Spatial Dissimilarity Index
- Introduced by
- Richard Morrill & David Wong
- Year
- 1993
- Type
- Boundary-aware index of residential segregation between two groups
- Family
- Segregation and concentration measures
What does it do?
The spatial dissimilarity index is a boundary-aware measure of residential segregation that corrects the classic index of dissimilarity for the fact that areal units are not isolated boxes but neighbours that share borders.
Developed by Richard Morrill in 1991 and refined by David Wong in 1993, it discounts the aspatial index by the degree to which adjacent units differ in group composition, so that two groups clustered into separate but neighbouring areas are recorded as less segregated than two groups locked into a checkerboard. It directly addresses the long-standing checkerboard problem that the aspatial Duncan index cannot see.
The intuition behind the method
The classic dissimilarity index counts how unevenly two groups are spread across units but is blind to where those units sit. Imagine a chessboard where black and white squares perfectly alternate, versus one where all black squares are on the left half and all white on the right: the aspatial index scores both as completely segregated, yet anyone can see the second is far more segregated than the first, where every household abuts the other group across a border. The spatial index fixes this by giving back 'credit' wherever adjacent units resemble each other — the alternating board, full of mixed borders, has its score pulled down, while the split board, with one long internal frontier, keeps a high score.
History
The index of dissimilarity, formalized by Otis and Beverly Duncan in 1955, dominated segregation measurement for decades but was repeatedly criticized for being aspatial — its checkerboard problem meant identical scores for utterly different geographies. Richard Morrill addressed this in 1991 in Geography Research Forum by subtracting a contiguity-based interaction term, defining the boundary-adjusted index D(adj). David Wong extended the idea in his 1993 Urban Studies paper 'Spatial indices of segregation', incorporating shared boundary lengths, unit perimeters, and shape, and situating the family within a broader treatment of spatial segregation measures. Their work, together with later contributions by Reardon and O'Sullivan on spatial segregation profiles, established the boundary-aware tradition now standard in segregation research.
When to use it
Use the spatial dissimilarity index when you measure segregation between two groups across areal units and care about the spatial arrangement of those units, not merely their compositions — for example, distinguishing a checkerboard of mixed neighbours from large, clustered ethnic enclaves that the aspatial index would score identically. It is appropriate when you have group counts per unit and a contiguity or shared-boundary structure, and when you want a measure robust to the checkerboard problem. It is less suitable when units are non-contiguous or boundary data are unavailable, when interest lies in exposure or isolation rather than evenness, or when a multi-group or entropy-based measure better fits the question.
Limitations and considerations
- Requires a contiguity or shared-boundary structure, which is unavailable or ambiguous for non-contiguous or point data.
- Remains a two-group, evenness-only measure; it does not capture exposure, isolation, or multi-group segregation.
- Like all areal indices it is sensitive to the modifiable areal unit problem and to how units are drawn.
- The adjustment can only lower the index, so it cannot represent forms of segregation that boundaries fail to smooth.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Resolves the checkerboard problem that makes the aspatial dissimilarity index blind to spatial arrangement.
- Built directly on the familiar, interpretable Duncan index, so it remains easy to communicate.
- Wong's boundary-length and perimeter weighting respects the real geometry of units, not just adjacency.
- Exposes the scale-dependence of segregation, distinguishing clustered enclaves from fine-grained mixing.
Applications
- Measuring residential segregation between ethnic or racial groups while respecting neighbourhood adjacency.
- Distinguishing clustered enclaves from fine-grained mixing that the aspatial index scores identically.
- Comparing segregation across cities or census years with a spatially honest, boundary-aware metric.
- Assessing how measured segregation changes with the spatial scale and zoning of areal units.
- Studying socioeconomic or linguistic separation where the geography of contact, not just composition, matters.
How it works
How it works
The baseline is Duncan and Duncan's classic index of dissimilarity D, half the summed absolute differences between each unit's share of group X and its share of group Y. It ranges from 0 (identical group proportions everywhere) to 1 (complete separation) and is interpreted as the fraction of one group that would have to relocate for the two distributions to match. Crucially, D depends only on the composition of units, not on where they lie relative to one another.
Formula
D = \frac{1}{2}\sum_i \left|\frac{x_i}{X} - \frac{y_i}{Y}\right|Morrill's adjustment subtracts from D a term built from a contiguity matrix c, whose entries flag which units share a boundary. For every adjacent pair the absolute difference in group proportions |z_i − z_j| is summed and normalized by the number of adjacencies, then subtracted from D. Where neighbouring units are similar, this correction is small and the index changes little; where neighbours differ sharply, the term is large and pulls the index down only when those differences are smoothed across borders, rewarding spatial clustering.
Formula
D(adj) = D - \frac{\sum_i \sum_j c_{ij}\,|z_i - z_j|}{\sum_i \sum_j c_{ij}}Wong refined the adjacency adjustment to use the geometry of the shared borders rather than mere adjacency. The interaction between two units is weighted by the length of their shared boundary b_ij relative to a unit's perimeter P_i (and, in fuller versions, by unit shape and area), so that units sharing a long common edge interact more than those touching at a point. This makes the index sensitive to the size, shape, and contact of units, not just to a binary neighbour relation.
Formula
D(w) = D - \sum_i \sum_j \frac{b_{ij}}{P_i}\,|z_i - z_j|Because the adjustments only ever subtract interaction across borders, the spatial index is at most equal to and usually below the aspatial D, and it falls furthest when similar groups occupy adjacent units — that is, when segregation is spatially clustered into compact zones. Comparing the spatial and aspatial values, and recomputing across spatial scales, reveals how much of the measured unevenness is genuinely spatial and exposes the scale-dependence that the aspatial index hides.
Common pitfalls
- Treating the spatial and aspatial indices as interchangeable when they answer different questions about arrangement.
- Using a binary adjacency where boundary lengths differ greatly, ignoring that some neighbours share far more border than others.
- Comparing spatial indices computed under different contiguity definitions (rook vs. queen) as if on the same scale.
- Forgetting that the measure still inherits modifiable-areal-unit and edge effects from the units it is built on.
Frequently asked questions
How does the spatial dissimilarity index differ from the classic index of dissimilarity?
The classic index of dissimilarity measures only how unevenly two groups are distributed across units and ignores where those units sit, so it scores a checkerboard of alternating groups the same as two solid, separated blocks. The spatial index starts from that same value but subtracts an interaction term across shared boundaries, lowering the score wherever neighbouring units resemble each other. As a result it falls below the aspatial value precisely when segregation is spatially clustered, capturing arrangement as well as composition.
What is the checkerboard problem?
The checkerboard problem is the failure of the aspatial dissimilarity index to distinguish spatial patterns that have the same unit compositions but very different geographies. A perfectly alternating checkerboard and a board split into two solid halves both yield maximal aspatial dissimilarity, even though the halves are far more segregated in any meaningful sense. The spatial index solves this by accounting for whether adjacent units are similar, rewarding the mixed borders of the checkerboard with a lower score.
What did Wong add to Morrill's adjustment?
Morrill's 1991 adjustment used a simple binary contiguity matrix: units either share a boundary or they do not. Wong's 1993 refinement made the correction geometric, weighting each pair's interaction by the length of their shared boundary relative to the unit's perimeter, and in fuller versions by unit area and shape. This means two units with a long common edge interact more strongly than two that barely touch, so the index reflects the real intensity of contact between units rather than treating all neighbours equally.
Related methods
Sources
Wong, D. W. S. (1993). Spatial indices of segregation. Urban Studies, 30(3), 559–572.
Open DOI referenceCopied.