Risk know-how Framework

Jordan Phasey, Water Treatment Professional, Australia

When making decisions, we need to clarify what the risks are to individuals as well as how many people could be affected, how likely they will be affected, the kind of threat, the doses and for what period of time.

For example, 2 cases of food poisoning, 2 months in which a valley flooded, 2 people killed by machinery.

For example, 2 out of 100 people tested, 2 months of flooding out of 24 months monitored, 2 out of all factory employees in a region.

The risk to homes from an earthquake might be assessed in one suburb but be relevant to all suburbs. What's more, other suburbs may be more likely to have substandard housing, and alter the risk calculation.

No hurricanes in Florida (yesterday)!

If the pollution in the river is reduced, it may have no benefit for fragile species that live in parts of the river that are still polluted.

Alejandra Albuerne, Architectural Engineer and Construction Historian, Global

After the 2015 [Nepal] earthquakes, the communities saw the concrete buildings that had survived as the buildings they would feel safe living in. [However,] in a different earthquake many of those would not have been safe either.

Gabriella D'Cruz, Founder of The Good Ocean, India

Much of the risk involving climate change related phenomena such as sea level rise, the increase in cyclones and the declining health of oceanic ecosystems are difficult to talk about.. these risks seem to have a longer time scale to pan out compared to immediate risks.

Compare the numbers to what is expected, what happened before or to other risks, and understand whether there will be new consequences.

In terms of population size, it might sound trivial if a computer hardware system has a 0.1% chance of failing over one year. But if it is used in 10 million computers, we would expect around 10,000 computers to fail in a given year.

In terms of period of time, a 1 ̊C raise in temperature for the next day might not feel like much, but if sustained for decades, it would transform the entire ecosystem of the Earth.

And in terms of consequences, a 50,000 ton drop in wheat production is meaningless without knowing that 760 million tons are produced per year, and whether the drop will affect one country's consumption.

Katelyn Jeterina, Founder of Your Local Epidemiologist, USA

People dont usually think in numbers. One thing that I learned is not to use percentages. Instead of 50%, you say one-in-two people, instead of 20%, you say one-in-five people.

Kees Balde, Senior Scientific Specialist, Sustainable Cycles, United Nations Institute for Training and Research (UNITAR), Netherlands

With our messages, we try and use relatable figures. For example, 50 million metric tons, I don't know how much it is. However, 7 kilograms per inhabitant per year, I can understand. We also always name the appliances... like your fridge, your desktop, your phone.

High and low, big and small, can be understood differently by different people and in different situations. Communities with risk know-how use numbers - such as percentages or fractions - to clarify what is meant, where this isn't obvious from the context.

The US weather service uses specific labels to state the probability of a tornado appearing within 25 miles. It defines "Marginal" as 2%, "Slight" as 5%, and "Enhanced" as 10% or more. Without the quantitative definitions, it would be difficult to concretely compare the meanings of "marginal," "slight," and "enhanced".

Sarah-Jane Ambler, Research Midwife, UK

We label all pregnancies ‘high risk’ and ‘low risk’. When women are screened for foetal abnormalities, results are most often reported as a statistical chance. All hospitals have a number they use as a cut-off... ths number is usually 1 in 150. This means that all women who have a result between 1 in 150 will be offered a diagnostic test.

The way a risk is presented influences how people respond to it. It's important to be aware of how comparisons or context might be doing that, and question whether the same information might seem different in another context.

A factory might boast that it has the lowest risk of accident in a region. However, if it's a shoe-packing factory in a region of dangerous wood cutting factories, the comparison could be misleading — the factory might actually have a terrible risk of accident compared to other shoe-packing factories.

Patricia Nanteza, Associate Director for Africa at the Alliance for Science, Uganda/Kenya

There is absolutely great value in considering the culture of people and what is influencing certain behaviours. They are making complex decisions that are influenced by cultures that we may not understand. Before putting out more and more information, we need to understand the story behind their behaviours.

The relative risk tells us how much higher or lower a risk is. The absolute risk tells us the actual chance of something happening.

