That's not the right way to think about this. Let's take binge-drinking, for example. Let's suppose that each binge-drinking episode has a .0001% chance of cau

That's not the right way to think about this. Let's take binge-drinking, for example. Let's suppose that each binge-drinking episode has a .0001% chance of causing a fatality. That is trivially low. However, in order to consider the trade-offs involved, you have to think about the expected value of your life in either scenario. So you must compare E(life | no drink) with E(life | drink). These are practically the same, but where they differ is that in the drink scenario, your expected value is broken down into two scenarios -- one if you die as a result of the binge drink, and one if you don't. So you can break E(life | drink) into 99.999(E(life + d)) + .001d, where d is the added enjoyment you get from that one night of binge drinking. The comparison, then, is ultimately whether E(life) >=< 99.999(E(life + d) + .001d You can tell from this that what really matters is the ratio of d to E(life), the enjoyment you get from drinking that one night, compared to your overall enjoyment of your ENTIRE subsequent life. As we extend your life expectancy, as long as we assume that life is a good thing, we make the value of your subsequent life higher and higher. This, in turn, should make any rational actor be less willing to take on risks that would cut it short.