Quadrant Count Ratio¶

Center paired quantitative observations at their sample means, score same-side pairs as concordant and opposite-side pairs as discordant, and normalize the signed count difference by sample size to obtain a coarse association statistic in [-1, 1].

Version: v3 · 2026-09-06 · History
Domain-specific #: 2578
Origin domain: statistics
Subdomain: descriptive statistics and statistics education
Aliases: QCR, Quadrant-count ratio

Core Idea¶

The quadrant count ratio (QCR) is a coarse statistic for the direction and predominance of association between two paired quantitative variables. It draws a vertical line through the sample mean of (X) and a horizontal line through the sample mean of (Y). Observations in the upper-right and lower-left quadrants are concordant relative to the means: both components lie on the same side of their respective centers. Observations in the other two quadrants are discordant relative to the means. QCR subtracts the discordant count from the concordant count and divides by the total number of paired observations.^[1]^[2]

For paired observations \((x_i,y_i)\), \(i=1,\ldots,N\), with sample means \(\bar{x}\) and \(\bar{y}\), define

\[ s_i=\operatorname{sgn}(x_i-\bar{x})\operatorname{sgn}(y_i-\bar{y}), \qquad q=\frac{1}{N}\sum_{i=1}^{N}s_i. \]

With (operatorname{sgn}(0)=0), a point on either mean line contributes zero to the numerator while remaining in the declared total (N). When no observation lies on a mean line, this is exactly

\[ q=\frac{(n_1+n_3)-(n_2+n_4)}{N}, \]

where (n_j) is the count in quadrant (j). This sign-product form exposes the relationship to covariance: each centered cross-product contributes only its sign, not its magnitude. Pearson's ®, by contrast, sums the magnitudes of centered cross-products and normalizes by the variables' centered sums of squares.^[3]

The statistic is unitless and lies in ([-1,1]). Positive values mean same-side pairs outnumber opposite-side pairs; negative values mean the reverse; zero means that the two counts balance after any mean-line zeros. These interpretations are intentionally limited. (q=1) says all counted pairs are mean-concordant, not that the points lie on a positive straight line. (q=0) says signed counts cancel, not that the variables are independent or unrelated. The QCR is used especially as an instructional bridge from scatterplot pattern to Pearson correlation, not as a replacement for a model, a test, or a general dependence measure.^[2]^[4]

The locked identity is:

paired quantitative sample + declared mean centers + four sign quadrants + concordant-minus-discordant count + division by total N -> bounded sign-count association statistic.

Changing the center rule changes the statistic. Median-centered quadrant counts are a documented teaching variant, including in GAISE II, but they must not be silently substituted for the mean-centered identity developed here.^[4]

Structural Signature¶

Sig role-phrases:

the paired quantitative sample — complete ((x_i,y_i)) observations with an explicitly declared analysis population and missing-data rule
the two sample means — \(\bar{x}\) and \(\bar{y}\), which establish the vertical and horizontal reference lines in the canonical definition
the four mean-centered regions — high–high, low–high, low–low, and high–low positions relative to the two means
the concordance coding — (+1) for upper-right or lower-left observations whose centered deviations share a sign
the discordance coding — (-1) for upper-left or lower-right observations whose centered deviations have opposite signs
the axis convention — zero contribution for an observation lying exactly on either mean line, with the denominator policy stated
the signed count contrast — concordant count minus discordant count, rather than a distance-weighted sum
the normalization denominator — the total paired-observation count (N), fixing the result to a comparable scale
the bounded readout — \(q\in[-1,1]\), interpreted as predominance and direction of mean-relative concordance, not as linear fit or causation

Locked signature: mean-center -> classify signs -> contrast concordant and discordant counts -> normalize by N -> bounded coarse association readout.

Recognition test: A reported quantity is the canonical QCR only if it uses paired quantitative observations, centers both variables at their sample means, assigns same-side and opposite-side observations to the two signed groups, forms their count difference, divides by the declared total sample size, and interprets the result as a coarse association statistic. Median splits, arbitrary business thresholds, products of deviation magnitudes, pairwise rank concordance, or quadrant-specific percentages may be useful relatives, but they are not this exact statistic unless explicitly labeled as variants.

The construction guarantees boundedness because every (s_i) is (-1), (0), or (1), so the absolute numerator cannot exceed (N). It guarantees invariance to translations and positive changes of unit: adding constants moves the means with the data, and multiplying either variable by a positive constant preserves every centered sign. Reversing the direction of exactly one scale multiplies (q) by (-1). It does not guarantee invariance under arbitrary monotone transformation, because a nonlinear transformation does not generally carry the sample mean to the transformed mean.

