3 ms·

What does sorting by performance look like to you? The curve determines rating and compensation, influences promotion decisions, projects people get to work on

What does sorting by performance look like to you? The curve determines rating and compensation, influences promotion decisions, projects people get to work on and what opportunities come their way. The seems like a sort to me given that budget and project opportunities are fixed in many places.

nightpool 4y ago
The key component of stack ranking / sorting by performance is that there's no absolute measure of your contribution, only a relative measure by which you have to outcompete everyone else in your team/department. If you fire your lowest 15% of engineers—no matter how good they are on an absolute level—that's a stack ranking system. It's a horrifying Red Queen's Race that punishes high-quality teams. I'm not necessarily saying that Meta's system is good, but at least it's not that. EDIT: It sounds like at least some people are claiming that Meta has an internal policy that X% of your engineers under you are supposed to be given "Meets Most Expectations" ratings, which would also be pretty close to stack ranking (if at least allowing for some management leniency) https://news.ycombinator.com/item?id=32930370 https://news.ycombinator.com/item?id=32930370
- joshuamorton 4y ago
  The other key component of stack ranking was that it's on a per team basis. A team of 20 people may have no low performers. An org of 400 probably has a couple, at some point the law of large numbers does really start to apply, and you're no longer in competition with the people you're working with on the daily. If one member of your two-pizza team is going to get fired each year, it makes sense to throw someone under the bus on the reg. If one member of your 100 person org is going to get fired each year, that strategy doesn't make sense.
  - evgen 4y ago
    Having endured several such calibration sessions at FB/Meta I can also assure you that we are not even talking about 'large' numbers before there is a strong expectation that the ranking totals within an org will meet the predicted curve when it comes to the percentage at exceeds, meets, and meets most. I distinctly remember wasting more than 15 minutes of my life arguing with a skip level manager that someone who had course corrected and delivered solid work while digging themselves out of a PIP was at least a meets before eventually realizing this manager had a quota of meets most to fill and the person we were discussing was destined for that bucket just to make the numbers look as expected.
    - dustingetz 4y ago
      as a manager what do you say there to the report
      - fn-mote 4y ago
        Polish your resume, the guillotine is coming and there's nothing I can do to save you. I tried for you, remember me when your next job is hiring managers.
        disgruntledphd2 4y ago
        Or if you know a team is hiring internally then push them towards that. Ideally you'd do this before calibration so the rating doesn't mess it up.
- epolanski 4y ago
  In other words, it's like if you get rid of the worst player in each NBA team. They are still super atheletes, but they are judged against other NBA teammates, not average professional basketballer in the world.
- UncleMeat 4y ago
  Sort of. I'm at Google, not Meta, but the same "we have an expected distribution of ratings" thing exists here and it causes the same confusion. With large enough populations you see fairly consistent statistical information across time. Little fluctuations even out. So at org sizes of, say, 500 people you'll see pretty consistent ratings distributions between cycles. This is where you get the "we have a roughly expected ratings distribution." If you've deviated far from this then it means either that your org is training people on the evaluation process incorrectly or there is some other cause of a weird distribution. When something doesn't match, people simply investigate for an alternative solution. What people seem to be frightened of is when somebody says to a manager of 10 people "hey you've got too many high ratings, drop some." Google and Meta are massive companies so I suspect that this happens somewhere but I've never even got the scent of this happening from my position at Google. I've had cycles where 100% of my team was in the top 20% of ratings without anybody raising an eyebrow.
  - rnk 4y ago
    It used to happen at Microsoft back in the day. I was in SQL Server and we openly talked about this happening. They literally told us being in a great, high functioning org like sql server had a certain penalty. Then top management said don't worry, we got extra money for our people. It was kind of obviously silly. On stack ranking & the penalty for being in a great team - yes, also happened, being in a great team but you still had to rank someone lower, better to be in a worse team to get higher ratings. It had obviously negative impact, but there was some combination or organizational pride that had been cultivated, plus we thought we'd all be millionaires anyway so what did it hurt. But I bet some people had to leave who were the unfortunate victim of these affects.

