What does the value of r mean?

r = +1: perfect positive linear relationship (points lie exactly on an upward line). r = −1: perfect negative linear relationship (points lie exactly on a downward line). r = 0: no linear relationship. General strength guidelines: |r| ≥ 0.8: Very Strong; 0.6 ≤ |r| < 0.8: Strong; 0.4 ≤ |r| < 0.6: Moderate; 0.2 ≤ |r| < 0.4: Weak; |r| < 0.2: Very Weak or No relationship. Note these are conventional thresholds — the appropriate benchmark depends on the field of study.

What is Spearman's rank correlation and when should I use it instead?

Spearman's rank correlation coefficient ρ (rho) is a non-parametric alternative to Pearson's r. Instead of using raw data values, it ranks both variables and then applies the Pearson formula to the ranks. Use Spearman's ρ when: (1) data are ordinal (rankings); (2) data are not normally distributed; (3) the relationship is monotonic but not linear; (4) significant outliers are present. Spearman's ρ is more robust to outliers and non-normal distributions than Pearson's r.

What is a scatterplot and why should I use one?

A scatterplot is a graph with variable X on the horizontal axis and variable Y on the vertical axis, with each data pair plotted as a point. It is the most important first step in correlation analysis because: (1) it reveals whether the relationship is actually linear (required for Pearson's r); (2) it shows outliers that might distort r; (3) it can reveal non-linear patterns (e.g., quadratic) that r would miss; (4) it shows whether the data splits into clusters. Anscombe's Quartet famously demonstrates four datasets with identical r = 0.816 but completely different scatterplot patterns — only one of which is truly linear.

What is Anscombe's Quartet and why does it matter for correlation?

Anscombe's Quartet (1973) is a set of four datasets that have nearly identical summary statistics (mean, variance, correlation r ≈ 0.816, regression line) but look completely different on a scatterplot. One is a proper linear relationship; one is a perfect curve (quadratic); one has an outlier dragging the correlation up; one has a perfectly vertical cluster. This demonstrates that Pearson's r alone is insufficient — always visualise your data with a scatterplot before drawing conclusions from the correlation coefficient.

What is Kendall's tau and how does it differ from Spearman and Pearson?

Kendall's tau (τ) is another rank-based correlation measure. While Spearman's ρ ranks data and applies Pearson's formula, Kendall's τ counts concordant pairs (both variables increase together) versus discordant pairs (one increases, the other decreases): τ = (C−D)/√[(C+D+T_x)(C+D+T_y)]. Kendall's τ tends to produce smaller absolute values than Spearman's ρ, but its p-values are more accurate for small samples. Use Kendall's τ when you need a more robust rank correlation that better handles tied values.

Correlation Calculator — Pearson r, R², & Step-by-Step Breakdown (Free Online Tool)

Statistics Pearson's r Correlation Coefficient R² Bivariate Analysis Data Science

The Pearson correlation coefficient is one of the most important numbers in all of statistics. It tells you in a single value — ranging from −1 to +1 — how strongly two variables move together and in which direction. Whether you're a student checking your dataset for a school project, a researcher exploring relationships in survey data, or a data analyst building a predictive model, knowing how to calculate and interpret correlation is a foundational skill.

The HeLovesMath Free Correlation Calculator computes Pearson's r and R² instantly from any two paired datasets you enter. No spreadsheet, no textbook, no manual arithmetic required. This guide also teaches you exactly what correlation means, derives the formula from first principles, covers Spearman and Kendall alternatives, discusses the classic pitfalls (correlation ≠ causation, Anscombe's Quartet), and provides four fully worked numerical examples with every step shown.

Free Online Pearson Correlation Calculator

Variable X values (comma-separated) Enter numbers separated by commas. At least 2 pairs required. Variable Y values (comma-separated) Must contain exactly the same number of values as X.

📈 Correlation Results

Pearson r

—

R² (Coeff. of Det.)

—

R² as %

—

Interpretation: —

What Is Correlation? — Measuring Linear Association

Correlation is a statistical measure that describes the degree to which two variables move in relation to each other. More precisely, it quantifies the strength and direction of the linear association between two continuous variables. The word comes from the Latin correlatio — "mutual relation."

The concept was formalised by Sir Francis Galton in the 1880s while studying the relationship between heights of parents and children (a phenomenon he called "regression towards mediocrity"). His student Karl Pearson gave it mathematical rigour in 1895–1896, deriving what we now call Pearson's r — the standard measure of correlation in modern statistics.

