BIOSTATISTICS • DESCRIPTIVE STATISTICS & VISUALIZATION

Interpreting Scatterplots — Interpret scatterplots and association patterns

Uncover the direction, form, and strength of bivariate relationships through systematic visual analysis of paired data.

Historical Context & Motivation

The impulse to plot one measurement against another is surprisingly modern. Although astronomers and surveyors recorded paired observations for centuries, the idea of using a two-dimensional plane to reveal the relationship between two variables did not crystallize until the late eighteenth century, when scientists began to suspect that visual patterns could expose regularities invisible in tabular data. The scatterplot—a graph in which each observation is represented as a point defined by its x- and y-coordinates—emerged from this tradition and became one of the most fundamental tools in statistics and the biological sciences.

1686
First Known Bivariate Graph
Edmund Halley plotted barometric pressure against altitude in a seminal attempt to visualize the relationship between two measured quantities, laying conceptual groundwork for bivariate displays.
1833
Herschel's Stellar Scatterplot
John Herschel plotted orbital positions of double stars, creating what many historians consider the first modern scatterplot, with individual data points marked on Cartesian axes.
1886
Galton & Regression
Francis Galton used scatterplots of parent and offspring heights to discover 'regression toward mediocrity,' demonstrating how a cloud of points could reveal a systematic trend and introducing the concept of correlation.
1901
Pearson's Correlation Coefficient
Karl Pearson formalized the product-moment correlation coefficient r, providing a numerical summary of the linear association pattern visible in scatterplots and establishing the foundation for modern bivariate statistics.
1977
Tukey's EDA Revolution
John Tukey's Exploratory Data Analysis championed graphical methods—including enhanced scatterplots with smoothers—as essential first steps before formal modeling, influencing biostatistical practice for decades.

In contemporary biostatistics, scatterplots serve as the front line of analysis whenever a researcher asks whether two continuous variables—such as drug dosage and biomarker response, or body mass index and blood pressure—are associated. The ability to read direction, form, strength, and outliers from a cloud of points remains an indispensable skill, one that precedes and informs every regression model, every hypothesis test, and every causal inference strategy a biostatistician will employ.

Core Principles & Definitions

Interpreting a scatterplot requires a systematic vocabulary. Every scatterplot encodes a set of paired observations (xi, yi), and the spatial arrangement of those points conveys information along four key dimensions: direction, form, strength, and the presence of unusual observations. Understanding each of these dimensions—and how they interact—constitutes the foundation of scatterplot literacy in biological and health sciences research.

1

Direction (Positive / Negative / None)

A positive association means y tends to increase as x increases. A negative association means y decreases as x increases. No direction implies no systematic trend.
2

Form (Linear / Nonlinear / Clustered)

Points may trace a roughly straight line (linear), follow a curve such as exponential or logistic (nonlinear), or separate into distinct groups.
3

Strength (Strong / Moderate / Weak)

Strength refers to how tightly points cluster around the underlying trend. A strong association shows little scatter about the trend; a weak one shows substantial vertical spread at each x-value.
4

Unusual Observations

Outliers deviate markedly from the overall pattern. Influential points (especially high-leverage observations) can disproportionately alter a fitted regression line or correlation coefficient.
KEY TAKEAWAY
Think of a scatterplot as a photograph of the joint behavior of two variables. Just as you might describe a photograph by noting the subject (direction), the composition (form), the focus (strength), and any anomalies (outliers), you systematically describe a scatterplot by the same four features—before you ever fit a model or compute a statistic.

Visual Explanation — Anatomy of a Scatterplot

The diagram below illustrates the anatomy of a scatterplot using simulated data that might represent the relationship between a continuous predictor—such as weekly exercise hours—and a continuous response—such as resting heart rate. Each labeled element corresponds to a feature you should evaluate every time you encounter a scatterplot in a biostatistics context.

Each cyan dot represents one individual's paired data point (exercise hours, heart rate). The dashed gradient line is the fitted trend line, revealing a strong negative linear association. The red-circled point is an outlier—a person who exercises very little yet has an unusually low resting heart rate, deviating from the overall pattern.

When you examine any scatterplot, begin by scanning the overall cloud from left to right: does it rise (positive), fall (negative), or remain flat (no direction)? Next, ask whether the trend appears to follow a straight path or a curve. Then assess how tightly the points cluster around that trend—this is the strength of the association. Finally, identify any points that break the pattern: outliers that lie far from the cloud, or influential points at the extremes of the x-range that could unduly affect a fitted line.

Mathematical Framework — Quantifying Association

While scatterplot interpretation is fundamentally visual, biostatisticians formalize what the eye detects using numerical summaries. The most common measure for linear scatterplots is the Pearson correlation coefficient r, which captures both direction and strength on a scale from −1 to +1. When the form is nonlinear, Spearman's rank correlation ρ or other measures may be more appropriate, but Pearson's r is the starting point for most analyses and connects directly to the visual pattern of a scatterplot.

PEARSON CORRELATION COEFFICIENT
r = Σ[(xᵢ − x̄)(yᵢ − ȳ)] / √{Σ(xᵢ − x̄)² × Σ(yᵢ − ȳ)²}
where xi and yi are individual observations, x̄ and ȳ are sample means. Values near +1 or −1 indicate strong linear association; values near 0 indicate weak or no linear association.
COEFFICIENT OF DETERMINATION
R² = r²
R² represents the proportion of variance in y explained by its linear relationship with x. For instance, r = −0.94 yields R² = 0.88, meaning 88% of the variation in resting heart rate is accounted for by exercise hours in the model.
LEAST-SQUARES REGRESSION LINE
ŷ = b₀ + b₁x, where b₁ = r × (sᵧ / sₓ) and b₀ = ȳ − b₁x̄
The slope b1 quantifies the average change in y per one-unit increase in x. The sign of b1 matches the sign of r, linking the regression line directly to the direction of association visible in the scatterplot.
⚠️ IMPORTANT CAVEAT
Pearson's r measures only linear association. A scatterplot showing a perfect quadratic (U-shaped) relationship can yield r ≈ 0. This is precisely why visual inspection of the scatterplot must always precede numerical summarization—the plot reveals the form of association, which r alone cannot.

Detailed Breakdown — Association Patterns in Biostatistics

In biological and health sciences data, association patterns rarely fall into a single neat category. The diagram below presents six canonical scatterplot patterns you will encounter repeatedly in biostatistics coursework and research practice. Each panel illustrates a distinct combination of direction, form, and strength, and understanding these archetypes will help you rapidly classify real-world scatterplots.

Six archetypal scatterplot patterns: (A) strong positive linear, (B) strong negative linear, (C) no association, (D) quadratic (U-shaped), (E) exponential growth, and (F) clustered subgroups. Note how Pearson's r can be misleading for patterns D–F, reinforcing the necessity of always plotting data before computing summary statistics.

Panel D deserves special attention in biostatistics because many dose–response relationships follow a nonlinear pattern: low and high doses of a substance may both produce elevated responses (hormesis), while intermediate doses do not. Computing r on such data yields a value near zero, falsely suggesting no association. Similarly, Panel F illustrates Simpson's paradox territory—the overall correlation may differ from the within-group correlations when confounding subpopulations exist, a common scenario in epidemiological data where age, sex, or treatment groups create natural clusters.

Common association patterns with biostatistics examples
PatternBiostatistics ExampleTypical r Range
Strong positive linearHeight vs. arm span in adults+0.85 to +0.99
Strong negative linearAge vs. FEV₁ (lung function) in older adults−0.80 to −0.95
Weak / noneShoe size vs. cholesterol level−0.10 to +0.10
Nonlinear (quadratic)Core body temperature vs. enzyme activityr ≈ 0 (misleading)
Nonlinear (exponential)Time vs. bacterial colony count in log phaser > 0 but underestimates
Clusters / subgroupsBMI vs. blood glucose with diabetic and non-diabetic groupsVaries by group

Worked Example — Drug Dosage and Serum Concentration

A pharmacologist administers varying oral doses (mg) of a new analgesic to 8 volunteer subjects and measures the resulting peak serum concentration (µg/mL) 2 hours post-administration. The paired data are: (10, 1.2), (20, 2.5), (30, 3.1), (40, 4.4), (50, 5.0), (60, 6.3), (70, 7.1), (80, 8.0). We will interpret the scatterplot and compute r to quantify the association.

Interpreting a Dose–Response Scatterplot
1
Step 1 — Identify Variables and RolesThe explanatory variable (x) is drug dosage in mg, and the response variable (y) is peak serum concentration in µg/mL. Dosage is placed on the horizontal axis because it is the variable the researcher controls.
2
Step 2 — Assess DirectionScanning the points from left to right, serum concentration increases as dosage increases. The association is positive.
3
Step 3 — Assess FormThe points fall roughly along a straight line with no obvious curvature. The relationship appears linear, consistent with first-order pharmacokinetics in this dose range.
4
Step 4 — Assess Strength and Compute rCalculate means: x̄ = (10 + 20 + … + 80)/8 = 45 mg; ȳ = (1.2 + 2.5 + … + 8.0)/8 = 4.70 µg/mL. Computing deviations: Σ(xi − x̄)(yi − ȳ) = (−35)(−3.50) + (−25)(−2.20) + … + (35)(3.30) = 560.0. Also, Σ(xi − x̄)² = 4200 and Σ(yi − ȳ)² = 76.34. Then r = 560.0 / √(4200 × 76.34) = 560.0 / 566.1 ≈ 0.989.
r ≈ +0.989 — very strong positive linear association
5
Step 5 — Check for Outliers and InterpretNo individual point deviates substantially from the linear trend; residuals are small and roughly symmetric. The R² = 0.978, meaning approximately 97.8% of the variation in serum concentration is explained by the linear relationship with dosage. We interpret: there is a very strong, positive, linear association between oral dosage and peak serum concentration in this study—each additional 10 mg of dosage corresponds to approximately a 0.97 µg/mL increase in peak concentration.
R² ≈ 0.978 — dosage explains ~98% of serum concentration variability

Strengths, Limitations, and Common Pitfalls

Strengths and limitations of scatterplot analysis
StrengthsLimitations
Reveals direction, form, and strength simultaneously without requiring assumptions about the data distribution.Can only display two variables at a time (without color/size encoding), limiting multivariate exploration.
Immediately exposes outliers, clusters, and nonlinear patterns that numerical summaries can miss.Overplotting occurs when many points share similar coordinates, obscuring density patterns in large datasets.
Intuitive and accessible to interdisciplinary audiences (clinicians, public health officials, non-statisticians).Subjective interpretation—two analysts may describe the same pattern slightly differently without formal numerical summaries.
Serves as an essential diagnostic for regression assumptions (linearity, homoscedasticity) via residual plots.Cannot establish causation. A strong scatterplot pattern confirms association, not that x causes y.
KEY TAKEAWAY
The most dangerous mistake in scatterplot interpretation is the temptation to skip the plot entirely and jump straight to r or a regression model. Anscombe's Quartet—four datasets with identical summary statistics (same means, standard deviations, and r = 0.816) but wildly different scatterplot patterns—stands as a permanent reminder that numbers without pictures can deceive. Always plot first, summarize second.

Connection to Advanced Methods

Scatterplot interpretation is the conceptual gateway to an entire family of advanced biostatistical methods. Every technique listed below has its roots in the same visual intuition you develop when reading a bivariate point cloud—the question is always how x relates to y, with increasing sophistication in how that relationship is modeled, adjusted, and tested.

