What is Pearson Correlation Coefficient? Definition, Formula, Assumptions, Interpretation, and Practical Applications

Getting your Trinity Audio player ready...

Contents

• Glossary of Key Terms
• Key Takeaways
• What Is the Pearson Correlation Coefficient?
• The Pearson Correlation Coefficient Formula
• How to Interpret the Pearson Correlation Coefficient
• What Are the Assumptions of Pearson’s Correlation Coefficient?
• The Role of the Scatterplot in Pearson Correlation Analysis
• Testing the Statistical Significance of r
• Key Properties of Pearson’s r
• How to Run a Pearson Correlation in SPSS
• Alternatives to Pearson’s r
• Variants and Extensions of the Pearson Correlation Coefficient
• Practical Examples of Pearson Correlation
• Limitations of Pearson’s r
• How to Report Pearson’s r Correctly
• Correlation vs. Regression: Key Differences
• Frequently Asked Questions

Glossary of Key Terms

The following definitions apply throughout this guide and are provided for quick reference before reading.

Term	Definition
Pearson correlation coefficient (r)	A numerical measure of the strength and direction of the linear relationship between two continuous variables, ranging from -1 to +1.
Bivariate correlation	Correlation involving exactly two variables; Pearson’s r is the most common form of bivariate correlation for continuous data.
Covariance	A measure of how much two variables change together; the Pearson r is the covariance normalized by the product of the two standard deviations.
Linear relationship	A relationship between two variables that can be represented as a straight line on a scatterplot; Pearson’s r measures only linear association.
Positive correlation	Both variables tend to increase or decrease together; r is greater than zero.
Negative correlation	As one variable increases, the other tends to decrease; r is less than zero.
Homoscedasticity	The spread (variance) of one variable remains roughly constant across all values of the other variable; a required assumption of Pearson’s r.
Scatterplot	A graph that plots pairs of values for two variables, used to visually inspect linearity, direction, strength, and outliers before computing r.
Coefficient of determination (r squared)	The square of the Pearson r; represents the proportion of variance in one variable that is explained by the other.
Null hypothesis (H0)	In correlation testing, the claim that the true population correlation is zero (no linear relationship).
Spearman’s rho	A non-parametric alternative to Pearson’s r that uses ranked data instead of raw values; appropriate when normality or linearity is violated.
Restricted range	A situation where the data represent only part of the possible range of a variable, which can artificially reduce the observed correlation.
Partial correlation	A variant of Pearson’s r that measures the relationship between two variables while statistically controlling for the effect of one or more additional variables.
Causation	A cause-and-effect relationship between two variables; correlation alone, however strong, does not establish causation.

Key Takeaways

• The Pearson correlation coefficient (r) measures the strength and direction of the linear relationship between two continuous variables on a scale from -1 to +1.
• A value of +1 indicates a perfect positive linear relationship; -1 indicates a perfect negative linear relationship; 0 indicates no linear relationship.
• Pearson’s r requires both variables to be continuous (interval or ratio level), linearly related, approximately normally distributed, homoscedastic, and independently observed.
• The sign of r shows direction; the magnitude shows strength. Conventional benchmarks: values below 0.3 are weak; 0.3 to 0.5 are moderate; above 0.5 are strong.
• Squaring r gives the coefficient of determination (r squared), which states the percentage of variance in one variable explained by the other.
• Pearson’s r measures association, not causation. A strong r does not mean one variable causes changes in the other.
• Outliers, non-linearity, and restricted range can all distort r, making visual inspection of a scatterplot an essential first step.
• When assumptions are violated, use Spearman’s rho (non-normal or ordinal data), Kendall’s tau (small samples with ties), or point-biserial correlation (one binary variable).
• Statistical significance of r is tested with a t-test; a significant p-value means the correlation is unlikely to be zero in the population, but does not reflect practical importance.
• SPSS, R, Python (SciPy), Excel, and most major statistical packages can compute Pearson’s r with a few commands or clicks.

What Is the Pearson Correlation Coefficient?

