Tuesday, August 27, 2013

Stack Ranking Continued

More on the Microsoft ranking system:
This was my problem. I had three reports, A, B, and C, and they neatly fit into three categories: C was good, B was great, and A was fantastic. They were all nice and retiring sorts—they weren’t self-promoters, which put them at a disadvantage at Microsoft—and I did want to do well by them. Based on their position in the stack rank, I thought that this would have been a fair assessment of them relative to the company in general:

My Ideal Distribution

A: Above Average
B: Above Average
C: Average
Above Average would get A and B nice bonuses and raises, while C might get a small raise and a decent bonus with an Average. That didn’t happen. My manager told me baldly that this was how it would go:

The Actual Distribution

A: Above average
B: Average
C: Below average

My desired rankings were out of the question, since my manager would then have had to steal that extra Above Average from some other manager. I thought that B could live with Average (we were all well-compensated, after all), but rating C as Below Average hurt.

So I argued for C, and my manager said there was exactly one alternative:

The Alternative Distribution

A: Average
B: Average
C: Average

But A had been at the very top of the stack! How could A do worse than people we’d all agreed were weaker programmers? I gave up and let C take the Below Average. This is the zero-sum game at work.

I still feel bad about this.


This very much shows how the ranking ended up being all about politics and balance between different managers.  You can immediately see how taking a weak person could be a great idea to protect the rankings of the people actually doing the work.  And how good teams would be the absolute worst place to possibly join. 

I was mistaken about Mark P being a proponent of the system in a previous work situation (it is the only way to make IT workers productive was the gist but I don't recall the details and the other obvious person from that era has drifted out of touch).  But it keeps getting clearer and clearer that this kind of ranking system can lead to perverse incentives.  The actual example however shows something else -- the managers are imposing the distribution of ability on a three person team.  That the whole organization should have a wide distribution of talent is likely (the law of large numbers and all) but that any three person team will happen to follow an ordered distribution balanced around the grand mean of the company is a rather heroic statistical assumption

Now consider education reform and imagine what would happen if you tried the same thing with teachers.  All of a sudden there are new and interesting incentives to keep "poor teachers" because the distribution is fixed by design. 

No comments:

Post a Comment