The work came back wrong again. You have drafted the message twice and deleted it twice. You have told yourself you will raise it at the next one-to-one, and then you did not, because the moment never felt right.
Three months later you are having a much larger conversation than the one you avoided, and the person on the other side of it is blindsided, because from where they sat nothing was wrong until suddenly everything was.
Why you are avoiding it
This has a name and it is old. Sidney Rosen and Abraham Tesser (1970) documented what they called the MUM effect: people are reliably reluctant to transmit undesirable information, even when transmitting it costs them nothing and the recipient needs it.
Avoidance is not kindness. It is the MUM effect wearing kindness as a disguise.
Recognising it as a near-universal default rather than a personal failing is the first move, because it stops you treating the reluctance as evidence that the conversation is a bad idea.
The move: Name the Problem, Hold the Person
Both halves are required. Full honesty about the problem, and active protection of the person's standing so they can actually receive it.
Method at a glance
Name the Problem, Hold the Person
- Have the conversation, and have it promptly. After the second slipped deliverable, not the tenth.
- Aim the message at the work, not the person. Describe what they did, never what they are.
- Hold the person while you deliver the problem. Pair the criticism with a credible signal of high standards and real belief in their capability.
- Calibrate the weight. It is landing harder than you think. One clear message, then stop.
Two. Work, not person
Karen Jehn (1995) drew the distinction that makes this actionable. Task conflict, disagreement about the work itself, behaves very differently from relationship conflict, which is personal and identity-level. "These three deliverables had errors" is task conflict, addressable and non-threatening. "You're unreliable" is relationship conflict, a character verdict that triggers defence.
The test is whether your sentence describes something the person did or something the person is.
This gets botched because character judgments feel more honest. When someone's work has slipped repeatedly, "you're unreliable" feels like the deeper truth and naming only the errors can feel like pulling the punch. It is not. The specific errors are the actionable truth. The character verdict is an inference layered on top, and it is one they can dispute endlessly while the errors are simply facts.
There is a quieter reason it matters. Character verdicts tend to become self-fulfilling. Tell a capable person they are unreliable often enough and you erode the identity that was driving their good work.
Three. Hold the person
This is the move most people skip, and it is the one with the cleanest evidence behind it.
David Yeager and colleagues (2014) tested what has become known as wise feedback. Attaching a short note to critical feedback, stating that the standard is high and that the reviewer believes the person can meet it, substantially changed how the same criticism was received and acted upon. The criticism did not soften. What changed was whether it was read as an attack or as help.
David Sherman, Leif Nelson and Claude Steele (2000) showed the mechanism from the other side: affirming someone's broader sense of self makes threatening information markedly easier to accept rather than defend against.
So securing their standing is not padding. It is creating the conditions under which the unsoftened truth can land.
Four. It lands heavier than you think
Roy Baumeister and colleagues (2001) assembled the evidence for a broad asymmetry: bad is stronger than good, across a wide range of psychological domains. Negative information carries disproportionate weight.
Two things follow. Deliver one clear message rather than a pile, because each additional criticism arrives with that multiplied weight. And do not read a strong reaction to a single piece of feedback as evidence the person is oversensitive. Their reaction is calibrated to the real weight, not to the weight you intended.
Say the hard thing once, clearly, then stop. It does not need driving home.
Where this shows up outside work
A friend's business plan. A relative's drinking. A partner's habit that has stopped being funny. The MUM effect does not respect the boundary between professional and personal, and neither does the fix.
Where this tips into something worse
The guardrail sits on move three, because holding the person is the easiest part to fake.
The honest version requires that the high standards be real and the belief genuine. The manipulative version manufactures them as a delivery mechanism: warmth deployed to soften someone before a verdict that was already decided, the classic sandwich where the praise is packaging. Manufactured warmth gets detected eventually, and once someone works out that your expressions of belief were a tactic, every future conversation is poisoned.
The second corruption runs the other way. Holding the person must never become softening the truth until it is no longer delivered, which is the MUM effect dressed as compassion.
One question tests the whole method: are you making the truth easier to receive, or easier to avoid?
Three things to fix before the conversation
The pile
Write down everything you were going to raise, then cut it to one. The rest can wait for a conversation that is not this one.
Every sentence containing "you are"
Rewrite each as something observable that happened. If you cannot, you are working from an inference rather than a fact.
The opening
If it does not contain a true statement about your standard and your belief in them, it is not held yet, and the criticism will land as a verdict.
Where to start
Pick the conversation you have already postponed twice. Write one sentence naming the specific behaviour, and one sentence naming your standard and your belief that they can meet it.
Then have it this week, while the thing you are describing is still small enough to be about itself.
Common questions
How do I give hard feedback without demotivating someone?
Pair it with a credible statement of high standards and genuine belief they can meet them. In studies of this approach, the same critical feedback produced markedly different responses depending on whether that assurance was attached (Yeager et al., 2014). The criticism does not need softening. The person needs holding.
Is the feedback sandwich a good idea?
Not as usually practised. If the praise is packaging rather than meant, it gets detected, and once someone works out the warmth was technique every later conversation is compromised. Real belief stated plainly works. Manufactured belief used as a delivery vehicle does not.
Should I raise several problems at once?
No. Negative information carries disproportionate weight (Baumeister et al., 2001), so each additional criticism lands heavier than intended and a pile becomes crushing quickly. Deliver one clear message, then stop.
Why do I keep putting the conversation off?
Because reluctance to transmit unwelcome information is a documented and near-universal default, known as the MUM effect (Rosen & Tesser, 1970). Treating the reluctance as a signal that the conversation is a bad idea is the mistake. Every delay makes the eventual version heavier.
Sources
Rosen, S., & Tesser, A. (1970). On reluctance to communicate undesirable information: The MUM effect. Sociometry, 33(3), 253–263. · Jehn, K. A. (1995). A multimethod examination of the benefits and detriments of intragroup conflict. Administrative Science Quarterly, 40(2), 256–282. · Yeager, D. S., Purdie-Vaughns, V., Garcia, J., Apfel, N., Brzustoski, P., Master, A., Hessert, W. T., Williams, M. E., & Cohen, G. L. (2014). Breaking the cycle of mistrust. Journal of Experimental Psychology: General. · Sherman, D. A. K., Nelson, L. D., & Steele, C. M. (2000). Do messages about health risks threaten the self? Personality and Social Psychology Bulletin. · Baumeister, R. F., Bratslavsky, E., Finkenauer, C., & Vohs, K. D. (2001). Bad is stronger than good. Review of General Psychology, 5(4), 323–370.