What It Is Not¶

Not Pearson's correlation coefficient. QCR discards centered magnitudes and retains only sign quadrants. Pearson's ® measures standardized linear co-movement using every centered cross-product.^[3]
Not covariance. Covariance retains joint scale and weights distant observations more heavily; QCR is unitless and gives equal (pm1) weight to all off-axis observations.
Not a general dependence measure. A U-shaped, circular, or other nonlinear relationship can produce balanced quadrant signs and (q=0).
Not Kendall's tau. Kendall's statistic compares all pairs of observations for rank concordance; QCR compares each observation only with the two chosen centers.
Not a median quadrant coefficient by default. Median-centered teaching variants have different threshold behavior and robustness properties and must declare their center rule.
Not a regression slope or prediction model. QCR provides neither units of change, fitted values, residuals, nor an equation for predicting (Y) from (X).
Not evidence of causation. It summarizes a sample pattern; design and subject-matter evidence are required for causal claims.
Not inferential uncertainty. A point estimate alone supplies no sampling distribution, standard error, confidence interval, or hypothesis test.

Scope of Application¶

The abstraction's main habitat is descriptive statistics and statistics education. In exploratory analysis it offers a fast, visually auditable summary of whether paired observations more often occupy same-side or opposite-side regions relative to their means. In teaching, it connects verbal statements—“above-average (X) tends to accompany above-average (Y)”—to a signed numerical statistic before the magnitudes, standardization, and linear-fit commitments of Pearson's ® are introduced. Holmes developed the count construction as an intuitive route from scatter diagrams toward correlation formulas, and Kader and Franklin explicitly use QCR's shortcomings to motivate Pearson's coefficient.^[1]^[2]

It can also serve as a deliberately coarse sign feature in an applied workflow when that loss of magnitude is understood and validated. Its suitable role is screening, explanation, or comparison under a fixed protocol—not automatic substitution for a conventional association estimator. Applications must preserve paired cases, state the center and axis convention, inspect the scatterplot, and report why a sign-count summary is fit for the question.

The name should not be generalized to every four-region score. A risk matrix split at regulatory cutoffs, a two-by-two contingency coefficient, and a median-sign test each instantiate related structures while answering different questions.

Clarity¶

QCR makes the sign of centered co-movement visible. The numerator can be read directly from a scatterplot: count observations that support positive association, subtract those that support negative association, then normalize. This reveals a conceptual skeleton that the compact Pearson formula can hide—association direction comes from the signs of paired deviations from center.

It also clarifies exactly what is being thrown away. Once a point is classified into a quadrant, its distance from either mean no longer matters unless moving it changes a mean line or crosses one. A point barely above both means and one far into the upper-right quadrant receive the same (+1). Naming that loss prevents a user from interpreting (q) as though it measured tightness around a line.

Manages Complexity¶

The statistic collapses an \(N\times2\) dataset into two centers, four counts, and one bounded number. That compression supports hand calculation, classroom discussion, and rapid comparison of small scatterplots. The calculation has a simple audit trail: means, quadrant membership, counts, numerator, denominator.

The same compression is its main limitation. It suppresses slope, curvature, clusters, heteroscedasticity, gaps, leverage, and within-quadrant geometry. Complexity is managed responsibly only when the scatterplot and protocol remain available beside the scalar. QCR is therefore a lossy index with a transparent loss function, not a sufficient summary of bivariate structure.

Abstract Reasoning¶

QCR can be written as an average of centered sign products. This representation separates three operations: centering chooses the reference state; quantization maps continuous deviations to (-1,0,+1); aggregation averages their products. Each operation creates a distinct sensitivity. Moving one point can shift a mean and reclassify other points; quantization makes within-cell distances invisible; aggregation lets positive and negative structures cancel.

That decomposition supports useful counterfactuals. Recompute with one case removed to expose center leverage. Compare mean- and median-centered variants to expose threshold dependence. Compare (q) with Pearson's ® to expose magnitude leverage and linearity. Inspect subsets to expose cancellation between groups. A disagreement among those views is not a nuisance to average away; it identifies which structural feature carries the association claim.

Knowledge Transfer¶

The transfer is Type C: instrument transfer. The literal statistic transfers wherever the same preconditions hold: paired quantitative cases, declared centers, sign classification, a count contrast, and a defensible association interpretation. A quality engineer, ecologist, or educator may calculate the same QCR without altering its mathematics.

What must transfer with the formula is the validity package. The mean lines are data-dependent; missing-pair handling changes (N); ties need an axis convention; subgroup mixtures can cancel; nonlinear structure can disappear; and extreme QCR does not certify an exact line. Importing only the formula while dropping these conditions produces a number with the name but not the abstraction's warranted inference.

