Uber's Employee Ratings Put Women at a Disadvantage, Suit Says

Decimation was a system for BUILDING cohesion. You don't flee battle because you might be decimated. You don't let your comrades flee battle because you might be decimated. It's for when you want to severely punish a collective group of people but can't afford to hurt them all.

Stack ranking is the opposite. You are actively competing against your colleagues. You are incentivized for your colleagues to fail.

But five minutes of thought will tell you that people will hate working on that environment and will quit or just not apply to work there.

It is perhaps a natural fear that a Stack Ranking system would lead to a "Rank and Yank", but the mere ordering of employees during a performance management process doesn't give me any particular concern, nor do I consider it a bad faith system.

In fact, it forces people to make a sometimes hard choice, which I believe has more benefits than drawbacks to an organization, else you run the risk of having a massive "undifferentiated middle" and no clear development plans because you have no feedback from rich discussion about why Jane is better than Joe; instead Jane and Joe are simply declared to be equal without any discussion about what Joe could do to improve or what Jane should continue doing, etc.

With Romans, specifically, if cowardice was punished, there would be noone left. They were disciplined and had great quality equipment, but their morale was not the highest. For instance, before they met the Germans on the field, having heard of the military prowerss of their enemy, the Romans would spend their time crying and making their will, certain that they'd die in battle - Cesar has written about this.

The specific problem with stack ranking is the one-on-one ranking method. I'm absolutely convinced that I myself would succumb to all sorts of biases when asked to rate people like that: I'd tend to think the silent, introvert guy who stays late and has no fashion sense is smarter than the outgoing jock leaving at 5pm sharp to be with his family. I'd somehow convince myself that the guy I enjoy eating lunch with is more important than the Belgian guy who doesn't speak english well enough to socialise... The article makes a similar point: “My observation was whoever was the best orator got their people ranked higher.”

On aggregate, it's quite obvious that these biases could amount to gender-specific discrimination, even if not a single evaluator had gender on their minds.

When managers are faced with stack ranking employees, they will often decide based on factors not purely related to the productivity or value of the employee.

----

That, and as an employee in a stack-ranked organization, you know you need to be competitive. Not in terms of productivity but in terms of "in the eyes of your manager"-- Perception is more important than performance.

Studies have shown that men compete more aggressively than women, on average. So it's certainly plausible that men enjoy a gender-driven advantage that isn't purely due to gender-discrimination. Rather, it is the system that is biased.

This reminds me of that "racist algorithm" debacle.

Most people who use ridesharing see it only as a cheaper, easier option, which it is.

Most of the people I know who use or drive either because they seem them as commodities and don't really care. They just pick the better priced or closest one.

It is a real trend, but doesn't represent 'most' users.

Problems like this are endemic to zero-sum schemes like stack ranking.

One where no one ever will complain? No, but so what?

But absolutely eliminating sexism isn't the bar most people are held to. Most people are held to the bar of taking reasonable steps to mitigate the issue and improve culture.

In the vast majority of cases, you can iterate your way to a fair system with a clear rubric that controls for unconscious biases and establishes a substantial trail of evidence to support each judgement of performance. Actually doing that, especially when giving criticism, tends to be a matter of managerial skill.

In turn, giving up on building a less sexist system because people will always complain is like giving up on building a more secure infrastructure because people will always complain.

Unless you are extraordinarily lucky at both the humans building what you're building and the inputs to your system (constraints about language and tech stack and problem domain on the security side; constraints about hiring sources and existing culture/public reputation and leadership on the employment side), you're always going to have security flaws in your infrastructure, and you're always going to have sexism in your employment processes. And even if you are this lucky, you're not going to eliminate people poking at your system and suggesting places where it seems to fall short. The important question is what you do about it, how severe the problems are now, and how the severity changes over time.

Just like there are plenty of ways to write software that take it as a given that implementors will screw up and attackers will be clever, there are also plenty of ways to design a performance review system that takes it as a given that some fraction of humans in your organization will have biases (conscious and unconscious) and will treat their coworkers according to their biases.

Honestly, some of the mitigations are the same: the first thing I'd reach for is defense-in-depth and auditing. If your system is consistently giving different results (whether it's hiring, or retention, or promotions, or salary) to people in different demographic groups in a way that seems to disprove the null hypothesis that the demographic is irrelevant to job performance, look very hard at it, add additional reviews to make sure the results are consistent even when different people are involved, and keep measuring. There might be good reasons that the disparity is legitimate (just like there might be good reasons that one of your servers is regularly opening outbound TCP connections to random machines in China), just make sure you are actually noticing that something unexpected is going on and try to find an explanation for it.

Uber seems to be full of bad ideas, so maybe it's a good fit.