📈 Positive Correlation

As X increases, Y tends to increase too. The scatterplot shows points rising from lower-left to upper-right. Examples: height and weight, study hours and test scores, income and spending.

📉 Negative Correlation

As X increases, Y tends to decrease. Points fall from upper-left to lower-right. Examples: temperature and heating bill, price of a product and demand, hours of TV watched and exam results.

➡️ Zero Correlation

No systematic linear relationship — points scatter randomly. r ≈ 0. Note: r = 0 does NOT mean no relationship exists; there could be a strong non-linear (e.g., curved) relationship that r fails to detect.

⚠️ Critical Warning: Pearson's r only detects linear relationships. A perfect U-shaped (quadratic) relationship between X and Y can give r = 0, leading you to falsely conclude "no relationship exists." Always visualise your data with a scatterplot before relying on r.

Pearson Correlation Formula — Derived & Explained

There are several equivalent ways to write the Pearson correlation formula. We'll present all of them, starting from the most conceptually meaningful and ending with the computationally convenient form used by this calculator.

Conceptual Definition — Standardised Covariance

Pearson's r is defined as the covariance of X and Y divided by the product of their standard deviations:

✦ Pearson r — Population Formula

\[\rho_{XY} = \frac{\text{Cov}(X,Y)}{\sigma_X \sigma_Y} = \frac{E[(X-\mu_X)(Y-\mu_Y)]}{\sigma_X \sigma_Y}\]

ρ = population correlation coefficient | Cov(X,Y) = population covariance | σ_X, σ_Y = population standard deviations | μ_X, μ_Y = population means | This formula shows r as the standardised version of covariance, making it dimensionless and bounded in [−1, +1].

✦ Pearson r — Sample Formula (z-score form)

\[r = \frac{1}{n-1}\sum_{i=1}^{n}\left(\frac{x_i - \bar{x}}{s_x}\right)\!\left(\frac{y_i - \bar{y}}{s_y}\right)\]

n = sample size | x̄, ȳ = sample means | s_x, s_y = sample standard deviations | This form makes clear that r is the average product of z-scores — it measures how much X and Y deviate from their means in the same direction simultaneously.

Computational Formula (used by this calculator)

Expanding the z-score formula algebraically yields a form that avoids computing the means first — ideal for hand calculation and computer implementation:

✦ Pearson r — Computational Formula

\[r = \frac{n\!\sum x_i y_i - \!\sum x_i \!\sum y_i}{\sqrt{\!\left(n\!\sum x_i^2 - \!\left(\!\sum x_i\right)^{\!2}\right)\!\left(n\!\sum y_i^2 - \!\left(\!\sum y_i\right)^{\!2}\right)}}\]

Understanding What Makes r = ±1

By the Cauchy-Schwarz inequality, |r| ≤ 1 always. Equality |r| = 1 holds when all data points lie exactly on a straight line. If the line has positive slope, r = +1; negative slope, r = −1. This is the mathematical guarantee that r is always bounded between −1 and +1.

✦ Pearson r in terms of Covariance and Standard Deviations

\[r = \frac{S_{xy}}{S_{xx}\cdot S_{yy}} \quad \text{where} \quad S_{xy} = \sum x_i y_i - \frac{(\sum x_i)(\sum y_i)}{n}\] \[S_{xx} = \sum x_i^2 - \frac{(\sum x_i)^2}{n}, \quad S_{yy} = \sum y_i^2 - \frac{(\sum y_i)^2}{n}\]

S_xy = corrected sum of cross-products (numerator of sample covariance × (n−1)) | S_xx = corrected sum of squares for X | S_yy = corrected sum of squares for Y

Coefficient of Determination R² — How Much Variance Is Explained?

✦ Coefficient of Determination

\[R^2 = r^2\]

R² ranges from 0 to 1 (or 0% to 100%). It represents the proportion of the total variance in Y that is associated with (linearly "explained" by) variance in X. Example: r = 0.85 → R² = 0.7225 → 72.25% of variance in Y is associated with X.

