk scales shown to allow for differences in severity of adverse events, or to include more sophisticated risk measures such as years of life lost, or to distinguish between different contexts. However, this could easily overburden what seems best kept as a simple tool for communication of basic risk. The scales should be limited to clarifying the presentation of probabilities of adverse events (such as death or injury), leaving deeper investigation to heavier equipment.
These risk scales are intended to help with the first steps of communication about risk. Of course, they would need to be tested. Their value entirely depends on if and how they improve people's ability to understand and communicate about risks. It is hoped that they will help to provide a language and some useful dialects for risk—risk scales with a human face.