The Pearson correlation coefficient, denoted r, is a statistical measure that quantifies both the strength and direction of the linear relationship between two continuous variables. It was developed by Karl Pearson in the early twentieth century, building on work by Francis Galton, and is sometimes called the Pearson product-moment correlation coefficient (PPMCC), bivariate correlation, or simply Pearson’s r.

The coefficient always falls between -1 and +1, inclusive. This bounded range makes r a scale-independent measure: it is not affected by the units of measurement of either variable. The same r value would result whether height is measured in inches or centimeters, for example.

Because of its simplicity, interpretability, and mathematical tractability, Pearson’s r is the most widely used correlation measure in statistics, appearing in research across medicine, psychology, economics, education, data science, and virtually every empirical discipline.

The Pearson Correlation Coefficient Formula

The formula for Pearson’s r can be expressed in several equivalent forms. The most common versions are the population formula and the sample formula.

Population Formula

rho = Cov(X,Y) / (sigma_X * sigma_Y)

Where rho (the Greek letter rho) is the population correlation coefficient, Cov(X,Y) is the covariance of X and Y in the population, and sigma_X and sigma_Y are the population standard deviations of X and Y.

Sample Formula (Computational Form)

r = [n(SUM xy) – (SUM x)(SUM y)] / SQRT{ [n(SUM x^2) – (SUM x)^2][n(SUM y^2) – (SUM y)^2] }

Symbol	Meaning
r	The sample Pearson correlation coefficient (the statistic we calculate from data)
n	The number of paired observations (data points)
SUM xy	The sum of the products of each paired x and y value
SUM x	The sum of all x values
SUM y	The sum of all y values
SUM x^2	The sum of the squared x values
SUM y^2	The sum of the squared y values

Alternative Sample Formula (Conceptual Form)

r = SUM[(x_i – x_bar)(y_i – y_bar)] / SQRT{ SUM[(x_i – x_bar)^2] * SUM[(y_i – y_bar)^2] }

This form makes the conceptual meaning explicit: r is the covariance of the two variables divided by the product of their standard deviations. Deviations of each data point from the mean are multiplied together and summed; the denominator scales the result to fall between -1 and +1 regardless of the original units.

Step-by-Step Calculation Procedure

1. Create a table with columns for x, y, xy, x squared, and y squared.
2. For each paired observation, compute the product (xy), the square of x, and the square of y.
3. Sum each column to obtain SUM x, SUM y, SUM xy, SUM x squared, and SUM y squared.
4. Count the number of pairs (n).
5. Substitute all values into the computational formula and calculate r.
6. Check the sign (positive or negative) and magnitude (0 to 1) to interpret the result.

Worked Numerical Example

Suppose a researcher records hours of study (X) and exam score (Y) for five students:

Student	Hours (X)	Score (Y)	XY
1	2	50	100
2	3	60	180
3	5	70	350
4	7	80	560
5	8	90	720
Totals	SUM x = 25	SUM y = 350	SUM xy = 1910

Additional values: SUM x squared = 151; SUM y squared = 25,100; n = 5.

r = [5(1910) – (25)(350)] / SQRT{ [5(151) – 625][5(25100) – 122500] }

r = [9550 – 8750] / SQRT{ [755 – 625][125500 – 122500] }

r = 800 / SQRT{ 130 * 3000 } = 800 / SQRT(390000) = 800 / 624.5 = 0.981

This result (r = 0.981) indicates a very strong positive linear relationship between study hours and exam scores in this sample.

How to Interpret the Pearson Correlation Coefficient

Interpreting r requires considering both its sign (direction) and its absolute value (strength). Neither element alone gives a complete picture.

Direction: What the Sign of r Tells You

Sign	Direction	Example
Positive (r > 0)	Both variables tend to move in the same direction: as X increases, Y tends to increase.	Height and weight; years of education and lifetime earnings.
Negative (r < 0)	The variables move in opposite directions: as X increases, Y tends to decrease.	Outdoor temperature and heating costs; hours of exercise and resting heart rate.
Zero (r = 0)	No linear relationship. The variables do not move together in a straight-line pattern.	Shoe size and intelligence test scores (in a general population).

