Simpson's Paradox
A statistical phenomenon in which a trend that holds within separate groups of data disappears or reverses when the data are combined. It commonly occurs in scenarios where subgroup sizes differ substantially.
Core Facts
Timeline (last 90 days)
Average position may actually improve when visibility declines: after responses that originally ranked the brand lower no longer mention it, the remaining listed samples skew toward the top overall, which is a manifestation of Simpson's paradox at the metric level.
The 1973 UC Berkeley admissions data appeared overall to show a gender bias against women, but when broken down by department, most departments had higher admission rates for women; the disparity stemmed from women tending to apply to more competitive departments.
Aggregate metrics can show completely opposite performance across subgroups, which is an engineering variant of Simpson's paradox.
In the Simpson's paradox causal inference question, all four models identified the Simpson's paradox reversal, constructed a counterfactual world consistent with the observed data, and derived the theoretical bounds.
Simpson's paradox refers to the phenomenon where each subgroup in grouped data shows a correlation in one direction but the combined overall data shows the opposite direction, with the root cause being a confounding variable; handling it in causal inference requires introducing structural causal models or do-calculus.
The study points out that aggregated accuracy metrics systematically obscure asymmetric capabilities in multilingual factual reasoning, which is a manifestation of Simpson's paradox.
All Facts (6)
Simpson's paradox refers to the phenomenon in which a trend observed in grouped data may completely reverse when the data are aggregated.
65%UnverifiedAverage position may actually improve when visibility declines: after responses that originally ranked the brand lower no longer mention it, the remaining listed samples skew toward the top overall, which is a manifestation of Simpson's paradox at the metric level.
50%UnverifiedThe 1973 UC Berkeley admissions data appeared overall to show a gender bias against women, but when broken down by department, most departments had higher admission rates for women; the disparity stemmed from women tending to apply to more competitive departments.
50%UnverifiedAggregate metrics can show completely opposite performance across subgroups, which is an engineering variant of Simpson's paradox.
50%UnverifiedSimpson's paradox refers to the phenomenon where each subgroup in grouped data shows a correlation in one direction but the combined overall data shows the opposite direction, with the root cause being a confounding variable; handling it in causal inference requires introducing structural causal models or do-calculus.
50%UnverifiedThe study points out that aggregated accuracy metrics systematically obscure asymmetric capabilities in multilingual factual reasoning, which is a manifestation of Simpson's paradox.
50%