Interpolation, statistics and discrete distinctions

VectorPaint zero-correlation scatterplot: a five-node square relation is still dependent

Use five equally likely inputs with Y=X² to verify zero correlation and distinguish a model proof from generalizing five observations.

English translation of the Chinese original, translated on . Original publication dates, attribution, code and version notes are retained.

Define a finite probability model: X is uniform on −2,−1,0,1,2 and Y=X². Plot five points, compute Pearson correlation and disprove independence through event probabilities. Five real observations with empirical zero correlation would not justify an unknown-population claim; this proof relies on the fully specified model.

Fix the probability model without inventing observations

Write the five equally likely nodes and deterministic relation first, each with probability 1/5, not photographic evidence supposedly establishing equal probabilities. Ordered pairs are (−2,4),(−1,1),(0,0),(1,1),(2,4); retain negative and positive inputs separately. Values are abstract, with no real survey, trial or measurement. Separate nodes, mean guides, calculations and conclusions, then save native work. Real data later need new identity, sampling premises and scope, not a copied complete distribution or independence conclusion.

Compute means before paired deviations

Sums are zero for x and ten for y, giving means zero and two. Keep original pairs while forming x deviations −2,−1,0,1,2 and y deviations 2,−1,−2,−1,2. Independently sorting columns would change the relationship. Draw faint guides x=0,y=2 for means, not a fitted conclusion. Equal left/right heights follow squaring. Mean y=2 need not occur at a node and implies no independence of deviations.

Keep a zero cross-sum distinct from positive square sums

Paired products −4,1,0,−1,4 sum to Sxy=0, while Sxx=10 and Syy=14. Thus r=Sxy/√(Sxx Syy)=0 with a positive denominator, not an assigned value for 0/0. Uniform weights cancel in the complete five-node model’s ratio as well; with observations alone this would be empirical arithmetic first. Retain square sums as domain checks rather than only r. Cancellation does not mean every individual product is zero.

X uniform on {-2,-1,0,1,2}; Y=X^2
x:       -2,-1,0,1,2
y:        4, 1,0,1,4
mean x=0; mean y=2
Sxy=-4+1+0-1+4=0; Sxx=10; Syy=14
r=Sxy/sqrt(Sxx*Syy)=0
P(X=0,Y=0)=1/5 != (1/5)*(1/5)=1/25

Plot five nodes without changing the probability support

Use a suggested 1800×1600 canvas, origin (900,1300), 250 pixels per x unit and 200 per y unit, explicitly unequal. Points map to (400,500),(650,1100),(900,1300),(1150,1100),(1400,500). Use distinct dots, not horizontal bars merging equal heights. Any parabola guide must say probability lives only at five nodes; smooth paths/fills cannot imply uniform support on all reals. Approximate from guides if numeric fields are absent without claiming pixel measurement precision.

Disprove independence with one event pair

Here P(X=0)=P(Y=0)=1/5, but P(X=0 and Y=0)=1/5, not their product 1/25, disproving independence. Equivalently P(Y=0 given X=0)=1 differs from 1/5, with positive conditioning probability. The proof uses the full model, not merely a bent-looking scatterplot or five observations of unknown origin. Zero Pearson correlation does not establish no relationship, no predictive information or independence. Keep this scope explicit beside the diagram. Y takes 0,1,4 with probabilities 1/5,2/5,2/5, not one third each. Aggregate inputs sharing an output to audit probability denominators rather than count distinct outputs. These probabilities come from the complete model, not unqualified observed frequencies.

Use linear and constant controls to test the definition

With the same five x nodes and Y=2X+3, points lie on a positive line and correlation is one, without proving real causation. Constant Y=2 makes Syy zero and Pearson correlation undefined, not automatically zero. Keep separate controls without changing main Y=X² points. Retain x support/pairing and state that only Y’s rule changes. This explicit change is more auditable than three unexplained shapes. No confidence interval, significance test or real-population inference is performed. The linear control has y values −1,1,3,5,7, mean three, Sxy=20,Syy=40 and Sxx=10, giving r=20/√400=1. Compare this positive denominator with the constant case, avoiding a claim that every deterministic relationship has correlation one.

Troubleshoot pairing, denominator and model identity

For nonzero r, check original pairing, product signs and mean y=2. A dependence claim without probability premises needs the full model restored or narrowed to empirical correlation. Do not sort columns separately, delete negative nodes or use colours to conceal a constant-case zero denominator. Preserve originals, repair one copy factor and recompute three sums plus event probabilities. Smooth paths prove no independence, and five graphic circles are not evidence of five random trials.

Reopen with separate evidence for zero correlation and dependence

Save native work and reopen SVG/PDF, retaining five points, mean guides, probability premises, square sums and denominator 25. At fit-to-window size distinguish main/control panels; inspect minus signs enlarged. Reopen points, guides and notes separately. Deliver distribution, arithmetic, event counterexample and limits on generalizing observations. Leave unperformed actions pending, without automatic statistics/nonlinearity recognition or unknown-population claims from five real observations.

VectorPaint scatterplots: map every point to two values; VectorPaint least-squares line: four vertical residuals with squared sum 0.70.

Method sources: Philip Stark, Berkeley SticiGui: correlation and association; OpenStax Introductory Statistics 2e §3.2: independence.

Version and operation scope: checked on 17 November 2026 against the official VectorPaint listing and Apple public lookup: version 1.0.8, requiring iOS / iPadOS 17.0 or later. Draw manually with published shapes, pen, Bézier nodes, grids/guides, alignment, layers/groups, text and file output through visible installed controls. Prepare definitions and calculations externally; infer no built-in formula, statistics or graph solver. Keep editable native work and inspect SVG, PDF or PNG as needed. SVG/PDF containing imported bitmaps do not automatically become entirely vector. Canvases and samples are original teaching settings, without physical-device execution or real measurement; mark unperformed actions pending.