What Is Attribution Bias? (Luck vs. Skill Accounting)

Attribution bias — sometimes called self-serving attribution bias — is the systematic tendency to attribute positive outcomes to one's own skill, judgment, or effort, and negative outcomes to external circumstances, bad luck, or market conditions.

The bias is not a character flaw. It's a feature of how the ego constructs and defends a coherent self-image. An identity built around competence cannot simultaneously incorporate the belief that outcomes are significantly luck-driven. So the ego presents a narrative: successes reflect skill, failures reflect conditions.

The Asymmetry and Its Consequences

The problem is the asymmetry's direction. If failures were attributed to skill (or lack thereof) and successes to luck, the bias would at least produce a stable, calibrated self-assessment. But the direction runs consistently the wrong way for learning:

Successes receive no update because they confirm the existing model. Failures receive no update because they're attributed to external factors the model doesn't control. Either way, the model itself is never revised.

In any domain with significant randomness — trading, investing, venture, competitive markets — this means feedback loops are systematically broken. The performer who never updates their model never improves their model. They keep the same strategy that worked in the last favorable environment, into the next environment that might not favor it, believing they're skilled when their performance was a favorable draw from the distribution.

The Practical Correction

Nassim Taleb's corrective: explicitly, deliberately over-attribute success to luck before evaluating a result. Before deciding whether an outcome reflects genuine skill, assume the luck hypothesis is more likely, then ask what evidence would push you toward the skill hypothesis.

This is uncomfortable. It's not how advice books are written. But in domains with meaningful randomness, it's the only attribution policy that keeps feedback loops working. An outcome that survives the luck hypothesis — that remains impressive even after assuming favorable conditions, favorable timing, and favorable variance — is more likely to be signal than an outcome that collapses under the luck hypothesis.

The check: "Would this result be impressive in a fair Monte Carlo simulation where all the luck was stripped out?" If yes, update toward skill. If not, the attribution bias may be doing the heavy lifting.

For the full framework, read Living With Randomness.