More than a third of feedback interventions make the recipient worse. Here is why.

Jonas reads the summary twice. Then he reads the method section. Then he counts how many people responded, works out who they might be, and concludes that two of them have reasons to be dissatisfied that have nothing to do with his leadership. The report goes into a folder. Six months from now he will not remember exactly what it said.

Nothing about this sequence is unusual. According to the research, it is the expected outcome.

The two previous articles in this series built an argument in two steps. We cannot see ourselves from the outside, and the gap between how we think we act and how we actually act is invisible from the inside. The conclusion has been that self-awareness requires feedback from others. This article is about the uncomfortable next step: feedback from others often works badly too. And the research knows fairly precisely when, and why.

The number that should worry every HR department

In 1996 Avraham Kluger and Angelo DeNisi published what is still the most comprehensive review of the effects of feedback. They gathered 607 effect sizes from 131 publications, based on 12,652 participants, and asked the simplest possible question: do people perform better on a task after receiving feedback?

The average answer was yes. The combined effect was moderately positive (Cohen's d = 0.41). But the average concealed the finding that made the study a classic: just over 38 percent of the measured effects were negative. In more than one feedback intervention in three, performance fell rather than rose.

This was not a matter of odd experiments or obviously botched interventions, and it was not about 360 instruments or leadership feedback in particular. It concerned feedback on task performance across a wide range of settings: school, working life, laboratory. The number does not mean that 38 percent of everyone who receives feedback gets worse, or that one manager conversation in three is harmful. It means something more fundamental: feedback does not improve performance on its own. Whether it does depends on how it is designed.

The study has been cited thousands of times and the main result has never been seriously challenged. Yet most organizations still build their feedback culture on the assumption that more feedback is better.

What decides it: where the attention goes

Kluger and DeNisi did not stop at documenting the variation. They tried to explain it, and the explanation is still the most useful model for understanding why feedback succeeds or fails.

Feedback, they argued, directs the recipient's attention to one of three levels:

1. The task, how something concrete was carried out and how it could be done differently.
2. The motivation for the task, effort, goals, persistence.
3. The self, who I am, what I am worth, how others see me as a person.

The closer to the self the feedback lands, the worse the effect. Feedback that directed attention to the self-image (praise, criticism of the person, anything that threatened self-esteem) had on average a low or negative effect. Feedback that stayed with the task, especially when it carried information about how to do better, had the strongest positive effect.

Praise is the surprising example. Intuitively, positive feedback ought to be harmless. But praise by definition directs attention to the person rather than the action, and in the data praise was associated with lower effect, not higher. John Hattie and Helen Timperley reached the same conclusion a decade later in their review of feedback in education: feedback at the "self level" was the least effective of four levels, whether it was positive or negative.

As soon as the feedback is about who you are rather than what you did, it stops working as information and starts working as a threat.

The link to Argyris: the defenses kick in

This is where the threads tie back to the previous article.

Chris Argyris described how competent, successful people react when something challenges their picture of themselves: they explain it away, reinterpret it, question the source. Not out of malice, but reflexively. He called them defensive routines, and he showed they are strongest in those who have rarely failed.

Kluger and DeNisi's result is the same mechanism seen from the other side. Feedback that threatens the self-image activates the defenses, and while the defenses are up we learn nothing. Jonas did exactly what Argyris predicts: he began by examining the method, went on to disqualify the senders, and ended by putting the report away. At every step his espoused theory was intact, he is a manager who wants honest feedback. His theory-in-use was to neutralize it.

The decisive point is that the defenses need something to push against. A judgement about the person ("unavailable") can be dismissed, because it is an interpretation. A description of a behavior in a specific situation ("often interrupts before the speaker has finished making their point") is considerably harder to explain away. It can still be uncomfortable, but it leaves the self-image alone and directs attention where Kluger and DeNisi showed it has to be: at what actually happens.

Why 360-degree feedback so often becomes a paper product

If external feedback is the answer to the problem with self-assessment, 360-degree instruments ought to be the gold standard. The research is less enthusiastic.