R² is arguably more interpretable than r because it has a direct percentage interpretation. Consider these examples:

Pearson r	R²	% Variance Explained	Interpretation
±1.00	1.000	100%	Perfect linear fit
±0.90	0.810	81.0%	Very strong — excellent prediction
±0.80	0.640	64.0%	Very strong
±0.70	0.490	49.0%	Strong — around half variance explained
±0.60	0.360	36.0%	Moderate-strong
±0.50	0.250	25.0%	Moderate
±0.30	0.090	9.0%	Weak
0.00	0.000	0%	No linear relationship

Interpreting r — Strength, Direction & Context

There is no single universal scale for interpreting r strength — the appropriate benchmark depends heavily on your field of study. Cohen (1988) proposed guidelines for social sciences; medical research often requires much higher r to be practically meaningful.

−1 (Perfect Neg.)−0.50+0.5+1 (Perfect Pos.)

Very Strong Neg. Weak/Mod. None Weak/Mod. Very Strong Pos.

\|r\| Range	Cohen (Social Science)	Medical/Clinical	Physical Sciences
0.8 – 1.0	Very Strong	Very Strong	Often expected
0.6 – 0.8	Strong	Moderate-Strong	Moderate
0.4 – 0.6	Moderate	Moderate	Weak-Moderate
0.2 – 0.4	Weak	Weak	Very Weak
0.0 – 0.2	Very Weak / None	Negligible	Essentially zero

💡 Field Context Matters: In psychology and social sciences, r = 0.30 is often considered a respectable finding. In physics or engineering, r = 0.99 might be the minimum acceptable. Always interpret correlation in the context of your field and the complexity of the phenomenon being measured.

Assumptions of Pearson's Correlation

📏 1. Linearity

The relationship between X and Y must be approximately linear. Pearson's r measures only linear association. If the true relationship is curved (e.g., quadratic or exponential), r will underestimate the true relationship strength. Check with a scatterplot.

🔔 2. Bivariate Normality

Both X and Y should be approximately normally distributed. This is important mainly for hypothesis testing (significance of r) and confidence intervals. With large samples (n ≥ 30), the Central Limit Theorem makes this assumption less critical.

📐 3. Homoscedasticity

The variance (spread) of Y values should be roughly constant across all levels of X — this is called homoscedasticity (equal scatter). If the data "fans out" as X increases (heteroscedasticity), Pearson's r may be misleading.

🚫 4. No Extreme Outliers

Outliers can dramatically inflate or deflate r. A single extreme data point can cause a genuinely uncorrelated dataset to appear strongly correlated, or wash out a true strong correlation. Always inspect your scatterplot for outliers before reporting r.

🔗 5. Paired Observations

Each X value must be paired with exactly one Y value — they must come from the same observational unit. You cannot pair 10 random X values with 10 random Y values from different subjects and expect r to be meaningful.

🎲 6. Independence

For valid statistical inference, each pair (xᵢ, yᵢ) should be independent of all other pairs. Time-series data or repeated measures on the same individual can violate this, inflating or deflating the apparent correlation.

Is My Correlation Statistically Significant? — The t-Test for r

A correlation r computed from a sample may differ from zero purely by chance. The significance test asks: "If the true population correlation ρ = 0, how likely am I to observe an r at least this extreme by random sampling?"

✦ t-Test Statistic for Pearson r

\[t = \frac{r\sqrt{n-2}}{\sqrt{1-r^2}} \quad \sim\; t_{n-2}\]

Fisher's z-transformation for confidence intervals: \(z_r = \frac{1}{2}\ln\!\left(\frac{1+r}{1-r}\right) = \text{arctanh}(r)\). This transforms r to a nearly normal variable: \(z_r \sim \mathcal{N}\!\left(\text{arctanh}(\rho),\; 1/\sqrt{n-3}\right)\), allowing the construction of confidence intervals for r that can then be back-transformed via \(r = \tanh(z_r)\).

A key caution: with large samples, even tiny correlations become statistically significant. r = 0.05 with n = 2,000 is statistically significant at p < 0.05 — but explains only 0.25% of the variance, making it practically meaningless. Always report both the correlation value and the sample size, not just the p-value.

Pearson vs. Spearman vs. Kendall — Which Correlation to Use?

Property	Pearson r	Spearman ρ	Kendall τ
Data type	Continuous, Interval/Ratio	Ordinal or continuous	Ordinal or continuous
Relationship type	Linear only	Monotonic	Monotonic
Normality required?	For inference, yes	No (non-parametric)	No (non-parametric)
Outlier robustness	Low	High	Very High
Interpretation	Most intuitive	Rank-based	Concordance-based
Typical \|value\| vs Pearson	—	Slightly lower	Noticeably lower
Best for small samples	If normal	OK	Best

✦ Spearman's Rank Correlation ρ

\[\rho = 1 - \frac{6\sum d_i^2}{n(n^2-1)}\]

dᵢ = rank(xᵢ) − rank(yᵢ) = difference in ranks for the i-th pair | This simplified formula applies when there are no tied ranks. With ties, the full Pearson formula applied to ranks is used instead.

