Attribution Bias: Why We Credit Skill and Blame Luck

Here is the pattern: a trader has a great year. They attribute it to their method, their insight, their process. The thesis worked. The system was right.

The following year, the same trader has a terrible year. The market changed. An unprecedented event occurred. A "ten-sigma" move happened. The conditions were anomalous. The thesis was still right — the market just didn't cooperate.

Nassim Taleb calls this attribution bias. Success goes to skill; failure goes to luck. The reverse — failure to incompetence, success to luck — almost never happens spontaneously. And the consequences are severe in any domain where randomness is significant.

The Asymmetry Is Structural

The bias isn't unusual. It's universal. In studies across domains, 80 to 90% of people rate themselves above average at most things. This is a mathematical impossibility without attribution bias — you can't have 85% of people performing above the median. The gap between the perceived and actual distribution is entirely composed of people who are attributing their performance to skill rather than to the luck component that contributed to it.

The mechanism is ego preservation. An identity built around competence — as a trader, an investor, an executive, an operator — cannot simultaneously hold that the performance is substantially luck-dependent. The ego demands the narrative: the performance reflects genuine skill, the failures reflect circumstances outside my control, and the next period will reveal the competence that the last period obscured.

This isn't vanity. It's a deep feature of how the mind constructs and maintains self-image. The alternative — honestly crediting luck when luck was the source — requires holding an uncomfortable uncertainty about one's own skill level that most minds are not built to sustain.

The LTCM Version

Long-Term Capital Management's collapse in 1998 is the case study Taleb discusses in detail. The firm was run by Nobel laureates. Their models were sophisticated. Their conviction was documented and explicit. When the collapse came — when Russia defaulted and correlations went to one — the principals' public explanation was the "unprecedented" nature of the event.

This is attribution bias in real time. The event was not unprecedented in any distributional sense. It was the rare event the strategy was implicitly short of — the tail the historical data had never sampled, but which the distribution contained. The failure to price it in was a model error, not an act of God.

But admitting "our model failed to account for a tail event that was in the distribution all along" would require attributing the failure to the strategy — to their own judgment. The attribution bias produced an alternative narrative: the event was unpredictable, the strategy was sound, the loss was bad luck visiting a good process.

The post-hoc rationalization was internally consistent. It also guaranteed that no genuine update occurred. If the loss was bad luck, the strategy requires no modification.

What It Costs You

Attribution bias is expensive because it prevents accurate calibration.

Genuine skill improvement requires distinguishing between situations where the process was wrong and situations where the process was right but the outcome was unlucky. If everything unfavorable is attributed to luck, the process never receives the feedback that would improve it. The trader who blames unusual market conditions for every losing year eventually runs a strategy that has never been genuinely evaluated.

The same mechanism operates in business. The founder who attributes every failure to market timing, bad macros, or competitor behavior — never to product quality, go-to-market strategy, or capital allocation — is accumulating no learning. The survivorship-selected stories of success have taught them to hold conviction. The attribution bias prevents the conviction from being updated by contrary evidence.

Building Feedback Loops That Work

Taleb's prescription is structural rather than aspirational. You can't simply decide to be less biased toward attributing success to skill. The ego defense mechanism is too automatic. But you can build external feedback loops that force exposure to the information the bias would otherwise filter out.

Maintain a decision journal. Record, before the outcome is known, the rationale for a decision and a prediction of what would make the decision wrong. After the outcome, review the record honestly. Did the prediction come true? Did the disconfirming signal appear? The journal makes the attribution comparison explicit and visible rather than subject to post-hoc reconstruction.

Seek structured outside evaluation. Peers who weren't emotionally invested in the decision can evaluate it more accurately than you can. A trading partner reviewing your closed positions, a board evaluating a strategy's execution, a friend reviewing a business decision — external perspectives are less subject to the ego-preservation mechanism.

Credit both directions. Taleb's personal rule: start every review by convincing everyone (starting with himself) that the team is "a bunch of idiots who happen to know we are idiots." The deliberate move toward over-crediting luck on successes counteracts the structural tendency to under-credit it. The goal isn't false modesty but calibration: the actual luck component of most successes is larger than the ego-narrative admits.

State the process, not the outcome. Evaluating whether a decision was good should be based on the process — the information available at the time, the reasoning applied, the alternatives considered — not the outcome. A decision that followed a good process and produced a bad outcome was still a good decision. An attribution system based on outcomes rather than process will systematically reward luck and punish rigor.

The honest accounting of luck versus skill is genuinely uncomfortable. It requires holding the belief that some significant fraction of your best results reflected favorable draws from a distribution, not pure insight. But without that accounting, the calibration fails — and in random-adjacent domains, miscalibrated self-assessment eventually meets the distribution.

For the full framework, read Living With Randomness.