Examples¶

Canonical — extreme QCR without perfect linearity¶

Consider six paired observations:

\[ (-3,-1),\;(-2,-4),\;(-1,-2),\;(1,5),\;(2,1),\;(3,4). \]

The means are \(\bar{x}=0\) and \(\bar{y}=0.5\). The first three observations lie below both means and the last three lie above both means. Thus \(n_1+n_3=6\), \(n_2+n_4=0\), and \(q=(6-0)/6=1\). Yet the points do not lie on one straight line: observations with nearby \(x\)-values have markedly different \(y\)-values. Using the Pearson formula gives \(r\approx0.771\), not \(1\). NIST's definition explains the difference: Pearson weights centered cross-product magnitudes and reserves \(|r|=1\) for an exact linear relationship.^[3]

The failure mode is reading “all signs agree” as “all distances agree.” The intervention is to retain the scatterplot and compute a magnitude-sensitive linear measure when linear tightness matters.

Mapped back: the paired sample supplies (N=6); the means define the two lines; every sign product is (+1); the signed contrast is six; normalization yields the upper bound. The noncollinearity demonstrates the guarantee's exact limit.

Applied / In Practice — GAISE II finch data and a declared variant¶

GAISE II introduces association using paired beak length and beak depth for 54 Medium Ground Finches. Its instructional display uses median lines: 44 observations occupy the two same-side quadrants and 10 occupy the opposite-side quadrants, giving \((44-10)/54\approx0.63\). The report uses this result as a Level B description of fairly strong positive association and later develops Pearson correlation at Level C.^[4]

This is a valuable application and a boundary case: the report explicitly uses medians and notes that means could instead be used. A careful analyst therefore reports “median-centered QCR = 0.63,” not simply QCR without qualification. Replacing the center can move observations between quadrants and change the value; the two versions are recognized variants, not numerically interchangeable implementations.

The failure mode is silent center substitution. The intervention is to publish the center rule, axis/tie policy, four counts, and (N), so another analyst can reproduce the statistic and compare the mean-centered canonical calculation.

Mapped back: finch measurements supply paired quantitative cases; the declared median variant supplies center lines; 44 same-side and 10 opposite-side cases form the signed contrast; division by 54 yields 0.63; the explicit variant label preserves the identity boundary rather than obscuring it.

Structural Tensions¶

T1 — Visual intuition versus inferential precision. QCR is easy to see and calculate, but a transparent point estimate is not automatically a calibrated estimator or test. Diagnostic: ask whether the task is description, pedagogy, screening, estimation, or inference, and add sampling uncertainty appropriate to the actual task.

T2 — Sign retention versus magnitude loss. Equal signs receive equal weight regardless of distance from the means, so linear tightness and leverage are suppressed. Diagnostic: compare the scatterplot and Pearson ®; large disagreement signals magnitude or geometry that QCR discarded.

T3 — Mean centering versus center leverage. Ignoring within-quadrant distances sounds robust, yet an extreme observation can move a sample mean and reclassify many other observations. Diagnostic: perform leave-one-out center and QCR checks, and report any median-centered sensitivity analysis as a distinct variant.

T4 — Bounded extremity versus perfect fit. \(q=\pm1\) requires sign unanimity, not collinearity. Diagnostic: inspect residual scatter around a fitted line and never translate QCR extremity into perfect prediction.

T5 — Balanced counts versus no relationship. Positive and negative patterns can cancel, and nonlinear dependence can be quadrant-balanced. Diagnostic: inspect the full scatterplot for curvature, clusters, or subgroup structure before interpreting \(q\approx0\) as absence of association.

T6 — Stable name versus variant center. Mean- and median-centered versions share a pedagogical family but can yield different counts and sensitivity. Diagnostic: require the center, axis convention, and denominator in every reported result.

T7 — Autonomy versus reduction. QCR is built from correlation, ratio, partition, sign coding, and aggregation, but a bare conjunction does not specify the mean-centered four-quadrant protocol, exact signed numerator, axis rule, bounded interpretation, pedagogical relation to Pearson, or extreme-value failure. Diagnostic: if a decomposition cannot reproduce the statistic and its distinctive warnings without reintroducing those specialist obligations, the domain node remains independently useful.

Structural–Framed Character¶

Quadrant Count Ratio is framed/domain-specific, not a substrate-independent prime.

Vocabulary travels only with statistical translation. Sample means, scatterplot quadrants, paired observations, association, and Pearson correlation belong to a statistical measurement frame.
Evaluative meaning is domain-bound. A larger absolute value is interpreted relative to association direction and use; it is not generically “better.”
Institutional origin matters. Statistics-education articles and curriculum frameworks stabilize the name, formula, and pedagogical role.^[1]^[4]
Human practice is constitutive. Analysts choose the sample, center, missing-data and tie rules, comparison statistic, and warranted interpretation.
Other domains import the instrument rather than discover the named object unchanged. A sign-count diagnostic becomes QCR only when its paired-variable and association protocol is preserved.

