Suppose a shortcut on the job has a one-in-a-thousand chance of going wrong. Not a big chance. You could take that shortcut every day for a month and almost certainly nothing would happen. You might start to think it is safe.
Here is the math that says otherwise, and the ideas safety engineers use to fight it.
The arithmetic of many days
A one-in-a-thousand chance means a 99.9 percent chance that it goes fine on any one day. To find the chance that it goes fine every day for many days, you multiply 0.999 by itself once for each day.
The results surprise most people:
| Times you take the risk | Chance something goes wrong at least once |
|---|---|
| 500 | about 39 percent |
| 1,800 (about 10 years of 12-hour shifts, at roughly 180 shifts a year) | about 83 percent |
| 3,600 | about 97 percent |
| 5,400 | more than 99 percent |
A risk that is tiny on any one day becomes close to certain over a career. And this is for one person. Spread the same habit across a crew of fifty, and it doesn't take years. It takes weeks.
Even a one-in-ten-thousand risk, taken 5,400 times, comes out to about a 42 percent chance of happening at least once. That is the reason good plants treat "it's never happened" as a warning sign, not proof of safety.
Why we misjudge it
Human beings are not built to feel this math. We learn from experience, and experience says: I did this a hundred times and nothing happened. Each uneventful day feels like evidence that the risk is smaller than we thought. In fact it is only evidence that we've been lucky so far.
The sociologist Diane Vaughan, studying the 1986 Challenger space shuttle disaster, gave this a name: the normalization of deviance. Engineers had seen damage to the shuttle's O-ring seals on earlier flights. Each time, the flight came back safely, so the damage came to seem acceptable, then normal. The standard drifted, one successful flight at a time, until the day it didn't hold.
The same drift can happen anywhere. An alarm that goes off so often people stop hearing it. A step skipped once to save time, then skipped every time. Each small step away from the rule feels reasonable because the last one turned out fine.
Holes in the cheese
The psychologist James Reason gave safety one of its best-known pictures, usually called the Swiss cheese model. Imagine a stack of cheese slices. Each slice is a layer of protection: the design of the equipment, alarms, procedures, training, permits, a second person checking the work. Every layer has holes, because nothing is perfect. A procedure is out of date. A gauge reads wrong. Someone is tired.
Most of the time the holes don't line up, so one layer catches what another missed. An accident happens when the holes in every layer line up at once, and the hazard passes straight through.
The model teaches two lessons. First, almost no accident has a single cause. Second, every layer matters, including the ones that seem redundant. A shortcut that skips one layer doesn't just add a small risk. It removes a slice of cheese, and makes it more likely that the remaining holes will line up.
What the rules are for
Seen this way, a lot of plant rules make more sense. The permit, the lockout, the gas test, the second signature: each one is a slice. They can feel like paperwork, especially on a day when nothing goes wrong. But the point of a layer of protection is that most days it isn't needed. It is there for the day the other layers fail.
After the 2005 explosion at the BP Texas City refinery, which killed 15 contract workers, the U.S. Chemical Safety Board found a long list of warning signs that had been treated as normal, and layers of protection that had worn thin. Investigations of major accidents around the world tell a similar story again and again.
What one person can do
You can't fix the math. But you can work with it.
- Count the exposures, not the incidents. Ask how often you take a risk, not whether it has bitten you yet.
- Notice drift. When you catch yourself saying "we always do it this way," ask whether "always" started as an exception.
- Treat near misses as gifts. A near miss is the hazard showing you which holes lined up, for free. Report it.
- Keep your slice solid. You are one of the layers. On a tired night, that matters more, not less.
- Use your stop-work authority. Stopping a job is a layer of protection too.
Safety people sometimes say that luck is not a strategy. The math says the same thing more precisely: given enough tries, luck runs out. The work of safety is to make sure that when it does, there is still another slice of cheese in the way.
Related on HERE
Sources
- Probabilities computed by HERE: chance of at least one event = 1 − (1 − p) raised to the number of exposures.
- Diane Vaughan, The Challenger Launch Decision (1996), normalization of deviance.
- James Reason, Human Error (1990) and later work, the Swiss cheese model.
- U.S. Chemical Safety Board, investigation report on the 2005 BP Texas City refinery explosion.