Strength: Benchmarks for the Magnitude of r

Absolute Value of r	Conventional Strength Label	Direction
0.00	No relationship	None
0.01 to 0.29	Weak	Positive or negative
0.30 to 0.49	Moderate	Positive or negative
0.50 to 0.74	Strong	Positive or negative
0.75 to 0.99	Very strong	Positive or negative
1.00	Perfect linear relationship	Positive or negative

These benchmarks are conventions, not universal thresholds. What constitutes a “strong” correlation depends on the research context. In particle physics, r = 0.3 might be remarkable; in psychology survey research, r = 0.7 might be considered typical for related constructs.

The Coefficient of Determination: r Squared

Squaring r yields the coefficient of determination, written r squared. This tells you what proportion of the variance in one variable is statistically explained by the other.

r value	r squared	Variance Explained
0.30	0.09	9% of variance is shared
0.50	0.25	25% of variance is shared
0.70	0.49	49% of variance is shared
0.90	0.81	81% of variance is shared
1.00	1.00	100% of variance is shared (perfect)

For example, if r = 0.70 between study hours and exam scores, then r squared = 0.49, meaning that approximately 49% of the variation in exam scores can be attributed to variation in study hours. The remaining 51% is explained by other factors not in the model.

What Are the Assumptions of Pearson’s Correlation Coefficient?

Five assumptions must be met before Pearson’s r can be validly interpreted. Violations of these assumptions do not always prevent calculation, but they compromise the accuracy and meaning of the result.

Assumption	What It Requires	How to Check
1. Continuous, interval or ratio data	Both variables must be measured on an interval or ratio scale. Ordinal or nominal data are not appropriate for Pearson’s r.	Review the measurement scale of each variable. Use Spearman’s rho for ordinal data.
2. Linear relationship	The relationship between the two variables must be linear (expressible as a straight line). Pearson’s r does not capture curved or U-shaped relationships.	Inspect a scatterplot. If the pattern is curved, consider transforming the data or using a non-linear approach.
3. Approximate normality	Both variables should be approximately normally distributed in the population, particularly important for inferential tests (p-value for r). Mild departures are acceptable with larger samples.	Shapiro-Wilk test; Q-Q plot; histogram. With large samples the central limit theorem reduces sensitivity to this assumption.
4. Homoscedasticity	The variance of one variable should be approximately constant across all values of the other. Heteroscedasticity (funnel-shaped scatterplot) can distort r.	Inspect the scatterplot for a funnel or fan shape. Residual plots can also reveal heteroscedasticity.
5. Independence of observations	Each pair of observations must be independent of all other pairs. Repeated measures on the same participants or clustered data violate this assumption.	Review the study design. Use mixed models or paired analyses for dependent observations.

A critical additional consideration is the absence of influential outliers. A single extreme data point can substantially inflate or deflate r, masking the true pattern in the data. Outliers should always be identified and examined before reporting r; they may represent data entry errors, measurement problems, or genuinely unusual cases that warrant separate analysis.

The Role of the Scatterplot in Pearson Correlation Analysis

Inspecting a scatterplot before computing r is not optional; it is an essential part of the analysis. A scatterplot reveals information that r alone cannot communicate.

What to Look For	Why It Matters for Pearson’s r
Overall pattern	Confirms whether a linear model is appropriate. A curved or non-linear pattern means r will underestimate (or misrepresent) the true relationship.
Direction of the trend	Positive slope confirms positive r; negative slope confirms negative r. Verifies that r’s sign matches visual evidence.
Strength of the clustering	Tightly clustered points near a straight line suggest high r; widely scattered points suggest low r.
Outliers	Even one or two extreme outliers can dramatically change r. Identify and investigate them before including them in the final analysis.
Homoscedasticity	A roughly rectangular scatter band across all X values suggests homoscedasticity. A funnel shape signals heteroscedasticity.
Restricted range	If the X or Y values span only a small portion of their potential range, r may underestimate the true population correlation.

