Scatter plots are the workhorse of exploratory data analysis, revealing relationships between two numerical variables in a way that summary statistics alone never can. A single glance at a well-constructed scatter plot can reveal correlations, clusters, outliers, and the overall shape of a relationship — information that would take a dozen computed metrics to approximate.
Correlation quantifies what scatter plots visualize. Pearson's correlation coefficient (r) measures the strength and direction of a linear relationship, ranging from -1 (perfect negative) to +1 (perfect positive). But correlation is not causation, and scatter plots are excellent at exposing the flaws in over-relying on correlation coefficients. Anscombe's quartet famously demonstrates four datasets with identical means, variances, and correlations that look completely different when plotted — one linear, one curved, one with an outlier, and one with no pattern except a single influential point.
Adding dimensions to scatter plots multiplies their analytical power. Color encodes a categorical third variable (species of flower, country of origin), size encodes a numerical fourth variable (population, revenue), and faceting splits the plot into small multiples by a fifth variable. Modern interactive scatter plots add tooltips that reveal the identity of individual points on hover, turning a static overview into an explorable data landscape.
Bubble charts extend scatter plots by mapping a third numerical variable to point size, but they come with a caution: humans are poor at comparing areas accurately. A bubble with twice the radius has four times the area, creating perceptual distortion. Use bubble charts for broad patterns rather than precise comparisons, and consider scaling bubble area — not radius — to the data value for honest visual encoding.