Answers are grounded in this paper and cite their sources.
Methodological study reveals properties of the Switching Degree in structural break detection, highlighting systematic post-selection bias under unknown boundaries.
This paper proposes the Switching Degree—a standardized effect size that measures “how large a regime switch is” in structural break detection—and establishes its complete axiomatic foundation, inference theory, and power analysis. The Switching Degree is defined as S = ‖Δθ‖_Σ / ‖θ_ref‖_Σ, where ‖·‖_Σ is the Mahalanobis norm induced by the second-moment matrix Σ of the design, a choice that makes the Switching Degree strictly consistent with the unified shape function h(τ) of Tang (2026af): S = √h(τ*) / ‖θ_ref‖_Σ. The paper explicitly distinguishes two complementary effect sizes: the local switching degree S (standardized by a domain-specific reference vector, suitable for within-domain comparisons) and the global effect size S_eff = ‖Δθ‖_Σ / σ (standardized by the noise scale, suitable for cross-disciplinary comparisons and power analysis), and establishes the equivalence relation f² = γ(1−γ)S_eff² linking both to Cohen's f². At the axiomatic level, the paper establishes that, given the data-driven inner-product structure Σ, a switching measure satisfying four natural norm axioms—non-negativity, positive homogeneity, triangle inequality, and the parallelogram law—together with one normalization condition is uniquely the Mahalanobis norm ratio; the role of the parallelogram law is to select the Euclidean norm from among all L^p norms. At the level of inference theory, Theorem F establishes consistency; Theorem G establishes √n(Ŝ−S) → N(0, σ²/[γ(1−γ)‖θ_ref‖_Σ²]); Theorem H provides a non-asymptotic shrinkage bound; Theorem I extends inference to the unknown-boundary setting, deriving the two-sided first-order bias expansion Ŝ_plug − S = −ηS + o_p(n^(−1/2)) and showing that boundary estimation error in either direction leads to systematic underestimation; Theorem J extends inference to the multi-regime setting, deriving the joint asymptotic normality of the switching degree vector Ŝ with a tridiagonal covariance matrix whose off-diagonal entries contain direction-cosine factors; Theorem K provides heteroskedasticity-robust variance corrections and plug-in estimators. Theorem D (Structural Invariance Theorem) proves that the Switching Degree is invariant under invertible Σ-isometries. The paper further introduces the switching coupling degree and proves Theorem C (Switching Degree Decomposition Theorem): the squared path switching degree equals the squared net switching degree plus the squared coupling switching degree; in the one-dimensional case, the relation between the switching coupling degree and mutual information is I(Δθ₁; Δθ₂) = −(1/2)log(1−cos²φ). The 2D Ising model provides numerical verification at the Onsager-calibrated critical temperature (S = 0.5, S_eff ≈ 25.8). Numerical simulations verify the finite-sample behavior of the inference theory and report a negative result: with unknown boundaries, the switching degree estimator exhibits a systematic negative bias (bias approximately 81% of the theoretical standard error), and residual bootstrap quantile intervals raise coverage only from 84.5% to 86.5%, far below the nominal 95%—demonstrating that post-selection bias is a substantive first-order problem that naive bootstrap cannot correct and that explicit corrections such as conditional bias expansion are required. The core methodology, research direction, and final decisions were independently directed by the author. DeepSeek assisted with code implementation, data presentation, and text drafting. The author takes full academic responsibility for the final content.
No takes yet. Share an insight, caveat, or question.
Shuiping Tang (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: