On the drive to work the other day, shortly after I narrowly avoided being flattened by a bus which started changing lanes, I realised there is a wealth of information which is not collected or considered when working in risk modelling. The data associated with events which did not happen, or near-misses.

If I had been flattened by the bus, there would have been an accident report filed, insurance claims made, drug tests, speed checks…and the T.A.C. would have another file to add to their crash statistics. They would have had a snapshot of how the crash happened.

But in risk modelling, and mitigation, not only do you need to know how/why the accident occurred, it’s important to be able to identify ways in which the accident could have been avoided, thus, reducing the risk. But if you only collect information on accidents which actually occur then you’ve got no record of any actions which prevent the accident from ever taking place.

In my case a blast of the horn was sufficient for the bus driver to realise that I was right next to him, thereby avoiding what would presumably have been a rather messy accident. Anyone who drives regularly has stories like this, where they avoided an accident somehow, but if this information is not collected and analysed then how is it possible to reliably reduce the risk of an accident occurring. This does not only apply to traffic accidents, it applies in any risk-modelling/mitigation scenario;

Understanding why something DIDN’T happen is just as informative as understanding why something DID happen.

Back in 1931 H.W. Heinrich had a similar realisation, with respect to workplace safety. After studying many workplaces he arrived at the ratio of 300:29:1. That is, for every 300 ‘near-misses’ there are 29 minor accidents and 1 major accident.

Heinrich’s Triangle
If this ratio or one similar were to hold for other ‘risk’ scenarios, and there is no reason to suspect it wouldn’t, then there is a large, untapped data source in ‘near misses’.

Obviously there are technical difficulties and high costs associated with collecting ‘near-miss’ data, but it is a resource that is worth considering.

When the goal of risk modelling and mitigation is to understand and reduce the risk, having information about real-life avoidance situations provides working examples of possible mitigation strategies and their effectiveness.

To return to my problem with the bus, I honked my horn, which alerted the driver to my presence, at which point he took action which reduced the risk to me and my car. While I’m not suggesting that everyone drives around with their horn blaring, it is a risk minimisation tool when used in the right situation.

Sometimes a database of events which didn’t happen is more useful than a database of events which did.