Probability Lessons For Workplace Safety
A Safety Program Is Not Just a List of Hazards
One way to misunderstand workplace safety is to treat it as a cataloging problem: identify the hazards, write them down, and hope the list remains accurate.
That resembles a probability model with a fixed sample space. Useful, sometimes. Also dangerously incomplete.
A workplace changes. Tools move. Procedures get improvised. New people arrive with different levels of training. A low-probability event can become much more likely when the conditions around it shift, even if nobody has formally declared a new hazard.
The observation problem In probability, we often begin by asking what outcomes are possible. In safety, the harder question may be: what outcomes are we failing to observe?
Near misses, awkward workarounds, confusing instructions, and small equipment problems are data. They are not proof that an injury will occur, but ignoring them creates a biased picture of the system. The visible incidents are only the portion that crossed a reporting threshold.
This is one reason prevention cannot be reduced to reacting to accidents. The absence of a recorded injury is not strong evidence that the process is safe. It may only mean that the process has been lucky, underreported, or not yet stressed in the relevant way.
A small mathematical warning Suppose an event has probability 1% on any given occasion. Seeing no event in 50 occasions is entirely possible: the probability is approximately 0.99⁵⁰, or about 60.5%.
So “nothing happened” may tell us less than intuition suggests. Repeated safe outcomes can coexist with a meaningful underlying risk, especially when the cost of one failure is high.
Of course, real workplaces are not independent-trial machines, and assigning a number like 1% is often unjustified. The calculation is a warning against false reassurance, not a complete safety analysis.
OSHA’s Safe + Sound Week 2026 emphasizes safety and health programs, hazard identification, and prevention. I like that framing because it treats safety as an ongoing feedback process rather than a document produced once and filed away.
In mathematics, the model improves when we find observations it cannot explain. In safety, the equivalent may be a worker saying, “This is how we actually do it.” That sentence can be more informative than a perfectly completed checklist.