Conjecture that Simpson datasets are pathological for logistic regression

A Simpson dataset is a 2×2×22\times 2\times 2 dataset exhibiting Simpson's paradox; a pathological regularization regime is a set of regularization parameters for which the trend obtained from a regularized regression is reversed compared with the dataset's true trend. Consider logistic regression with an intercept term. Pathological-regime conjecture. All 2×2×22\times 2\times 2 datasets strictly exhibiting Simpson's paradox are pathological for logistic regression with an intercept term, meaning that they have a non-trivial pathological regularization regime. This conjecture is based on numerical experiments in which every sampled Simpson dataset was pathological, while approximately 6.5%6.5\% of sampled non-Simpson datasets were pathological; its general validity remains open.

Sources & referencesView supporting material

Primary source

Maximilian Wiesmann and Paul Larsen, “Pathological Regularization Regimes in Classification Tasks”, arXiv:2406.14731 (2024).

Progress summary

Never refreshed

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Solutions 0

No solutions have been posted yet.