Wharton management professor Peter Cappelli has spent decades studying the complicated dynamics of employment. In a post-recession world, his research is more timely than ever, as companies large and small struggle to adapt to a new normal that relies on fewer employees handling a larger, shifting workload. One practice that has persisted in this changing business landscape is the performance evaluation, which Cappelli describes as universally despised by both supervisors and subordinates.

In their latest research, Cappelli and Martin Conyon, a professor at Bentley University and a senior fellow at Wharton, question the usefulness and accuracy of performance appraisals and find some surprising answers. Cappelli, who is also director of Wharton’s Center for Human Resources, discussed their findings in a recent segment on the Knowledge@Wharton show on Wharton Business Radio on SiriusXM. The following is an edited transcript of the conversation.

Knowledge@Wharton: Is the performance appraisal as important now as it was 20 years ago?

Peter Cappelli: It’s more important in the sense that more people have to do it. If you look around the United States, there haven’t been a lot of recent studies. But the last ones show that more than 90 percent of the workforce has a performance appraisal. I think the big change is that it used to be kind of a U.S. thing, but now you see them all around the world.

K@W: Your work with Martin Conyon on this paper is trying to bring together information about performance appraisals from a variety of sources?

Cappelli: The thing about performance appraisals is, they’re ubiquitous. There’s probably nothing in management that’s more common. And there’s also almost no practice in business that people hate more. The evidence on this is pretty overwhelming.

It’s also surprising how little we actually know about it. How do job appraisals actually work inside companies? Quite remarkably, almost nobody has looked at this. We got data from a large Fortune 50 company on all their performance appraisals over a 10-year period. There’s a view among a lot of executives that employment is like a contract. At the beginning of the year, you set goals; then we assess how well you’ve done. At the end of the year, we give you a pay increase based on how many of your goals you’ve met and how well you’ve done, maybe, compared to everybody else.

But there’s another view that it’s not like a contract—that it’s really kind of a relationship. If you think about employment, you don’t really have a contract with your boss. The boss is telling you to do different things all the time. Your circumstances are unpredictable, too. It could be, “We’ve got this goal.” But then business collapses, and we change the goal. Or we have to adjust the target.

One of the questions we looked at was to what extent is a performance appraisal a contract, and to what extent is it a relationship used to encourage you. We also wanted to see if people who perform well always perform well. And people who perform poorly, do they always perform poorly? This matters because there is a very prominent theory in management—something that Jack Welch made famous—about the A-player, B-player, C-player model. The folks at McKinsey & Co. made a similar case that there are really good executives and there are lousy ones. If you believe that, you want to hire the good ones and get rid of the bad ones. If that’s the story, then management’s kind of simple, right?

As far as we can tell, no one has ever looked at this before, or at least published it. Are the people who do well always doing well, or not? If we know your scores this year for everybody in the company, how much of next year’s score could we predict or explain? If the good people are always good and the bad people are always bad, we can explain 100 percent of your scores, because next year’s score will be identical to this year’s score. If it’s random, which would be kind of astonishing, then it would be zero—no relationship between how people on average perform this year and how they perform next year. The good people could be good or bad; the bad people could be good or bad.

K@W: But you’d think they would follow a pattern. If you’re good in 2014, unless something has drastically changed, you’ll be pretty good in 2015 as well.

Cappelli: Right. It’s between zero and 100 percent. If you think this A-player, B-player, C-player model is right, it’s going to be closer to 100. If you think it’s all just random or people vary a lot, you’re closer to zero. So that’s the question.

K@W: I’m going to say it would probably be closer to 70 or 75 percent.

Cappelli: That’s a very common answer. People in human resources guess 80 percent. The correct answer is 27 percent, so it’s way closer to zero than it is to 100 percent.

K@W: Why so much lower? I would think it would be on the higher end.

Cappelli: Many people seem to believe that, especially in human resources. But when I ask them if they’ve ever actually looked at it, the answer is no. Maybe they assume it’s that way because that’s what you hear from the A-player, B-player, C-player story, and you could see some of this is a cognitive bias. There’s something in psychology known as the fundamental attribution error: When we see somebody behave in a particular way, we’re inclined to assume it’s because of who they are rather than the circumstances. The classic example is somebody racing by you on the road going home. They’re driving on the shoulder and your inclination is to say, “That guy’s a jerk,” rather than to entertain the idea that maybe it’s an emergency. We seem to be wired to think everything is due to the person. If you believe that, then you would be inclined to think the A-player, B-player, C-player model is right and good players this year are going to be good players next year.

The other thing we looked at was whether it actually changed your appraisal scores when you got a new supervisor, because the other view is that once you get comfy with a particular supervisor, your scores are sort of the same. You get a new supervisor, and they can really sort out whether you’re good or bad. Well, we didn’t see that, either.

It turns out there’s a lot of variation in how people perform. This calls into question forced ranking systems—they call them “rank and yank” or “rack and whack.” General Electric used to force out the bottom 10 percent because they believed in the A-player, B-player, C-player model. If your company’s doing that, you might want to actually look to see whether it’s true that your bottom 10 percent this year are the same as your bottom 10 percent next year. The problem is, if you keep firing your bottom 10 percent, you’re never going to know, because you’ll never know what those guys would have done. But you could at least look at the appraisal scores for everybody else and see whether they remain constant over time. If they’re bouncing around, it’s insane to fire the bottom 10 percent, because there’s no reason to think those guys are going to be bad next year.

K@W: In certain situations, the boss is asking the employee questions such as, “What kind of job did I do over the course of the year?”

Cappelli: 360-degree feedback is the formal way in which that gets done, where you ask people all the way around you, “How do you think I did as a boss?” That has not had a terrific track record, partly because there’s a lot of venting going on. If you’re a subordinate, it’s hard to be objective about your boss. It’s hard, always, to like your boss.

Let me tell you the punch line of what we found on the academic side: Things don’t look very much like a contract, and supervisors tend to reward people for improvements as well as the level of performance. Also, counter to the prevailing view, companies over-reward high performers. It’s not a linear relationship. If you’re a poor performer, they whack your merit pay increases. And if you’re a better performer, they load them up.

K@W: That increases the separation between the upper end and the lower end in the course of the job.

Cappelli: Within the job, right. And it’s true that as you move up the organization, the average scores increase. One explanation is that you’re selecting better people if it’s promotion from within, so it’s not surprising that the scores would go up. And is it bias once you get near the top? They say that CEOs always give their personal assistants the top score.

Published as “The Problem With Job Performance Appraisals” in the Fall 2016 issue of Wharton Magazine.