✦ Kendall's τ

\[\tau = \frac{C - D}{\frac{1}{2}n(n-1)}\]

C = number of concordant pairs (both X and Y increase or both decrease together) | D = number of discordant pairs (one increases while the other decreases) | Denominator = total number of pair comparisons = n(n−1)/2

Correlation ≠ Causation & Anscombe's Quartet

The most important lesson in statistics: correlation measures association, not causation. Two variables can be strongly correlated for many reasons that have nothing to do with one causing the other.

Why Correlation ≠ Causation

🔀 Common Cause (Confounding)

Ice cream sales and drowning deaths are strongly positively correlated — both increase in hot weather. Neither causes the other; a third variable (temperature / summer) drives both.

🔄 Reverse Causation

If sick people take more medicine, correlation between medicine use and illness would appear positive — but illness caused the medicine intake, not the other way round.

🎲 Spurious Correlation

Per capita cheese consumption in the US correlates with deaths by bedsheet tangling (r ≈ 0.947). No real relationship exists — this is pure numerical coincidence over time.

Anscombe's Quartet — Why You Must Plot Your Data

In 1973, statistician Francis Anscombe created four famous datasets that all have nearly identical summary statistics — same mean of X, mean of Y, variance of X, variance of Y, correlation r ≈ 0.816, and the same regression line — yet look completely different on a scatterplot:

Dataset	r	What the Scatterplot Shows	Appropriate Analysis
I	0.816	Genuine linear relationship — Pearson r is appropriate	Linear regression
II	0.816	Perfect curved (quadratic) relationship — r is misleading	Polynomial regression
III	0.816	Perfect line except ONE extreme outlier — r highly distorted	Remove outlier, re-run
IV	0.817	All X values identical except one — not a real correlation	Do not use Pearson r

⚠️ The Lesson: Anscombe's Quartet proves that r alone can be profoundly misleading. Always create a scatterplot of your data before computing, reporting, or drawing conclusions from a correlation coefficient.

Real-World Applications of Correlation Analysis

📊 Finance & Economics

Portfolio diversification relies on asset correlation. Stocks with low or negative correlations reduce portfolio risk. The correlation matrix of asset returns is central to Modern Portfolio Theory (Markowitz, 1952).

🔬 Medical Research

Epidemiologists correlate risk factors (blood pressure, cholesterol) with health outcomes (heart attacks, strokes). Correlation informs which variables to include in multivariate regression models predicting disease risk.

🤖 Machine Learning

Correlation matrices help identify redundant features in datasets. Highly correlated features can be dropped to reduce dimensionality and prevent multicollinearity in linear models. Correlation is also used in feature selection for predictive modelling.

📚 Educational Research

Researchers correlate study habits, socioeconomic factors, and teaching methods with student performance. r helps identify which inputs most strongly predict outcomes, guiding curriculum design and resource allocation.

🌡️ Climate Science

Climate scientists correlate atmospheric CO₂ concentration with global mean temperature, sea ice extent, and ocean pH. Long-term correlations over decades provide evidence for climate trends.

🏭 Quality Control

Manufacturing engineers correlate process parameters (temperature, pressure, speed) with product quality metrics to identify which process variables most affect output quality, enabling targeted process improvements.

Worked Examples — Pearson r Step by Step

Example 1 — Small Dataset: Study Hours vs. Test Score

Data: X = [2, 3, 5, 7, 8] (study hours), Y = [40, 55, 65, 80, 90] (test score %)

n = 5. Σx = 2+3+5+7+8 = 25. Σy = 40+55+65+80+90 = 330.

Σxy = (2×40)+(3×55)+(5×65)+(7×80)+(8×90) = 80+165+325+560+720 = 1850.

Σx² = 4+9+25+49+64 = 151. Σy² = 1600+3025+4225+6400+8100 = 23350.

Numerator = n·Σxy − Σx·Σy = 5×1850 − 25×330 = 9250 − 8250 = 1000.

Denom Part 1 = n·Σx² − (Σx)² = 5×151 − 625 = 755 − 625 = 130.