Testing the Statistical Significance of r

Pearson’s r is calculated from a sample. To determine whether the observed r reflects a true correlation in the population (rather than a result of sampling error), a significance test is applied.

The Hypothesis Test

• Null hypothesis (H0): The population correlation coefficient (rho) equals zero; there is no linear relationship.
• Alternative hypothesis (H1): The population correlation coefficient is not equal to zero (two-tailed); or is greater than zero, or less than zero (one-tailed).

The Test Statistic

t = r * SQRT(n – 2) / SQRT(1 – r^2)

This t-statistic follows a t-distribution with n minus 2 degrees of freedom under the null hypothesis. The resulting p-value indicates the probability of observing an r at least as extreme as the one computed if the true population correlation were zero.

Critical Values of r for Statistical Significance

Sample Size (n)	Critical r (alpha = 0.05, two-tailed)	Critical r (alpha = 0.01, two-tailed)
10	0.632	0.765
20	0.444	0.561
30	0.361	0.463
50	0.279	0.361
100	0.197	0.256
200	0.139	0.182

With very large samples, even a tiny r (e.g., 0.10) can be statistically significant, yet explain only 1% of the variance and have no practical importance. Statistical significance and practical significance are not the same thing.

Confidence Intervals for r

A 95% confidence interval for r should always accompany the point estimate. Because the sampling distribution of r is skewed (especially when r is far from zero), Fisher’s Z transformation is used to compute the interval, then the result is converted back to the r scale. Statistical software handles this automatically.

Key Properties of Pearson’s r

Property	Description
Bounded range	r always falls between -1 and +1, inclusive.
Symmetry	The correlation between X and Y equals the correlation between Y and X: r(X,Y) = r(Y,X).
Unit independence	Multiplying either variable by a positive constant does not change r. Adding a constant to either variable also leaves r unchanged.
Sensitivity to linear scale only	r measures only linear association. A curvilinear relationship can produce r = 0 even if a strong relationship exists.
No causation implied	r measures statistical association; it provides no evidence of whether X causes Y or Y causes X.
Effect of outliers	A single outlier can dramatically inflate or deflate r, making the scatterplot check essential.
Restricted range effect	When data are collected from a restricted part of the population’s range, r tends to underestimate the true population correlation.

How to Run a Pearson Correlation in SPSS

SPSS provides a straightforward menu path for computing Pearson’s r. The procedure is the same whether you are running a single bivariate correlation or a full correlation matrix.

Step-by-Step Procedure in SPSS

• Open your dataset in SPSS Data View.
• From the menu bar, navigate to Analyze, then Correlate, then Bivariate.
• Move the two (or more) variables of interest into the Variables box on the right.
• Under Correlation Coefficients, ensure Pearson is checked (it is checked by default).
• Under Test of Significance, select Two-tailed (or One-tailed if your hypothesis is directional).
• Check Flag significant correlations to have SPSS mark significant results with asterisks.
• Click OK.

Reading the SPSS Output Table

SPSS produces a correlation matrix, which is a square table. For a two-variable analysis, the relevant output is a 2×2 matrix.

Output Row	What It Shows
Pearson Correlation	The r value for each pair of variables. The diagonal always shows 1.000 (each variable correlates perfectly with itself).
Sig. (2-tailed)	The p-value for the two-tailed significance test of r. Values below 0.05 are flagged with one asterisk; values below 0.01 with two asterisks.
N	The number of pairs used in the calculation. If N differs across cells, some data may be missing.

APA-Style Reporting of SPSS Output

The standard format for reporting Pearson’s r in academic writing is:

r(df) = [r value], p = [p value]

Where df equals n minus 2. For example: r(48) = .72, p < .001, indicating a strong positive correlation between the two variables. Always report r squared alongside r in dissertation or thesis work.

Alternatives to Pearson’s r

When the assumptions of Pearson’s r cannot be met, or when the data structure calls for a different approach, several well-established alternatives are available.

Alternative	When to Use	Key Difference from Pearson’s r
Spearman’s rho	Ordinal data; non-normal distributions; presence of outliers; non-linear but monotonic relationships	Uses ranks of the data rather than raw values; measures monotonic (not only linear) association
Kendall’s tau	Small samples; ordinal data with many ties; when robustness is more important than efficiency	Based on concordant and discordant pairs; more conservative than Spearman’s rho
Point-biserial correlation	One continuous variable and one genuinely dichotomous (binary) variable	Mathematically equivalent to Pearson’s r when one variable is coded 0/1
Biserial correlation	One continuous variable and one artificially dichotomized variable (underlying normal distribution assumed)	Corrects for the artificiality of the dichotomization; larger than point-biserial
Partial correlation	When you want to measure the relationship between two variables while controlling for one or more third variables	Removes the effect of control variables from both X and Y before computing r
Phi coefficient	Both variables are genuinely dichotomous	Special case of Pearson’s r for 2×2 tables
Distance correlation	Detecting any form of statistical dependence, including non-linear and non-monotonic relationships	Equals zero if and only if the variables are truly independent; not bounded by linearity

Variants and Extensions of the Pearson Correlation Coefficient

The standard Pearson r has been extended in several ways to handle specific data structures or analytical goals.

Variant	Purpose and Use
Partial correlation	Computes r between X and Y after removing the linear effect of one or more control variables (Z). Useful for identifying direct relationships without confounding influences.
Semi-partial (part) correlation	Removes the effect of control variables from only one of the two variables. Used in multiple regression to understand the unique contribution of each predictor.
Weighted correlation	Assigns different weights to observations based on their reliability or importance, reducing the influence of less trustworthy data points.
Adjusted r (correction for bias)	Applies a correction formula to reduce the positive bias that can occur in small samples, producing a more accurate estimate of the population correlation.
Circular correlation	Used for data that are directional or cyclical (e.g., time of day, compass bearing, angle measurements), where standard linear correlation is not meaningful.
Correlation matrix	A square table showing Pearson’s r for all possible pairs of variables in a dataset. Useful for exploring multivariate datasets and identifying patterns before regression or factor analysis.

Practical Examples of Pearson Correlation

The following examples illustrate the range of contexts in which Pearson’s r is applied.

Field	Variable X	Variable Y	Typical r and Interpretation
Medicine	Daily sodium intake (mg)	Systolic blood pressure (mmHg)	r = 0.45; moderate positive correlation
Education	Hours of study per week	Final exam score	r = 0.70; strong positive correlation
Economics	GDP per capita (USD)	Life expectancy (years)	r = 0.80; very strong positive correlation
Psychology	Perceived stress score	Sleep quality index	r = -0.55; strong negative correlation
Sports science	Body fat percentage	VO2 max (aerobic capacity)	r = -0.65; strong negative correlation
Marketing	Advertising spend (USD)	Monthly revenue (USD)	r = 0.60; strong positive correlation

Limitations of Pearson’s r

Understanding what Pearson’s r cannot do is as important as knowing what it can.

• Correlation is not causation. A high r between two variables does not establish that one causes the other. Both could be caused by a third, unmeasured variable (a confounder or “lurking variable”).
• Only linear relationships: Pearson’s r can be exactly zero even when a strong, consistent relationship exists between X and Y, if that relationship is curved or non-linear.
• Sensitivity to outliers: Even one extreme observation can dramatically distort r. This makes outlier detection and scatterplot inspection essential.
• Restricted range bias: If the sample does not represent the full range of the population, r will be attenuated (pulled toward zero), underestimating the true population correlation.
• Not appropriate for non-continuous data: Using Pearson’s r with ordinal (ranked) or nominal (categorical) variables produces misleading results.
• Does not indicate proportionality: r = 0.80 is not twice as strong as r = 0.40 in any simple linear sense. The coefficient of determination (r squared) is the more interpretable measure of explained variance.
• Sample size sensitivity in significance testing: With large samples, even trivially small correlations become statistically significant; the p-value must be considered alongside the magnitude of r and r squared.

How to Report Pearson’s r Correctly

Complete reporting of Pearson’s r is a common weakness in research. Committees and journal reviewers expect more than r and p.

What to Include in a Full Report

• The r value, rounded to two decimal places (e.g., r = .72, not r = 0.724).
• The degrees of freedom in parentheses: r(df) format, e.g., r(48) = .72.
• The p-value: either exact (p = .003) or as a comparison (p < .001).
• The coefficient of determination r squared, with a statement of what percentage of variance is explained.
• The direction and strength of the relationship in plain language.
• A confidence interval for r (95% CI) when reporting for publication.
• A note on sample size and whether assumptions were verified.

Example of Correct APA Reporting

A Pearson correlation was computed to assess the relationship between weekly study hours and final exam scores. There was a strong positive correlation between the two variables, r(48) = .72, p < .001, 95% CI [.56, .83]. Study hours accounted for approximately 52% of the variance in exam scores (r squared = .52).

Correlation vs. Regression: Key Differences

Correlation and regression both describe relationships between variables, but they answer fundamentally different questions. Pearson’s r asks how strongly two variables are linearly associated; linear regression asks how much Y changes for a given change in X, and enables prediction.

The two analyses are closely related mathematically: in simple linear regression with one predictor, the standardized regression coefficient (beta) equals Pearson’s r exactly. Despite this, their purposes, outputs, and interpretive implications differ in important ways that researchers must understand before choosing between them.

Pearson Correlation vs. Linear Regression

Feature	Pearson Correlation (r)	Simple Linear Regression
Primary question	How strongly and in what direction are X and Y linearly associated?	How much does Y change for each one-unit increase in X? What value of Y does a given X predict?
Output	A single coefficient r between -1 and +1; r squared	An intercept (a), a slope (b), standard error of the estimate, and r squared
Symmetry	Symmetric: r(X,Y) = r(Y,X). It does not matter which variable is X and which is Y.	Asymmetric: regressing Y on X gives a different slope than regressing X on Y. The choice of dependent variable matters.
Role of variables	Neither variable is formally designated as predictor or outcome; both are treated equally.	One variable is explicitly the predictor (independent) and one is the outcome (dependent).
Units	Unit-free: r has no units and is directly comparable across studies regardless of measurement scale.	The slope b carries units (units of Y per unit of X) and cannot be compared across studies without standardization.
Prediction	Cannot generate predicted values of Y for a given X.	Produces a prediction equation (Y-hat = a + bX) that estimates Y for any given value of X.
Multiple variables	Extended to a correlation matrix when more than two variables are present; each pair is assessed separately.	Extends naturally to multiple regression, allowing several predictors to be assessed simultaneously in one model.
Shared metric	r squared indicates the proportion of shared variance between X and Y.	r squared (R squared in regression notation) indicates the proportion of variance in Y explained by the model.
Causal assumption	No causal direction is assumed or implied.	A predictor-outcome structure is assumed, though causation still cannot be inferred without experimental design.
Sensitivity to outliers	Highly sensitive: one outlier can substantially shift r.	Also sensitive: outliers exert leverage on the slope and intercept, especially if extreme on the X axis.
When to prefer it	When the goal is to quantify association strength for comparison or hypothesis testing, with no designated predictor or outcome.	When the goal is prediction, estimation of effect size in units, or modeling the influence of multiple variables on an outcome.

A common source of confusion: a significant Pearson r and a significant regression slope always appear together in simple linear regression because they share the same t-test. The distinction becomes meaningful when moving to multiple regression, where r is a bivariate statistic but regression coefficients are partial effects controlling for other predictors.

• Use Pearson’s r when: both variables are continuous; you want a symmetric, scale-free measure of linear association; you are comparing relationship strength across different pairs of variables.
• Use linear regression when: one variable is conceptually the predictor; you want to predict Y from X; you need to express the effect in original units; or you plan to extend the analysis to multiple predictors.

Frequently Asked Questions

What Does a Pearson Correlation Coefficient of 0 Mean?

A value of r = 0 means there is no linear relationship between the two variables. This does not mean there is no relationship at all. A strong curvilinear relationship (such as a U-shape) can produce r = 0 because Pearson’s r only measures how well a straight line describes the association. Always examine a scatterplot: if the data show a clear pattern that is not a straight line, r = 0 is misleading rather than informative.

Can Pearson’s r Be Used With Ordinal Data?

No; Pearson’s r requires continuous data measured on an interval or ratio scale. Ordinal data (such as Likert scale responses) have an ordering but no guarantee of equal spacing between categories. Using Pearson’s r with ordinal data can produce incorrect results. Spearman’s rho is the appropriate alternative: it converts raw values to ranks and is designed for ordinal measurement levels.

Does a High Pearson r Mean One Variable Causes the Other?

No. Correlation does not imply causation. Even a perfect correlation of r = 1.00 provides no evidence that X causes Y. The association could reflect: the reverse direction (Y causes X); a third variable causing both X and Y simultaneously; coincidence in the sample; or a confounding variable not measured in the study. Establishing causation requires experimental designs with random assignment, or rigorous causal inference methods such as instrumental variables or regression discontinuity.

What Is the Difference Between Pearson’s r and Spearman’s Rho?

Pearson’s r measures the strength and direction of the linear relationship between two continuous variables, using the raw values directly. Spearman’s rho is a non-parametric measure that converts both variables to ranks and then computes Pearson’s r on those ranks. Spearman’s rho is therefore more robust to non-normality, outliers, and non-linear (but monotonic) relationships. The two statistics give similar results when the data are approximately normal and the relationship is linear; they can diverge substantially when outliers are present or the relationship is monotonic but not strictly linear.

How Does Sample Size Affect Pearson’s r?

Sample size affects Pearson’s r in two important ways. First, the reliability of r as an estimate of the population correlation increases with sample size: with a very small sample (n less than 10), even a large observed r may not replicate. Second, the statistical significance of r is directly influenced by n: with large samples (n > 200), even a very small r (such as 0.10) may be statistically significant at alpha = 0.05, yet explain only 1% of the variance. Reporting r squared alongside the p-value guards against over-interpreting significance from large samples.

What Should I Do If My Data Contain Outliers?

Outliers should be identified before computing r, using the scatterplot and standardized residuals. Once identified, the recommended approach is: first, verify that the outlier is not a data entry error or measurement mistake; second, report r both with and without the outlier to show its influence; third, consider whether the outlier represents a genuinely different subgroup that should be analyzed separately. Removing outliers arbitrarily to improve r is not acceptable practice. When outliers are a persistent feature of the data, Spearman’s rho provides a more robust correlation estimate.

What Is the Difference Between Pearson’s r and the Regression Slope?

Pearson’s r and the regression slope (b) both describe the linear relationship between two variables, but they measure different things. Pearson’s r is standardized: it has no units and always falls between -1 and +1, making it directly comparable across different studies and variables. The regression slope b is unstandardized: it tells you how many units Y changes for each one-unit increase in X, and its value depends on the scales of both variables. The standardized regression coefficient (beta) in simple linear regression equals Pearson’s r exactly. The relationship is: b = r multiplied by (standard deviation of Y divided by standard deviation of X).

How Do I Handle Multiple Correlations Without Inflating Type I Error?

When computing Pearson’s r for many pairs of variables simultaneously, the probability of at least one spuriously significant result increases with the number of tests. For example, computing 20 correlations at alpha = 0.05 yields an expected one false positive by chance alone. Common approaches to this multiple comparisons problem include: the Bonferroni correction (divide alpha by the number of tests; e.g., use 0.05/20 = 0.0025 as the significance threshold); the false discovery rate (FDR) approach, which controls the expected proportion of false positives among significant results; and limiting the number of planned comparisons based on a priori hypotheses rather than exploratory mining. Reporting the full correlation matrix transparently, with a note that results should be interpreted cautiously, is also recommended.