I once rated myself 3 out of 5 for SQL in a job interview. We were in a Caffe Nero in a train station, which felt appropriately chaotic for the situation. I asked the interviewer what a 5 would look like. I could actually do everything they described. Turns out I was a 5. A mix of Irish modesty and being fairly junior meant I had no real frame of reference for what good actually looked like.
This is the core problem with self-assessment. It's not that people are dishonest. It's that the number is almost meaningless without context.
We Compare Ourselves To The Wrong Things
When someone rates their own skills, they're not measuring against an objective standard. They're measuring against whatever they happen to have been exposed to. Someone who has only ever worked in small teams might rate their project management skills highly, because relative to the people around them, they're good. Put them in a different context and the number shifts entirely.
This is what researchers call the big fish little pond effect. First described by Marsh and Parker in 1984, the idea is that our self-concept is shaped not just by our ability but by our comparison group. A strong student in a weaker school will rate themselves higher than an equally strong student in a stronger school. The skill is the same. The perception isn't.
It applies directly to the workplace. Someone who has always been the most technically capable person in their team will rate themselves differently to someone with identical skills who works alongside people more experienced than them. Neither number is wrong exactly. But neither number is particularly useful either.
The More You Know, The More You Know You Don't Know
There's a pattern that shows up consistently in research on self-assessment. People with less knowledge in a domain tend to overestimate their ability. People with more expertise tend to underestimate it.
Kruger and Dunning's 1999 paper is the most cited starting point - the finding that the skills needed to perform well at something are often the same skills needed to recognise that you're performing poorly. If you don't know enough to know what good looks like, you can't accurately judge where you sit.
It's worth being honest that this research has been debated since - some argue parts of the effect are statistical rather than purely cognitive. But the practical reality holds up: the most junior person in a room is often the most confident about their abilities, and the most senior person is often the most uncertain. My Caffe Nero moment was a good example of this working in reverse - I knew enough to know I wasn't sure, but not enough to know I actually was.
Thinking About Thinking Is A Skill In Itself
Metacognition - the ability to think accurately about your own thinking and learning - develops over time and with experience. It's not something people are born with, and it's not evenly distributed.
Research on metacognitive accuracy consistently finds a gap between how well people think they understand something and how well they actually do. A 2024 study in engineering education found a significant mismatch between students' self-assessed cognitive skills and their actual performance - and crucially, this mismatch was largest among those who had the weakest actual skills.
What this means practically is that the people whose self-assessments you most need to be accurate - the ones who might be struggling, the ones who need support - are the ones whose self-assessments are least likely to be reliable. The people who are doing fine will probably tell you they're doing fine. The people who aren't may genuinely not know.
So What Do You Do With This?
None of this means self-assessment is useless. Asking someone to reflect on what they know, where they feel confident, and where they feel uncertain is a useful starting point for a learning conversation. It surfaces what someone thinks, which matters even if it isn't accurate.
The problem is treating the number as the answer. When a survey score becomes a training record. When a 1-5 rating becomes evidence of competence.
A few things that actually help:
- Give people a frame of reference. If you're going to ask someone to rate themselves, show them what each level looks like first. Not a vague descriptor - specific examples. What does a 3 actually do? What does a 5 do that a 3 doesn't? The number becomes more useful when people have something concrete to calibrate against.
- Ask questions instead of ratings. Asking someone to describe a situation where they applied a skill, or explain what they'd do in a given scenario, tells you more than a number. It's harder to answer and harder to game.
- Have a conversation. When you can ask a follow-up question - when you can push on something that sounds thin - you get much closer to what someone actually understands. A survey can't do that.
The number someone gives you tells you how they see themselves relative to their own frame of reference. That's useful context. It's just not the same thing as knowing what they can do.