Denom Part 2 = n·Σy² − (Σy)² = 5×23350 − 108900 = 116750 − 108900 = 7850.

Denominator = √(130 × 7850) = √1,020,500 ≈ 1010.2.

r = 1000 / 1010.2 ≈ 0.9899. R² = 0.9801 (98.0% of variance in test score explained by study hours).

✅ r ≈ 0.990 — Very Strong Positive linear relationship. Enter this data above to verify.

Example 2 — Negative Correlation: Temperature vs. Hot Drink Sales

Data: X = [5, 10, 15, 20, 25, 30] (°C), Y = [120, 95, 70, 50, 30, 15] (units sold)

n=6, Σx=105, Σy=380, Σxy = (5×120)+(10×95)+(15×70)+(20×50)+(25×30)+(30×15) = 600+950+1050+1000+750+450 = 4800.

Σx²=5²+10²+15²+20²+25²+30² = 25+100+225+400+625+900 = 2275. Σy²=14400+9025+4900+2500+900+225 = 31950.

Num = 6×4800 − 105×380 = 28800 − 39900 = −11100. (Negative → negative correlation ✓)

D1 = 6×2275 − 105² = 13650 − 11025 = 2625. D2 = 6×31950 − 380² = 191700 − 144400 = 47300.

Denom = √(2625 × 47300) = √124,162,500 ≈ 11142. r = −11100/11142 ≈ −0.9962.

✅ r ≈ −0.996 — Very Strong Negative. R² ≈ 0.992 (99.2% explained). Temperature strongly predicts hot drink sales.

Example 3 — Weak Correlation: Shoe Size vs. IQ

Data: X = [7, 8, 9, 10, 11] (shoe size UK), Y = [105, 98, 112, 101, 108] (IQ score)

n=5, Σx=45, Σy=524, Σxy = 735+784+1008+1010+1188 = 4725. Σx²=339, Σy²=54954.

Num = 5×4725 − 45×524 = 23625 − 23580 = 45.

D1 = 5×339 − 2025 = 1695 − 2025 = −330 ... wait, correction: (45)² = 2025, 5×339=1695, D1=1695−2025 = ... D1 should be positive. Re-check: Σx² = 49+64+81+100+121 = 415. D1 = 5×415 − 2025 = 2075−2025 = 50.

Σy² = 11025+9604+12544+10201+11664 = 55038. D2 = 5×55038 − 524² = 275190 − 274576 = 614.

Denom = √(50 × 614) = √30700 ≈ 175.2. r = 45/175.2 ≈ 0.257.

✅ r ≈ 0.257 — Very Weak positive correlation. R² ≈ 0.066 (only 6.6% of IQ variance associated with shoe size). Essentially no meaningful relationship.

Example 4 — Significance Test: Is r = 0.65 with n = 15 significant at p < 0.05?

Given: r = 0.65, n = 15. Degrees of freedom: df = n − 2 = 13.

\(t = r\sqrt{n-2}/\sqrt{1-r^2} = 0.65\times\sqrt{13}/\sqrt{1-0.4225} = 0.65\times 3.606/\sqrt{0.5775}\)

= 0.65 × 3.606 / 0.7600 = 2.344 / 0.7600 = 3.084.

Critical t at α=0.05 (two-tailed), df=13 is t_critical = 2.160. Since 3.084 > 2.160, we reject H₀: ρ=0.

✅ r = 0.65 with n=15 IS statistically significant at p < 0.05. The correlation is unlikely to be due to chance.

Frequently Asked Questions

What is the Pearson correlation coefficient?＋

Pearson's r measures the strength and direction of the linear relationship between two continuous variables. It ranges from −1 (perfect negative linear relationship) to +1 (perfect positive linear relationship), with 0 indicating no linear relationship. Developed by Karl Pearson in 1895, it is the most widely used correlation measure in statistics, research, data analysis, and machine learning.

What is the formula for Pearson's correlation coefficient?＋

The computational formula is: \(r = \frac{n\Sigma xy - \Sigma x\Sigma y}{\sqrt{[n\Sigma x^2-(\Sigma x)^2][n\Sigma y^2-(\Sigma y)^2]}}\). The conceptual formula is: \(r = \frac{S_{xy}}{\sqrt{S_{xx}S_{yy}}}\) where S terms are corrected sums of squares/cross-products. Both are algebraically identical. This calculator uses the computational formula because it avoids computing means first.

