Statistics Glossary → Distance measure (Euclidean, Manhattan, Gower, great-circle)
Distance measure (Euclidean, Manhattan, Gower, great-circle)concept
The definition that turns how dissimilar two observations are into a number, used in clustering, scaling and spatial analyses.
Euclidean distance is the square root of the sum of squared differences (√Σ(xᵢ − yᵢ)²), the straight-line distance. Manhattan distance is the sum of absolute differences (Σ|xᵢ − yᵢ|) and gives a single large difference less weight than Euclidean does; Chebyshev is the largest single difference. Gower distance handles mixed data by scaling each numerical, ordinal or categorical variable to the 0-1 range and averaging. Great-circle (haversine) distance is the distance over the surface of a sphere between latitude-longitude coordinates.
The chosen measure directly shapes the result: the same data can give different clusters with different distances. Euclidean and Manhattan are sensitive to the scale of the variables; if income (in thousands) and age (in tens) are used together without standardizing, income dominates the distance. Applying plain Euclidean distance to geographic coordinates in degrees distorts distances, especially at high latitudes and over large areas.
An everyday example
Two cafés in a city are 1 km apart as the crow flies (Euclidean), but because the streets form a grid the walk is 1.4 km (Manhattan). For the distance between two cities, rather than a straight line through latitude-longitude degrees, the great-circle distance that allows for the Earth's curvature is used.
In MerQur
'Distance measure' in Hierarchical Clustering (Euclidean, Manhattan, Chebyshev, Gower and others); 'Distance metric' in DBSCAN / HDBSCAN (including Haversine); the distance options in Multidimensional Scaling, t-SNE, UMAP and k-Nearest Neighbours; the 'Distance metric' (including Mahalanobis) of the kNN method in Discriminant Analysis; and Euclidean or great-circle distance in SAR, the Spatial Error Model and GWR.
Analyses where this concept appears
Discriminant Analysis (LDA / QDA)K-Means ClusteringHierarchical ClusteringDBSCAN / HDBSCANt-SNE (t-Distributed Stochastic Neighbor Embedding)Multidimensional Scaling (MDS)UMAP (Uniform Manifold Approximation and Projection)SAR (Spatial Autoregressive Lag)Spatial Error Model (SEM)GWR (Geographically Weighted Regression)k-Nearest Neighbors (KNN)Moran's I (Spatial Autocorrelation)DBSCAN Spatial ClusteringVariogram + Kriging (Spatial Interpolation)Point Pattern Analysis (Ripley K/L, G/F/J, CSR)
Related concepts
z-score (standardization)Mahalanobis distanceLinkage method (Ward, complete, average, single)