MerQuR+Statistics Glossary

Statistics Glossary → Scatter plot

Scatter plotconcept

A graph that shows each observation of two numerical variables as a point whose coordinates are its two values.

In a scatter plot the horizontal axis shows one variable and the vertical axis the other; each point is one observation. The direction of the relationship (increasing, decreasing), its form (linear, curved), its strength (how tightly the points gather around a line), subgroups and outliers can all be seen at a glance. Points can be coloured or shaped by group, and a regression or smoothing curve can be overlaid.

A correlation coefficient is a single number, and very different patterns can produce the same r (Anscombe's quartet), so the scatter plot should be inspected before a correlation or regression is interpreted. With many observations, points overlap; transparency, smaller markers or hexagonal binning help. The plot shows association, not causation.

An everyday example

A café owner plots 60 days of data with the day's temperature on the horizontal axis and cold drinks sold on the vertical axis. The points rise to the right, but the rise levels off above 30 degrees, and two hot days with very few sales stand out: on those days the café closed early.

In MerQur

'Additional output: Scatter plot' in Bayesian Correlation; the cluster scatter under the 'Chart' option of DBSCAN / HDBSCAN; the fit plot (a scatter plot) under the 'Chart' option of Quantile Regression; the scatterplot matrix in 'Additional output: Assumption checks' of Correlation Analysis; and the canonical score plot (LD1 × LD2 with group centroids) in Discriminant Analysis.

Analyses where this concept appears

Correlation AnalysisDiscriminant Analysis (LDA / QDA)DBSCAN / HDBSCANQuantile RegressionBayesian Correlation

Related concepts

Pearson correlation coefficient