nightpool 4y ago

The key component of stack ranking / sorting by performance is that there's no absolute measure of your contribution, only a relative measure by which you have to outcompete everyone else in your team/department. If you fire your lowest 15% of engineers—no matter how good they are on an absolute level—that's a stack ranking system. It's a horrifying Red Queen's Race that punishes high-quality teams. I'm not necessarily saying that Meta's system is good, but at least it's not that. EDIT: It sounds like at least some people are claiming that Meta has an internal policy that X% of your engineers under you are supposed to be given "Meets Most Expectations" ratings, which would also be pretty close to stack ranking (if at least allowing for some management leniency) https://news.ycombinator.com/item?id=32930370 https://news.ycombinator.com/item?id=32930370

joshuamorton 4y ago
The other key component of stack ranking was that it's on a per team basis. A team of 20 people may have no low performers. An org of 400 probably has a couple, at some point the law of large numbers does really start to apply, and you're no longer in competition with the people you're working with on the daily. If one member of your two-pizza team is going to get fired each year, it makes sense to throw someone under the bus on the reg. If one member of your 100 person org is going to get fired each year, that strategy doesn't make sense.
- evgen 4y ago
  Having endured several such calibration sessions at FB/Meta I can also assure you that we are not even talking about 'large' numbers before there is a strong expectation that the ranking totals within an org will meet the predicted curve when it comes to the percentage at exceeds, meets, and meets most. I distinctly remember wasting more than 15 minutes of my life arguing with a skip level manager that someone who had course corrected and delivered solid work while digging themselves out of a PIP was at least a meets before eventually realizing this manager had a quota of meets most to fill and the person we were discussing was destined for that bucket just to make the numbers look as expected.
  - dustingetz 4y ago
    as a manager what do you say there to the report
    - fn-mote 4y ago
      Polish your resume, the guillotine is coming and there's nothing I can do to save you. I tried for you, remember me when your next job is hiring managers.
      - disgruntledphd2 4y ago
        Or if you know a team is hiring internally then push them towards that. Ideally you'd do this before calibration so the rating doesn't mess it up.
epolanski 4y ago
In other words, it's like if you get rid of the worst player in each NBA team. They are still super atheletes, but they are judged against other NBA teammates, not average professional basketballer in the world.
UncleMeat 4y ago
Sort of. I'm at Google, not Meta, but the same "we have an expected distribution of ratings" thing exists here and it causes the same confusion. With large enough populations you see fairly consistent statistical information across time. Little fluctuations even out. So at org sizes of, say, 500 people you'll see pretty consistent ratings distributions between cycles. This is where you get the "we have a roughly expected ratings distribution." If you've deviated far from this then it means either that your org is training people on the evaluation process incorrectly or there is some other cause of a weird distribution. When something doesn't match, people simply investigate for an alternative solution. What people seem to be frightened of is when somebody says to a manager of 10 people "hey you've got too many high ratings, drop some." Google and Meta are massive companies so I suspect that this happens somewhere but I've never even got the scent of this happening from my position at Google. I've had cycles where 100% of my team was in the top 20% of ratings without anybody raising an eyebrow.
- rnk 4y ago
  It used to happen at Microsoft back in the day. I was in SQL Server and we openly talked about this happening. They literally told us being in a great, high functioning org like sql server had a certain penalty. Then top management said don't worry, we got extra money for our people. It was kind of obviously silly. On stack ranking & the penalty for being in a great team - yes, also happened, being in a great team but you still had to rank someone lower, better to be in a worse team to get higher ratings. It had obviously negative impact, but there was some combination or organizational pride that had been cultivated, plus we thought we'd all be millionaires anyway so what did it hurt. But I bet some people had to leave who were the unfortunate victim of these affects.