Its character: a framed descriptive-statistics instrument whose sign-count mechanism is simple and portable but whose identity and validity depend on a specialist bivariate-analysis protocol.

Structural Core vs. Domain Accent¶

Structural core. Two attributes are centered, each observation is reduced to a pair of signs, same-sign and opposite-sign cases are contrasted, and the result is normalized into a bounded scalar.

Domain accent. The carriers are paired quantitative observations; centers are sample means in the canonical definition; the four cells are scatterplot quadrants; the scalar is interpreted as coarse association; and its use is bounded against covariance, Pearson linear correlation, nonlinear dependence, causation, and inference.

Removal test. Strip away the statistical vocabulary and a generic normalized sign-count remains, but it no longer tells the analyst which centers to use, what counts as concordance, why ([-1,1]) has an association interpretation, or why (q=1) does not mean a perfect line. Those lost conditions determine correct calculation and interpretation, so the domain accent is constitutive.

correlation — proposed strict subsumption parent. QCR is a specific statistic for systematic co-variation between paired variables, with a mean-relative sign-count implementation and deliberately coarse interpretation.
ratio — proposed strict subsumption parent. Its readout is an ordered numerator—the signed count difference—divided by the nonzero total paired count (N), with both roles and scope required for interpretation.
covariance — closest declined neighbor. Both center paired variables and combine the signs of cross-products, but covariance preserves deviation magnitudes and joint scale while QCR discards them.
aggregation — related constituent. QCR collapses (N) signed classifications into one scalar, but adding a third direct parent would be nonminimal because the ratio numerator already specifies the count aggregation.
partition — related constituent. Mean lines divide the off-axis plane into four mutually exclusive regions; it is an internal construction rather than the association or ratio genus.
measurement — related practice. Units cancel in QCR, but cases, variables, missingness, and measurement quality still determine what the statistic can mean.

Relationships to Other Abstractions¶

Current abstraction Quadrant Count Ratio Domain-specific

Parents (2) — more general patterns this builds on

Quadrant Count Ratio is a kind of Correlation Prime

correlation — proposed strict subsumption parent. QCR is a specific statistic for systematic co-variation between paired variables, with a mean-relative sign-count implementation and deliberately coarse interpretation.
Quadrant Count Ratio is a kind of Ratio Prime

ratio — proposed strict subsumption parent. Its readout is an ordered numerator—the signed count difference—divided by the nonzero total paired count (N), with both roles and scope required for interpretation.

Hierarchy paths (2) — routes to 2 parentless roots

Quadrant Count Ratio → Correlation

Show alternative path (1)

Neighborhood in Abstraction Space¶

Quadrant Count Ratio sits in a sparse region of the domain-specific corpus (79^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Statistical Adjustment & Estimation Effects (14 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Pearson's ®. Tell: if centered products are magnitude-weighted and standardized by sums of squares, the statistic measures linear correlation rather than sign-count predominance.
Covariance. Tell: if the result retains product units and changes under positive rescaling, deviation magnitude has not been discarded.
Kendall's tau. Tell: if concordance is determined by comparing all pairs of cases rather than each case with fixed centers, the estimator is rank-pair based.
Quadrant correlation with medians. Tell: if the dividing lines are medians, report a median-centered variant and its tie convention rather than silently using the canonical mean-centered label.
Fourfold or phi correlation. Tell: if the inputs are intrinsically binary categories summarized in a two-by-two table, the object is not a dichotomization of paired quantitative deviations at their means.
Regression. Tell: if the output includes a slope, intercept, fitted values, or predictions in outcome units, a model has been fit rather than a quadrant count summarized.
General dependence. Tell: if the task is to detect any nonlinear relationship, a balanced sign count is insufficient and a broader diagnostic is required.
Causal effect. Tell: if the claim concerns what changing (X) would do to (Y), study design and causal assumptions—not the value of QCR—carry the inference.

References¶

[1] Holmes, P. (2001). “Correlation: From Picture to Formula.” Teaching Statistics, 23(3), 67–71. registry ↩a ↩b ↩c

[2] Kader, G. D., & Franklin, C. A. (2008). “The Evolution of Pearson's Correlation Coefficient.” Mathematics Teacher, 102(4), 292–299. registry ↩a ↩b ↩c

[3] National Institute of Standards and Technology. Correlation — Dataplot Reference Manual. Formula and interpretation for Pearson's linear correlation coefficient. Accessed 2026-08-26. registry ↩a ↩b ↩c

[4] Bargagliotti, A. E., Franklin, C. A., Arnold, P., Gould, R., Johnson, S., Perez, L., & Spangler, D. A. (2020). Pre-K–12 Guidelines for Assessment and Instruction in Statistics Education II (GAISE II). American Statistical Association, especially pp. 60–61. registry ↩a ↩b ↩c ↩d