Survivorship Bias

Survivorship bias was first systematically articulated by statistician Abraham Wald during World War II when analyzing bullet hole patterns on aircraft.

The phenomenon of training only on success stories while lacking knowledge of failure patterns is statistically known as survivorship bias.

Survivorship bias is a cognitive bias that stems from focusing only on samples that have passed through a selection process while ignoring eliminated cases.

Survivorship bias is a common logical fallacy: people only see the samples that survived successfully after some kind of screening, while ignoring the samples that failed despite making the same attempt and therefore never came into view.

For young people without family capital as a backstop, the rags-to-riches startup story is more like a case of survivorship bias.

The archaeological record is a product of survivor bias—what survives is stone, pottery, and metal, rather than textiles, wood, and food, so the variety and quantity of traded goods are severely underestimated.

For a list-driven pipeline, historical sequences are a product of survivorship bias, retaining only entities that remain consistently popular, and these happen to be the slowest-changing batch.

Survivorship bias arises from keeping only samples that "survived to today"; the approach to handling it is to retain the complete history, fill the end of delisted companies' windows with delisting returns, and link tickers before and after name changes through entity resolution.

Survivorship bias refers to the phenomenon where predictions that came true are repeatedly cited while failed predictions are ignored, leading to a systematic overestimation of technological judgment.

A more rigorous framework for evaluating someone's prediction accuracy is to examine all of their public predictions as a whole sample, rather than relying on a single tweet that turned out to be correct.

Survivorship bias refers to the phenomenon of only observing samples that have 'survived' after a selection process while ignoring the large number of eliminated samples, thereby drawing distorted conclusions.

To judge whether a source of investment advice is reliable, one should examine its complete, traceable, and publicly disclosed full recommendation records and overall win rate and drawdown, rather than individually cherry-picked cases selected after the fact.

'Displaying track records + funneling traffic to private channels' is the standard structure of stock-recommendation schemes, and selectively showcasing individual stocks with good performance after the fact constitutes survivorship bias.

During World War II, statistician Abraham Wald studied the distribution of bullet holes on damaged returning aircraft and found that the parts truly in need of reinforcement were those with no bullet holes.

The commentary points out that the original evaluation contains a systematic bias in baseline construction that has been overlooked—survivorship bias.

The commentary article concludes that the differences between humans and LLMs observed in the original study may be partly or even largely a product of survivorship bias, rather than a true reflection of creative generation ability.

幸存者偏差最著名的例子来自二战时期，统计学家亚伯拉罕·瓦尔德指出军方只能检查安全返航飞机上的弹孔分布，而被击落的飞机无法被观察到

幸存者偏差是一种认知偏误，源于只关注通过某个筛选过程的样本而忽略被淘汰的案例，二战时期统计学家亚伯拉罕·瓦尔德指出军方只能检查安全返航飞机上的弹孔分布