What do different r values mean?＋

|r| ≥ 0.8: Very Strong; 0.6–0.8: Strong; 0.4–0.6: Moderate; 0.2–0.4: Weak; 0–0.2: Very Weak/None. The sign indicates direction: positive r means both variables increase together; negative r means one increases as the other decreases. These thresholds are conventions — the appropriate benchmark depends on your field (see the comparison table above).

What is the coefficient of determination R²?＋

R² = r². It represents the proportion of variance in Y that is linearly associated with variance in X, expressed as a percentage. Example: r = 0.8 → R² = 0.64 → 64% of variance in Y is "explained" by X. The remaining 36% is due to other factors. R² is more directly interpretable than r for communicating practical effect size.

Does correlation imply causation?＋

No — this is the most important caution in statistics. Correlation only establishes statistical association, not causation. Two variables can be correlated because: (1) X causes Y; (2) Y causes X; (3) a third variable causes both; or (4) it's pure coincidence (spurious correlation). Only a properly designed randomised controlled experiment can establish causation. Always think critically about the mechanism before interpreting correlation as evidence of cause and effect.

What are the assumptions of Pearson's correlation?＋

The main assumptions are: (1) Linearity — the true relationship between X and Y is linear; (2) Bivariate normality — both variables approximately normally distributed (for inference); (3) Homoscedasticity — consistent spread of Y values across X; (4) No significant outliers — extreme values can distort r dramatically; (5) Paired observations — each x paired with exactly one corresponding y; (6) Independence — observations are independent of each other.

When should I use Spearman instead of Pearson?＋

Use Spearman's ρ (rank correlation) when: (1) your data are ordinal (ranked but not truly continuous); (2) data are not normally distributed; (3) the relationship is monotonic but not linear; (4) there are significant outliers that you cannot remove. Spearman's ρ ranks both variables and then applies Pearson's formula to the ranks — making it non-parametric and more robust. Formula: \(\rho = 1 - 6\Sigma d^2/[n(n^2-1)]\) where d = rank difference for each pair.

What is the difference between correlation and covariance?＋

Covariance \(Cov(X,Y) = \Sigma[(x_i-\bar{x})(y_i-\bar{y})]/(n-1)\) measures how two variables vary together in original data units — making it scale-dependent and hard to compare across studies. Pearson's r = Cov(X,Y)/(s_x · s_y) is the standardised version — dimensionless, always in [−1,+1], and directly comparable across different scales and datasets. Think of r as "covariance measured in standard deviation units."

How do I test if a correlation is statistically significant?＋

Use the t-test: \(t = r\sqrt{n-2}/\sqrt{1-r^2}\) with df = n−2. Compare the computed t to the critical value from the t-distribution at your significance level α. An alternative uses Fisher's z-transformation for confidence intervals. Important caveat: with very large n, even r = 0.05 becomes statistically significant — always report both r and n, and evaluate practical significance (R²) alongside p-value.

What is a scatterplot and why is it important for correlation?＋

A scatterplot plots each (x, y) pair as a point on a graph — X on the horizontal axis, Y on the vertical. It is the single most important step before computing correlation because it reveals: (1) whether the relationship is actually linear; (2) the presence of outliers; (3) non-linear patterns (curves) that r would miss; (4) data clusters or structural breaks. Anscombe's Quartet (1973) famously shows four datasets with identical r = 0.816 but completely different scatterplot patterns — only one of which is truly appropriate for Pearson's r.

What is Anscombe's Quartet?＋

Anscombe's Quartet (1973) is four datasets with nearlyidentical means, variances, correlation r ≈ 0.816, and regression lines — yet completely different scatterplot patterns. Dataset I: genuinely linear. Dataset II: perfect quadratic curve. Dataset III: perfectly linear except one outlier. Dataset IV: all X identical at 8 except one point at 19. The lesson: never rely on summary statistics alone — always plot your data before computing and reporting correlation.

What is Kendall's tau and when is it used?＋

Kendall's τ = (C−D)/[n(n−1)/2] where C = concordant pairs and D = discordant pairs. It is a rank-based correlation that counts how often pairs rank in the same order across both variables. Compared to Spearman's ρ, Kendall's τ produces smaller absolute values but has more accurate p-values for small samples and handles tied values better. Use Kendall's τ for small samples, heavily tied data, or when you need the most robust rank correlation available.