In 2005 James Smither, Manuel London and Richard Reilly reviewed 24 longitudinal studies of multisource feedback, that is, feedback from manager, peers and direct reports at the same time. Their conclusion: improvements over time were on average small. Not absent, but small, and widely spread. Some recipients changed markedly. Many did not change at all.

A few years earlier, Joan Brett and Leanne Atwater had examined how recipients actually react. The result was uncomfortable but logical: those rated lower by others than they had rated themselves, that is, those with the largest gap between self-image and the view from outside, reacted with anger and discouragement, judged the feedback as less accurate and saw it as less useful. In other words: those with most to learn were least inclined to take it in.

In a follow-up article DeNisi and Kluger identified the most common design faults. Tying a 360 instrument to pay or promotion turns feedback into appraisal, which moves attention to the self. Questions phrased as traits rather than behaviors give the defenses something to push against. And reports delivered without follow-up become, as with Jonas, a folder in a desk drawer.

The problem, then, is not that feedback from several sources is the wrong idea. The problem is that the instrument is often built in a way that triggers precisely the reactions that render feedback useless.

Five things that set working feedback apart

Read the research together and a pattern emerges. Feedback that actually changes behavior has a few traits in common.

1. It is about behavior, not about the person. This is the most robust finding in the whole literature. Descriptions of what someone does in concrete situations can be received. Judgements about who someone is trigger defenses.

2. It is descriptive, not evaluative. Feedback that ranks (better, worse, above, below) shifts the focus from information to status. Feedback that describes how a behavior is perceived, without grading it, lets the recipient decide what to do with the information.

3. It comes from several observers. One person's experience is an opinion. When five or eight people independently describe the same pattern, it is a fact about how you are perceived, whatever you make of it yourself. That is also when the defenses have the hardest time finding a single sender to disqualify. Simine Vazire's research on what others know about us that we do not concerns observable behavior precisely, and that is exactly what several observers can describe.

4. It is set against the self-image. A pattern recurs in Smither and colleagues' review: those who rated themselves higher than those around them did are the group that changes most over time, provided they engage with the result. The gap is not a problem to be smoothed over. It is where the learning is.

5. It is followed up in conversation. Alan Walker and James Smither followed 252 managers across five annual measurements and found that those who sat down with their people and discussed their feedback improved more than those who only read the report. What the recipient did with the feedback mattered more than what it contained.

None of this is technically complicated. But it is rarely how feedback is actually given.

How the IDI method is built

IDI was developed from a simple starting point: what matters is not how you describe yourself, but how others experience your behavior. In practice the method is an application of the points above.

An IDI profile is built from the answers of people who work closely with you. The questions concern observable behavior, how you act in interaction with others, not traits or personality. The result describes your behavioral style as those around you perceive it and sets it against your own rating. The gap between them is not an error in the measurement. It is the measurement's main result.

No behavioral style in IDI is better than another. The profile does not sort people. It describes a pattern in how you are perceived, and how well you adapt it to the situation and the person in front of you. That is why the method can deliver clear answers without activating the defenses Argyris described and Kluger and DeNisi measured.

The profile is not delivered as a report in a folder. The feedback happens in conversation with a certified IDI facilitator, whose job is to help the recipient understand what those around them see, and what can be done with that insight.

Jonas would probably have reacted differently if the report had not called him unavailable but described how often he is perceived to interrupt, and if someone had been sitting beside him as he read it.

Feedback is a method, not a gift

There is a popular saying that feedback is a gift. The research suggests that is an unfortunate image. Gifts are received politely and put on a shelf. Feedback that works is closer to an instrument, and like every instrument it only works if it is built correctly and used correctly.

We cannot see ourselves. Others can. But others seeing is not enough. What they see has to be described in a way that can be received, set against what we believe ourselves, and followed up by someone who knows what the next step is.

That is the difference between feedback that changes behavior and feedback that ends up in a folder.