Promote people at random

Five years ago Amanda Page (then my PhD student) and I published a paper (free pdf here) that was computational statistical physics but was inspired by the problem of protein crystallisation. It is one of my favourite papers. An hour ago I got some automated email saying that another paper had cited this work of Amanda’s and mine. I don’t know why I got the email, I don’t think I signed up anywhere, but whatever, it is always nice that someone has referred to your work – presumably this means they have read it. Then I read the title.

The title is “Efficient promotion strategies in hierarchical organizations“. My first thought was that this must be a mistake, what has Amanda’s and my work got to do with promoting people? So I took a look and it is not a mistake. They do reference our work, as they do a number of other statistical physics papers because they are also using statistical physics, only not to understand protein solutions but to understand the effect of different promotion strategies. So there is a connection, truly statistical physics is a useful thing, it takes us from protein crystals to promoting people in a single step.

It is quite an entertaining paper. They are inspired by the idea that is often called the Peter Principle, which is that people are promoted and promoted until they reach a level at which point they are clearly incompetent, at which they stick and are not promoted again.

Basically, what the authors Pluchino et al. do is study on a computer a simple model in which they can change the promotion strategy, i.e., how people are selected for promotion. They then run the model with different strategies and see what organisation efficiencies come out.

Anyway, under certain assumptions, promoting people at random is actually the best strategy, it produces more efficient organisations than promoting the best person! From what I can tell of the model this is because of two problems with promoting the best person. The first problem is that you are then promoting someone very good at their job, who is then replaced by someone who is almost certainly less good at that job – which hurts efficiency.

The second problem is that it is also likely that this promoted person is less good at their new job than they were at the old job. This assumes that people are not equally good at two different jobs, the old and the new one they are promoted too. Then if there is a bit of variability, what is called regression to the mean takes over.

Anyway, you could argue that the model is too simple, makes too many assumptions etc. But who knows, maybe promoting people at random is the best strategy. It would also save time. No more writing cases for promotion, updating cv’s etc. Just pick a lottery ticket out of a hat….