From scatterplot interpretation to advanced biostatistical modeling
Concept in This LessonAdvanced ExtensionWhat Changes
Visual assessment of direction & formSimple linear regression (SLR)The trend line is formally estimated via least squares; inference (confidence intervals, p-values) is added.
Nonlinear scatterplot patternsPolynomial & nonlinear regression; GAMsCurved functional forms replace the straight line; generalized additive models (GAMs) use flexible smoothers.
Pearson's r and R²Multiple regression, partial correlationMultiple predictors are modeled simultaneously; partial correlations measure association after controlling for confounders.
Outlier detection (visual)Influence diagnostics (Cook's D, leverage, DFFITS)Formal numerical measures quantify each observation's impact on model estimates and predictions.
Cluster patterns / subgroupsStratified analysis; interaction terms; mixed-effects modelsGroup structure is explicitly modeled via indicator variables, interaction terms, or random effects for hierarchical data.

As you advance through biostatistics coursework, remember that even the most complex regression model produces residuals that should be plotted and inspected as a scatterplot. The skills you build now—recognizing linearity, detecting heteroscedasticity (fan-shaped spread), spotting outliers—are precisely the skills that validate or invalidate advanced models. Mastery of scatterplot interpretation is therefore not an introductory exercise to be outgrown; it is a permanent diagnostic discipline that accompanies every stage of statistical modeling.

Practice Problems

PROBLEM 1CONCEPTUAL
A researcher creates a scatterplot of patient age (x-axis) versus systolic blood pressure (y-axis) and observes that the cloud of points drifts upward from left to right in a roughly straight band. Describe the association using the four-feature framework (direction, form, strength, unusual observations) and state what additional information you would need to classify the strength more precisely.
PROBLEM 2BASIC CALCULATION
Given five paired observations—(2, 8), (4, 6), (6, 5), (8, 3), (10, 2)—compute the Pearson correlation coefficient r. Interpret the direction and strength of the association.
PROBLEM 3INTERMEDIATE
An epidemiologist plots daily particulate matter (PM₂.₅ in µg/m³) against hospital admissions for respiratory illness across 30 cities. The scatterplot shows a generally positive trend, but the spread of points around the trend widens considerably for cities with high PM₂.₅ values. (a) What statistical term describes this widening spread? (b) Does this pattern affect the validity of Pearson's r? (c) What remedial action might the analyst consider?
PROBLEM 4APPLIED
A clinical trial investigator plots drug concentration (ng/mL) versus tumor reduction (%) for 40 cancer patients. The scatterplot reveals two distinct clusters: one group of 25 patients with moderate drug concentrations and modest tumor reduction, and another group of 15 patients with high concentrations and large tumor reduction. Within each cluster, there appears to be little association. The overall Pearson r is 0.72. (a) Explain why computing a single r is potentially misleading. (b) What biological or methodological explanation might account for the clusters? (c) Propose an appropriate analysis strategy.
PROBLEM 5CRITICAL THINKING
An observational study reports r = 0.85 between daily fruit consumption (servings) and self-reported happiness score across 500 participants. A journalist writes: 'Eating more fruit causes greater happiness.' Provide a rigorous critique of this claim using at least three distinct methodological arguments grounded in scatterplot interpretation and statistical reasoning.

Lesson Summary

A scatterplot displays paired observations as points on a Cartesian plane and is the primary visual tool for exploring bivariate association in biostatistics. Every scatterplot should be evaluated along four dimensions: direction (positive, negative, or none), form (linear, nonlinear, or clustered), strength (how tightly points cluster around the trend), and the presence of outliers or influential observations. The Pearson correlation coefficient r quantifies linear association on a [−1, +1] scale, and represents the proportion of variance explained.

Critical pitfalls include relying on r for nonlinear relationships (where it can be near zero despite a strong curved trend), ignoring heteroscedasticity, and conflating association with causation. The foundational lesson—reinforced by Anscombe's Quartet and the history of exploratory data analysis—is that visualization must always precede and accompany numerical modeling. Scatterplot interpretation is not merely an introductory skill; it is a permanent diagnostic practice that validates every regression, every correlation test, and every causal inference in the biological and health sciences.

Varsity Tutors • Biostatistics • Interpreting Scatterplots — Interpret scatterplots and association patterns