Let’s suppose your child wants to take a martial arts class. Being a conscientious parent, you check out the local dojos and find two good places. Both are suitable and well equipped. Both practice fighting with contact – but there’s one major difference. One dojo insists on a full range of protective padding – hands, feet, chest protectors, shin guards – the whole works. The other takes a much lighter approach – hands and feet, and sometimes not even those.

To the conscientious parent, the first place is going to look much safer, right? But when you look at the injury rates of the two dojos, you notice something odd: They’re about the same. The kids covered in foam padding are getting just as many bruises, scrapes, and sprains as the kids wearing almost none. What could be going on here?

What’s happening is a process known as risk compensation. It’s a tendency in humans to increase risky behavior proportionately as safeguards are introduced, and it’s very common. So common, in fact, as to render predictions of how well any given piece of safety equipment will work almost useless.

In the instance of the mini-ninjas, those with pads are likely hitting and kicking harder and more wildly than those without, and the adults supervising them are likely to be allowing it. Why would we do such a strange thing? Dr. Gerald Wilde of Queens University in Ontario proposes a hypothesis he calls risk homeostasis. In a nutshell it proposes that human beings have a target level of risk with which they are most comfortable. When a given activity exceeds their comfort level, people will modify their behavior to reduce their risk until they are comfortable with their level of danger. So far, that’s not exactly a controversial observation. But risk homeostasis proposes another half to that continuum – according to Dr. Wilde, if a given person’s level of risk drops too far below their comfort level, they will again modify their behavior. This time though, they will increase their level of risk until they are once again in their target zone.

It seems an odd proposition, but Dr. Wilde and his colleagues have assembled an impressive array of data to support it. For instance, a study of Munich taxicab drivers conducted while the taxicab fleet was being changed over to ABS braking systems. The drivers were tracked by observers unaware of which kind of brakes each cab had. Against the expectations of safety experts who recommend ABS brakes as a safety advance, the drivers with ABS brakes actually had more accidents per vehicle mile than those without. The drivers braked more sharply, made tighter turns, drove at higher speeds, and made a number of other adjustments to their driving, all of which more than compensated for their supposedly safer cabs.

Fortunately for us, risk homeostasis does not seem to apply in all cases. Safety innovations that are invisible tend not to provoke changes in behavior – for example changing windshields to safety glass does not alter most peoples’ driving behavior. The difference in the windshield is effectively invisible to the driver, and so doesn’t affect the driving. The taxicab drivers, by contrast, were intimately familiar with their cabs, and the difference in braking was apparent to them. Risk compensation can also be affected by motivation. A taxicab driver has every reason to try to get from A to B faster, but someone out for a Sunday drive to see the scenery would be less likely to go quickly in response to better braking.

Unfortunately for those whose job it is to make us all safer, risk homeostasis and risk compensation are not easy to study. It is difficult to predict how people will alter their behavior in response to a given piece of safety equipment, and thereby equally difficult to figure out what to be looking at in a study. Seatbelts, for instance, have an modestly positive effect on driver and passenger safety (though somewhat less than models predicted). It’s not until you look at pedestrian safety that you really begin to see where risk compensation may be having its way.

An additional complication for the already beleaguered safety engineers is that risk homeostasis is dependent not upon actual danger, but rather the perception of risk. Much of the gender and age differences in risk-taking behavior appear to stem less from differing desires for risk, and more from the individual’s different evaluation of risk. Young people, and particularly young men, tend to evaluate their level of risk as much lower than older people would, even in identical situations. This implies that promoting safer behavior depends more upon altering the perceptions of the target population, rather than improving the safety of the environment⁠— a much trickier proposition.

What it all boils down to is that the law of unintended consequences is extraordinarily applicable when talking about safety innovations. Sometimes things intended to make us safer may not make any improvment at all to our overall safety, and in rare instances they may actually make us less safe. The human tendency to take risks may trump all the efforts of the safety engineers. In the end, no one can save us from ourselves.