The proposed functional form for collinearity as a function of correlation and dimension

From papers

Let cc denote the collinearity parameter, Δ\Delta the correlation parameter, and pp the number of variables in the latent linear-regression framework described above. Proposed functional-form conjecture. Based on empirical observations, cc takes the form

c(Δ,p)=2Δ+(p2)Δ214Δ+2pΔ+p2Δ23pΔ2+3Δ2.c(\Delta, p) = \frac{2\Delta + (p-2)\Delta^2}{1-4\Delta + 2p\Delta + p^2\Delta^2 -3p\Delta^2 + 3\Delta^2}.

This postulate is intended to specify the theoretically expected dependence of the collinearity parameter on Δ\Delta and pp, following the derived closed-form expression for LOCO importance. The supplied text gives empirical motivation but does not state a proof or resolution.

Progress summary

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

Sources & referencesView supporting material

Primary source

Kelvyn K. Bladen, D. Richard Cutler and Alan Wisler, “Mathematical Theory of Collinearity Effects on Machine Learning Variable Importance Measures”, arXiv:2510.00557 (2025).

Solutions 0

No solutions have been posted yet.