Math

Simpson's paradox

Construct Simpson’s paradox with editable group weights and success rates, and compare subgroup and aggregate results.

Actual application image

Try this

Method A is blue; method B is pink. Each has 400 observations divided between easy and hard groups. Adjust group sizes and success rates: weighting can reverse the aggregate comparison even when A leads in both groups. This is a synthetic example, not causal evidence.

Equations used in the module

Equations
Overall rate = (n₁p₁+n₂p₂)/(n₁+n₂)
A and B may have different group weights
A₁>B₁ and A₂>B₂ can coexist with Aoverall<Boverall
The aggregate comparison alone cannot establish a causal effect

Related experiments

Back to catalogue

Math

Counting necklaces: Burnside's lemma

Count necklaces and bracelets with Burnside’s lemma, inspect symmetry operations and verify the result by enumerating representatives.

Math

Euler's identity in motion

Connect the unit circle, sine and cosine with the vector sum of the complex exponential series and Euler’s identity.

$Distance-field fractals — Actual application image$

Physics / Math

Distance-field fractals

Surfaces with no mesh: a distance estimator and sphere tracing draw the whole thing.