Self-Efficacy: The Four Sources, Ranked by the Evidence

Key takeaways

Self-efficacy is deliberately narrow — a judgement about a specific performance, which makes it measurable. “I self-sabotage” is unfalsifiable by construction: succeed and it was luck, fail and the belief was right.
The belief does predict performance: across 114 studies and 21,616 participants, the weighted average correlation with work performance is 0.38.
But that is a between-person figure. Within-person — the test that matters for one person trying to change — a meta-analysis found it weak or absent, and warned prior performance may confound the whole literature.
A cross-lagged meta-analysis in education found the loop runs both ways, with which direction looked stronger depending on measurement order. Reciprocity appeared in adults and not in children.
Bandura ranked the sources of self-efficacy: enactive mastery first, vicarious experience second, verbal persuasion explicitly weaker. Writing a belief down, disputing it and replacing it is verbal persuasion from start to finish.

The method is always the same. Identify the belief holding you back. Write it down. Find where it came from. Challenge it. Replace it with a better one. Repeat until the belief changes and the life follows.

“Limiting belief” is not a research term and there is no literature under that name. There is a large literature on the thing it is gesturing at — and its most consistent finding is that you do not get at it that way round.

What has actually been measured

The construct is self-efficacy: your belief that you can execute a particular course of action. Not confidence, not self-esteem, not a global sense of your own worth. Bandura’s 1977 formulation is deliberately narrow — it is domain-specific, task-specific, and it can be high in one area and low in the next without contradiction.

That specificity matters immediately, because “I am not the kind of person who succeeds” is not a self-efficacy belief. It is too big to be measured and too big to be tested, which is part of why it survives.

The construct predicts a lot. A meta-analysis of 114 studies, 157 correlations and 21,616 participants found a weighted average correlation between self-efficacy and work-related performance of 0.38 — a substantial relationship by the standards of this field.

So the belief matters. The question the advice never asks is which way the arrow points.

Why the size of the belief is the first problem

Before the direction of causation, there is a problem of scale that the method creates for itself.

Self-efficacy is measured as a judgement about a specific performance: can you make this call, run this distance, sit this exam. It is deliberately bounded, because a bounded claim can be checked against what happens next.

The beliefs the coaching exercise surfaces are the opposite. “I self-sabotage.” “I don’t deserve success.” “I’m not a leader.” These are unfalsifiable by construction — no single outcome can confirm or refute them, because any evidence can be absorbed. Succeed and it was luck or an easy case; fail and the belief was right all along.

Which means the exercise has taken a construct that works because it is specific and applied its language to statements that survive precisely because they are not. That is a category problem, and it happens before any question about what causes what.

The direction of the arrow

Between people, high self-efficacy goes with high performance. That is the 0.38, and it is what makes the coaching case.

Within a single person over time, the picture changes.

Weak to null

A meta-analysis of the within-person self-efficacy domain found the relationship between a person’s self-efficacy and their own subsequent performance to be weak or absent — and warned that increases in self-efficacy caused by prior improvements in performance may confound how the whole relationship has been interpreted.

Read that carefully, because it is the entire argument. Comparing people, believers do better. Following one person, believing more this week does not reliably predict doing better next week — and much of what looks like belief driving performance may be performance having already driven belief.

A cross-lagged meta-analysis in education found the loop genuinely runs both ways, with a wrinkle that should make everyone cautious: which direction looked stronger depended on the order in which the researchers happened to measure the two variables. Reciprocity showed up in adults and not in children.

What this does and does not overturn

It does not show that belief is irrelevant. The between-person association is real and large, the reciprocal path exists, and nobody in this literature argues that confidence does nothing.

What it undermines is the specific causal story the advice depends on: that changing the belief first is how you change the behaviour. The within-person evidence is the test of that claim, and it is the evidence that comes back weakest.

Bandura already ranked the ways to change it

The most useful part of this is nearly fifty years old and is skipped by almost everything written on the subject.

The 1977 paper does not only define self-efficacy. It sets out where self-efficacy comes from, in order of strength:

Enactive mastery — having actually done the thing, or something like it. The strongest source by a distance, because it is based on personal experience of succeeding rather than on inference.
Vicarious experience — watching someone comparable to you do it. Weaker, and it depends heavily on how similar you judge them to be.
Verbal persuasion — being told, by yourself or anyone else, that you can. Explicitly weaker still, because it is not grounded in an authentic experiential base.
Physiological states — how aroused, tense or exhausted you feel, read as evidence about your capability.

Now look at what the limiting-beliefs method actually consists of. Writing the belief down, tracing its origin, disputing it, replacing it with an affirming alternative: that is verbal persuasion, from start to finish. It is the technique the founding framework ranks third of four.

What the ranking implies you should do instead

If mastery is the strongest input and persuasion is among the weakest, the practical move is not to argue with the belief. It is to arrange evidence against it that you cannot dismiss, at a size small enough to actually happen.

That is a duller instruction than reframing, and it has the advantage of being aimed at the thing the research says moves.

It also explains a common and demoralising experience: doing the belief work sincerely, feeling briefly better, and finding the belief exactly where it was a fortnight later. That is not a failure of commitment. It is what the ordering predicts.

Try this

Take the belief and make it specific enough to be wrong. Not “I’m bad at difficult conversations” but “I cannot raise the rota problem with my manager.”

Then find the smallest version of that action which is still genuinely the action — a two-line email rather than a meeting — and do it this week.

You are not looking for a result. You are looking for one instance of enactive mastery, which is the input the evidence ranks first. The belief is downstream of that, and arguing with it directly is the input ranked third.

The parts of this that are genuinely hard

Self-efficacy’s specificity cuts both ways. It makes the construct measurable and it makes it narrow: succeeding at the two-line email tells you about the two-line email, and generalisation to a whole domain is not automatic. Anyone promising that one small win rewrites your self-concept is overselling exactly what makes this evidence usable.

The within-person finding is also contested rather than settled. It is a meta-analytic argument about how a decades-old literature should be read, and Bandura and colleagues disputed that reading. What is not disputed is that the between-person and within-person questions are different, and that the advice industry has only ever quoted the between-person answer.

And some beliefs are not miscalibrated at all. A person who believes they cannot get a mortgage on their salary, or cannot leave a job in this labour market, may simply be right. The framework in this article is for beliefs that a small experiment could disconfirm. Treating an accurate assessment of a constraint as a limiting belief to be dissolved is not empowerment; it is telling somebody their circumstances are their attitude.

The bottom line

The belief and the performance are related — a weighted correlation of 0.38 across 114 studies is not nothing. But the within-person evidence, which is the test that matters for a person trying to change, is weak, and much of what looks like belief driving behaviour is behaviour that has already driven belief.

Bandura ranked the sources in 1977 and the ranking has not moved: doing the thing beats watching someone do it, which beats being told you can. The entire limiting-beliefs method is the third one. The first one is available, and it is smaller than you think.

Sources (4)

Every factual claim in this article comes from one of these sources.