If the risk of plane crashes has halved in a decade (relative risk), it sounds like a dramatic decline. However, there are so few plane crashes — 1 for around 7.7. million flights (absolute risk) — that this would amount to no change to the safety level for almost everyone who flies.

Rocio Perez Benavente, Maldita Ciencia, Spain

People were worried about the AstraZeneca vaccine causing blood clots, so we wanted to offer some context for the numbers. We used an infographic from the Winton Centre showing the risks to different ages, and kept the information in the context of other side effect risks and risks from Covid-19 itself.

Averages are calculated in different ways, known as mean (add then divide the values), median (set out the range of values and find the one in the middle) and mode (the most common value). Things rarely happen in the same predictable way, so averages are the only way to calculate a risk. But they can hide important information about what is happening at the extremes.

Average pest damage to corn of 500 plants per farm might hide the fact that pests had wiped out one farm and left others unaffected. Averages also hide differences in the intervals between risk events. A '1-in-30 year flood' is an average taken from past data. It doesn't tell people about the timing — two of these floods could happen in back-to-back years followed by 50 years with no floods.

Probabilities are all made in reference to something else: a 20% chance of something happening comes from adding up the times it did happen, compared to the times it didn't. It is important to know what the reference group for the risk is.

Sometimes the reference can be ambiguous. When we say that there is a 30% chance of rain in a region, the reference is other times when the region had the same weather conditions — it doesn't mean that it will rain in 30% of the region or 30% of the day.

Rebecca Blaylock, Research and Engagement Lead, Centre for Reproductive Research & Communication, British Pregnancy Advisory Service (BPAS), UK

Women we spoke to found it very difficult to contextualize risk. For example, women who are living with obesity are constantly being told that they are at higher risk of a poor outcome but are never told by how much. Risks are extrapolated from large scale epidemiological studies and then applied to an individual in a clinical setting, but without describing how that risk is calculated and translated to an individual, it's very difficult for them to contextualize it.

Few things are 100% certain – and everything less than this is uncertain, so people have to ask how confident they can be. Where uncertainty can be calculated, it should be clear if this is incorporated in risk predictions. Often uncertainty cannot be quantified, but decisions still need to be taken. If more information becomes available, the risk calculation and the decision might have to change. Talking about this possibility is helpful for communities navigating risks.

Assessments of large bridges calculate when they are likely to fail. Making the calculation in different ways gives a range of answers. We can then see whether an answer - say, 'in 50 years' - is vastly different to the others and calculate how likely it is to be the true answer.

Tim Gill, Global Advocate for Children’s Play and Mobility, UK

Humans can have an ambivalent relationship with uncertainty. In the workshops I run even just talking about uncertainty explicitly and expressing their fears can unburden people.

Conditional probabilities present the likelihood of a particular outcome given something else that's happened.

For example, calculating the risk of a system failure following an alarm sounding. The calculations are complex because the references for the initial event (the system alarm) and the conditioned outcome (the system failure) are different. It's more useful for most people to understand how much or how often something is happening, but all the same it is important to know that people communicating news about risk often miscalculate conditional probabilities, especially when talking about monitoring and tests.

When we see changes in two things happening at the same time, it doesn't necessarily mean that one causes the other — other factors may be involved.

When a change in fishing activities coincides with a rise in fish stocks, one didn't necessarily cause the other. More questions need to be asked: What other factors affect fish stocks, and how can we obtain data about those factors? Is there a third factor that is common to both - e.g. a new environmental agency changed the fishing permits and also stopped toxic waste disposal.

Inilek Wilmot, Marine Biologist, Jamaica

The fishing community will often point to the expansion of tourism as the root cause of reef decline because there is a correlation between tourism growing and the fish population declining. A lot of them will be very adamant that fishing has nothing to do with it. [Because] people make their living from catching fish, they have a resistance to any information that might indicate what they’re doing is causing the problem.

Adjustments are made to correct imbalances in the data used to calculate risks. People with risk know-how ask whether biases and limits have been taken into account and what has been done to eliminate them.

We might want to adjust survey results so that they represent the population being discussed rather than just the people sampled, if more men or employed people were asked to take part for example.