How many times have you heard of something — or someone — being an “accident waiting to happen”? This common idiom is applied to individuals, collective activities like industrial manufacturing, and even specific locations like traffic intersections. Clearly, it states a supposition about the future, yet no one really knows when or if any accident will occur or how serious it may be. This statement presents an assessment based on data or intuition or both. It’s a generalized prediction that perceives an imminent risk, and is often spoken in frustration, disgust, or concern.
And it’s not all that unusual. Most of us are able to spot those “accidents waiting to happen.” Personally, I believe that there is no such thing as a random failure or a fortuitous accident. They happen for one or more of these reasons:
• We don’t have the right measurements in place to identify precursor events.
• We lack the knowledge to identify precursor patterns.
• The risk-mitigation measures we took were inadequate.
Accident prevention relies on the ongoing cycle and discipline of measurement, analysis, and action. On the surface, this task sounds easy, but in reality the processes involved represent a demanding part of corporate strategy and execution. The historical difficulties in collecting and storing information before recent technology advancements have been replaced today by equal — if not more challenging — decisions related to what data to collect and what actions should be taken from the final analyses. And since risk is a function of time, data requirements can and do change. If you think that’s easy to accommodate, just take one look at your database administrator’s face when you request a new field added to a table. You might just get a glimpse of how hard dynamic measurement really can be.
There’s no silver bullet for a perfect measurement strategy, but there is a transparent, robust, risk-based format that can help identify data trends related to the events that can precede accidents. And there are ways to capture information in the moment even as data is constantly moving. This format can be a useful part of a measurement strategy.
Start with the event-frequency-versus-severity coordinate system I introduced in my September article. Using that basic system (risk = frequency x severity), we can plot data computed over successive time intervals. For example, event frequencies — or the probability of an event occurring — can be the quotient of the total number of events and total time or independent calculations for each time period depending on the situation. Event severity is usually measured by other statistics, such as an average value to provide some stationary information to enable comparisons between time intervals. Regardless of the specific units, the primary function of this framework is to provide time-dependent information regarding event frequency, severity, and risk.
To demonstrate, I’ve presented three examples in the chart below, say for the risk of events (or injuries) occurring at a manufacturing plant that could result in a loss of production. Five points corresponding to the same five time intervals are plotted in each case. The frequency and severity axes are shown as logarithmic scales displaying the constant risk relationships as straight lines. Frequency is given as the number of events per plant per year, and severity as the average cost of an event. The subsequent risk is shown as the event cost per plant, per year.
The examples illustrate how to identify potential trends using data displayed visually. The data is fictitious yet emblematic of actual situations. In the frequency trend, notice that the point always moves to the right, even though severity moves up and down. An analogous pattern is seen in the severity trend. In the risk example, the points always move in the same upper-right direction as time advances.
This graph shows the data in a simple way. It also has an interesting property since managers can infer trends in frequency, severity, and risk with a higher confidence than any certainty they would have over the actual position of individual points. In these types of plots, points are often defined by coordinates that are statistics, such as proportions or averages. For example, severity is typically defined as the average loss per event, while frequency is shown as a proportion describing the number of events per unit time. In order to identify suspected trends, focus only on the direction of movement between time periods — not on point locations. Even in its rawest form, this type of graph can prove valuable to risk identification and mitigation.
Here’s the idea. For each coordinate, assuming the data points are independent, there’s a 50-50 chance the value will be larger or smaller for the next time period. In other words, the probability that a point will move higher or lower in either severity or frequency is 1/2. For trends in risk, the probability that the movement is in the same quadrant is 1/4. Therefore, the probability of moving in the same direction in frequency or severity for four consecutive time periods is 1/16 (which is one-half to the fourth power). This is analogous to the situation of getting all heads (or tails) in four flips of a coin. Of course, the conclusion that a trend actually exists is yours, but the math tells you the rarity of the sequence. And that rarity allows fairly safe conclusions regarding whether or not a trend exists.
For more-detailed information, the graph can be enhanced to show degrees of uncertainty for each point or other information.
In most business-analysis plots, data represents known history that we use as portals to peer into the future. The frequency, severity, and risk plotting format gives us an integrated and visual peek into this same future. Such graphically displayed information can help stakeholders identify patterns utilizing their knowledge, intuition, and experience that are not as easily recognized from columns of numbers. The plots can also help communicate risk-based concepts and measurement results to a large audience. As such, they are invaluable tools in obtaining consensus for risk mitigation and other business actions.
And with the right measurement processes in place, applied with the right amount of discipline and creativity, maybe — just maybe — you can prevent that accident waiting to happen.
Rick Jones has spent the past 30 years applying risk analysis and management techniques to industrial and business problems. He has presented at several conferences and is the author of numerous articles and technical research papers. His third book, 20% Chance of Rain: Exploring the Concept of Risk, will be published in November.