One obvious example is that smoking 1.4 cigarettes is one micromort.
Doesn't this assume the risk accumulates linearly and irreversibly? I'm pretty sure you can't do that.
It does assume linearity but not irreversibly.
You can calculate conditional probabilities based on chances of quitting and not quitting cigarettes in the future.
Linearity is a more interesting issue. There the model of cancer being caused by a single mutation. In that model a single cigarette has a certain chance of creating a single mutation. If you deeply believe in that model you can justify linearity to some extend.
In reality cancer however seems more complicated. But it's hard to calculate the micromorts of a single cigarette. We don't have a good the...
Some of us here are already familiar with Micromorts - a unit that stands for a 1 in a million chance of dying. The wikipedia page lists a number of sample values. One obvious example is that smoking 1.4 cigarettes is one micromort. This is a good tool for comparing the relative dangerousness of activites - for example, if you fly in a jet in the US, your micromorts per mile from increased background radiation are twice the micromorts per mile from terrorism. And you can compare activities to baseline average risks of death, given as about 39 per day (averaged over all age groups and sexes).
However, people suck at imagining small probabilities. So a different unit, which we used in a group exercise at the Secular Solstice in Leipzig, is the number of expected future heartbeats. While Micromorts are a step away from empirical reality, expected heartbeats are another step in the same direction. But the concept got good feedback, it makes people think about life in a new way, so I thought I'd just share it here.
The average human heart gets to beat about 2.5 billion times - about 100000 times per day. So a micromort is around 2500 expected heartbeats. So you can translate, say, smoking a cigarette into a cost of about 1800 expected heartbeats (or about 80 seconds of life expectancy). And maybe that'll help people optimize their behavior in ways that Micromorts, due to their microprobability nature, aren't very good at doing especially for those who aren't very habitual Bayesians.