Skip to content
All apps

Simpson's paradox 2.0

About this app

Rebuild 2.0 (preview): three real cases — kidney stones, Berkeley 1973, exercise & cholesterol. Subgroups as bubbles, equalise the case mix, and a causal panel asking which view is right.

The statistics

In Simpson’s paradox a relationship that holds in every subgroup reverses in the pooled view. It is no calculation error: a group’s overall rate is a weighted average of its subgroup rates, and the weights, the case mix, differ between groups. Standardising to a common case mix removes this effect. Which view is right depends on the causal model: if the third variable Z is a confounder (X ← Z → Y), you split; if Z is a mediator on the path X → Z → Y, the pooled view gives the total effect.

What you see

Three cases are available: kidney stones (treatment success by stone size), admissions to the six largest departments of UC Berkeley in autumn 1973, and an exercise and cholesterol example with age as the third variable. Each subgroup is a bubble; its height is the rate and its area the number of cases. A bar with a diamond marks the overall rate, with the case mix shown below. A slider gradually equalises the case mix, and a causal panel asks which view is correct for each case.

Things to try

Open the kidney stone case and switch between pooled and split views: within each stone size one treatment leads, overall the other. Then drag the case-mix slider to the right; the subgroup rates stay, only the weights move until both groups share the same mix. In the Berkeley case you compare women and men within departments. In the exercise example a slider sets how strongly age drives training hours; at 0 % the reversal disappears.

Frequently asked questions

What does the 1973 Berkeley case show?
Overall, UC Berkeley admitted men more often than women in 1973. Within most departments, however, women’s rates were similar or higher. Women had applied more often to highly competitive departments with low admission rates; Bickel, Hammel and O’Connell analysed this in Science in 1975.
What is the kidney stone example of Simpson’s paradox?
In a well-known study one treatment was more successful for both small and large kidney stones, yet less successful overall. The reason: it was used more often on the harder large stones. Stone size influences both the choice of treatment and the outcome, which makes it a confounder.
What does standardisation of rates mean?
Standardisation weights the subgroup rates of both groups by the same case mix instead of each group’s own. This removes differences in group composition. Any difference that remains is due to the rates themselves.
What is new in version 2.0 of the Simpson app?
Version 2.0 is a rebuild (preview) with three real cases: kidney stones, Berkeley 1973, and exercise and cholesterol. Subgroups appear as bubbles, a slider equalises the case mix, and a causal panel explains when splitting is right and when the pooled view is.

Subject: Statistics

More from Statistics

A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps