Confidence Is Not Competence, and Our Own Study Says So

The numbers, in full

Fifty-seven participants learned the same feedback model, Situation, Behaviour, Impact, in one of three ways: an instructional slideshow, an instructional video, or an online coaching session with a role-play coach, each format capped at 20 minutes. Each then held a simulated feedback conversation, which two independent assessors scored blind on a 1 to 7 scale. Confidence was self-rated on a 1 to 7 scale at three points: before the intervention, straight after it, and after the simulation. The full method and its limits are on our research page.

Every group climbed at every step. The format had no significant effect on confidence (p = .958), and the interaction between time and format was not significant either (p = .541). Read that carefully: it is an absence of evidence for a difference between the formats, not evidence that they are the same. We will come back to that distinction, because we got it wrong ourselves.

Now the other measure. Blind-rated delivery averaged 5.25 out of 7 for the rehearsal group, 4.26 for slides and 3.82 for video. Rehearsal was significantly better than both passive formats, with Cohen's d of 0.75 against slides and 1.08 against video, and slides and video did not differ from each other. Put the two rows side by side. Final confidence sat between 5.38 and 5.53 across the three groups. Rated delivery sat between 3.82 and 5.25. The people who had watched a video reported almost the same final confidence as the people who had rehearsed, and were rated a long way behind them.

The study also checked whether a participant's final confidence rating predicted their assessed performance. It did not. The paper discusses this gap between confidence and performance through the lens of the Dunning-Kruger effect, but cautions that, without a control condition, the effect does not fully apply here. The study did not rely on the question "how confident are you?" alone: it asked it at three points and had two independent assessors score a recording of each conversation.

• Slideshow: 4.46 before the intervention, 5.00 after it, 5.39 after the simulation.
• Video: 4.69, 5.13, 5.38.
• Rehearsal with a role-play coach: 4.27, 5.33, 5.53.
• All self-rated, 1 to 7. The differences between the three groups were not significant. Source: Laumann (2020), UCL, unpublished research paper, Figure 4.

Why the form still gets sent

Confidence is cheap to collect and it goes up. That is the whole problem. The TalentLMS 2026 L&D Benchmark Report, a September 2025 survey of 101 HR managers in the United States, finds business impact selected by 37% of them as a measure of L&D success, and satisfaction with training by 28%. The Department for Education's Employer Skills Survey 2024 puts UK employer training spend at £1,700 per employee, £53.0bn in total. A confidence score is a thin receipt for that.

Even the framework most evaluation forms claim to follow does not treat confidence as behaviour. Kirkpatrick Partners define Level 2, Learning, as the degree to which participants acquire "the intended knowledge, skills, attitude, confidence, and commitment" they need, and Level 3, Behaviour, as the degree to which they actually perform the critical behaviours in their own environment. Confidence is filed under Level 2 by the people who maintain the model. A confidence question is a Level 2 instrument at best, and in our study a participant's final confidence rating did not predict how their simulated conversation was scored. Our explainer on the Kirkpatrick model sets out the four levels.

What the study cannot carry

This is one study of 57 people, run in 2020 for a taught module at UCL, unpublished and not replicated. It has no control condition, so it cannot say how much anyone improved, and no measurement of ability before the intervention, only of confidence. The practice condition was a short online coaching session, not our programme, and we do not present it as a measurement of what we deliver. The delivery score comes from one simulated conversation held immediately afterwards, which is not the same thing as behaviour at work a month later. What the study can say is narrow and, within that, clear: three formats produced confidence scores with no significant difference between them, and delivery scores that differed significantly.

We have had to correct our own wording on this. Earlier drafts of our research page described the rise in confidence as identical across the groups, and a later draft called it equal. Both overstate it. A non-significant difference is a failure to find a difference, not proof of sameness, and the phrasing now on the page is "with no significant difference between them". If we want buyers to read evaluation data carefully, we have to read our own the same way.

What we record instead

In our new manager programme the fundamentals are watched while people work, one manager at a time, rather than collected on a satisfaction form at the end. Each one-to-one conversation with an actor is played, debriefed, rewound to the moment that mattered and played again, and only the second attempt shows whether the manager can do something other than their default when the pressure comes back. The questions are behavioural: did the manager notice their own reaction and slow down, did they make it safe for the actor to push back, did they leave with an owner and a date. None of those is a feeling.

Being straight about the limit of that too: what someone does in the room under pressure is still not what they do on an ordinary Tuesday. That is Kirkpatrick Level 3, behaviour on the job, and it is a separate piece of measurement, set out in our guide to measuring behaviour change. What we do not do is ask people how confident they feel and call the answer evidence. Our own study says that answer went up in every condition, including the two whose delivery was rated significantly weaker.

Sources: Sidestream, The Study Behind the Method · Kirkpatrick Partners, The Kirkpatrick Model · TalentLMS, The 2026 Annual L&D Benchmark Report (September 2025) · Department for Education, Employer Skills Survey 2024