Biweight midcorrelation¶

A robust correlation measure that centers each variable at its median and downweights observations far from the median using a redescending biweight.

Version: v2 · 2026-09-19 · History
Domain-specific #: 3483
Domain group: Formal Sciences
Origin domain: Experimental Design & Statistics
Subdomain: Robust Statistics → Experimental Design & Statistics
Aliases: Bicor

Core Idea¶

The tuning constant and zero-MAD convention affect results, robustness does not make nonlinear dependence visible and pairwise missingness can break positive-semidefinite correlation matrices. Standardized median deviations determine compact-support weights, central paired deviations contribute to a weighted covariance-like numerator and weighted norms scale the result, suppressing observations beyond the cutoff. The abstraction is therefore identified by a declared carrier, a transformation or constraint over that carrier, and an invariant that tells an analyst whether the named structure is genuinely present.

How would you explain it like I'm…

Don't-Get-Fooled Matching

Say you want to know if taller kids also have bigger feet. The Biweight midcorrelation checks whether two lists of numbers go up and down together. It listens most to the kids near the middle, listens less to the unusual ones, and ignores a kid who is super far from everyone else, so one weird kid can't fool it.

The Middle-Listening Match Score

A correlation is a number that tells you whether two things tend to rise together, fall together, or neither. The ordinary kind can be pulled way off by one or two wild values. The Biweight midcorrelation first finds the middle (median) of each list and how spread out the values usually are. Then it gives every point a weight: points close to the middle count a lot, points farther out count less, and points past a cutoff count for nothing. The answer still runs from -1 to +1, with the sign telling you whether the two go together or opposite.

Outlier-Resistant Correlation

The Biweight midcorrelation is a robust alternative to the usual Pearson correlation between two paired variables. For each variable it measures how far each value is from the median, in units of the median absolute deviation (MAD), and scaled by a tuning constant. Those standardized distances feed Tukey's biweight function, which gives weight near 1 to central points, smoothly shrinking weight to farther points, and exactly zero weight beyond the cutoff. The weighted, centered cross-products form the numerator, and weighted sums of squares for each variable normalize it, so the result lies between -1 and +1. Unlike rank correlation, it still works with the actual values; unlike Pearson, a few extreme points cannot dominate it. It does not, however, reveal curved (nonlinear) relationships, and the tuning constant and how you handle a zero MAD change the answer.

Biweight midcorrelation estimates association between paired samples x and y while bounding the influence of outlying observations. For each variable, compute the median and MAD, then the standardized deviation u = (x - median)/(c * MAD) with tuning constant c. Tukey biweight weights w = (1 - u^2)^2 apply when |u| < 1 and are zero otherwise, so the weights have compact support. The numerator is a weighted cross-product of centered deviations over valid pairs, and the denominator normalizes by the square roots of the weighted sums of squares for each variable, placing the result in [-1, 1] with an interpretable sign. Results depend on the choice of c, on how pairs with missing values are handled, and on a convention for variables whose MAD is zero. Compared with Pearson correlation it is resistant to outliers; compared with Spearman or Kendall rank correlation it retains metric information near the center. Robustness does not make nonlinear dependence visible, and a matrix assembled from pairwise-complete biweight midcorrelations need not be positive semidefinite.

Scope of Application¶

Biweight midcorrelation belongs to robust statistics and is useful where the analyst can specify the typed robust statistics carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets, then evaluate the paired sample vectors and valid-pair rule, component medians and median absolute deviations, tuning constant, standardized deviations, Tukey biweight weights and cutoff, weighted centered cross-product, normalization by weighted sums of squares, range and sign, zero-scale handling, robustness and comparison with Pearson and rank correlation are explicit.

Clarity¶

The abstraction clarifies a crowded vocabulary by making the paired sample vectors and valid-pair rule, component medians and median absolute deviations, tuning constant, standardized deviations, Tukey biweight weights and cutoff, weighted centered cross-product, normalization by weighted sums of squares, range and sign, zero-scale handling, robustness and comparison with Pearson and rank correlation are explicit the center of the account. A claim should name the carrier, the governing operation or relation, the applicable assumptions, and the recognition test.

Manages Complexity¶

Without the abstraction, an analyst must reason directly over many local details: the carrier roles, admissibility assumptions, competing conventions, derived invariants, boundary cases, and proof or validation obligations specific to Biweight midcorrelation. Biweight midcorrelation compresses them into the roles in the structural signature. That compression permits comparison across instances without erasing the variables that determine validity. It also exposes which details may be varied safely and which are constitutive.

Abstract Reasoning¶

Identify the carrier. State what the elements, states, objects, or observations are: the typed robust statistics carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets. Reject examples whose alleged carrier belongs to a different problem.
Lock the constitutive rule. Express the paired sample vectors and valid-pair rule, component medians and median absolute deviations, tuning constant, standardized deviations, Tukey biweight weights and cutoff, weighted centered cross-product, normalization by weighted sums of squares, range and sign, zero-scale handling, robustness and comparison with Pearson and rank correlation are explicit independently of one notation or implementation.

Knowledge Transfer¶

Knowledge transfers strongly among subfields of robust statistics because they reuse the typed robust statistics carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets, Standardized median deviations determine compact-support weights, central paired deviations contribute to a weighted covariance-like numerator and weighted norms scale the result, suppressing observations beyond the cutoff., and type the carrier, state every parameter and convention in the definition, test that the paired sample vectors and valid-pair rule, component medians and median absolute deviations, tuning constant, standardized deviations, Tukey biweight weights and cutoff, weighted centered cross-product, normalization by weighted sums of squares, range and sign, zero-scale handling, robustness and comparison with Pearson and rank correlation are explicit, compare the nearest accepted identity, and report counterexamples, uncertainty, and limiting cases.

Relationships to Other Abstractions¶

Current abstraction Biweight midcorrelation Domain-specific

Parents (1) — outgoing typed relations

Biweight midcorrelation is a kind of Measurement Prime

It remains its own entry because its identity is fixed by the paired sample vectors and valid-pair rule, component medians and median absolute deviations, tuning constant, standardized deviations, Tukey biweight weights and cutoff, weighted centered cross-product, normalization by weighted sums of squares, range and sign, zero-scale handling, robustness and comparison with Pearson and rank correlation.

Hierarchy path (1) — routes to 1 parentless root

Biweight midcorrelation → is a kind of Measurement

Neighborhood in Abstraction Space¶

Biweight midcorrelation sits in a crowded region of the domain-specific corpus (25^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Descriptive Statistics & Correlation Measures (40 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-25