Isolation Index
- Introduced by
- Wendell Bell (formalization of P* indices)
- Year
- 1954
- Type
- Exposure-dimension segregation index
- Family
- Segregation measurement
What does it do?
The isolation index measures the exposure dimension of segregation: the extent to which members of a minority group are exposed only to one another rather than to members of other groups.
It answers the question 'what is the own-group share of the typical neighbor (or classmate, or coworker) that a member of the focal group encounters?' Unlike evenness measures, it depends on the relative size of the group as well as its spatial distribution.
The intuition behind the method
Pick a random member of the minority group and look at the composition of their unit — their neighborhood or school. The isolation index averages, over all such members, the proportion of their unit that is also minority. If minority members live almost entirely among other minority members, the index is near 1 (high isolation); if they are spread among the majority, it is near the group's overall share (low isolation). The complementary interaction index instead measures the average exposure to the other group. Crucially, even a perfectly even distribution leaves a small group with low isolation simply because it is small, so isolation reflects both distribution and group size.
History
The exposure-based P* indices were formalized by Wendell Bell in his 1954 Social Forces paper on a probability model for ecological segregation, building on earlier work by Shevky and Bell on the composition of neighborhoods. Stanley Lieberson revived and extended the P* framework in the late 1970s and 1980s, clarifying the asymmetry between isolation and interaction. Massey and Denton's 1988 dimensions framework classified isolation as the canonical measure of the exposure dimension, distinct from the evenness captured by the dissimilarity index.
When to use it
Use the isolation (and interaction) index when the substantive concern is exposure — the actual chance of within-group versus cross-group contact in neighborhoods, schools, or workplaces — rather than mere evenness. It is the standard measure of the exposure dimension and is appropriate when group size itself matters for the experience of segregation. It is less appropriate when you want a size-independent measure of evenness (use dissimilarity) or a multigroup decomposition (use the entropy index), and the raw index can be misleading when comparing groups of very different sizes unless the normalized form is used. Like other aspatial indices, it ignores the arrangement of units in space.
Limitations and considerations
- Depends on the group's overall size, so a small group can show low isolation even when highly clustered, unless normalized.
- Aspatial: ignores where units lie relative to one another.
- The basic form handles exposure between two groups; multigroup exposure needs generalized P* matrices.
- Sensitive to unit definition and the modifiable areal unit problem like other segregation indices.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Directly captures lived exposure — the group composition members actually encounter — which evenness measures miss.
- Interaction and isolation forms together describe both within- and cross-group contact, summing to one in the two-group case.
- Simple to compute from standard cross-tabulated unit counts.
- The standard measure of the exposure dimension, complementing dissimilarity's evenness.
Applications
- Measuring the own-group exposure of minority residents across neighborhoods.
- Assessing school segregation in terms of the racial composition the average student experiences.
- Studying workplace exposure across occupations or firms.
- Tracking how minority population growth and segregation jointly change exposure over time.
- Comparing exposure-based segregation across metropolitan areas.
How it works
How it works
The isolation index is the population-weighted average, over members of the focal group X, of the proportion of their unit that is also group X. Each unit contributes its share of the total X population times the within-unit proportion of X. The result is the own-group exposure experienced by the average group-X member.
Formula
{}_{x}P^{*}_{x} = \sum_{i=1}^{n} \left(\frac{x_i}{X}\right)\left(\frac{x_i}{t_i}\right)The interaction index is built the same way but uses the within-unit proportion of the other group Y, giving the average exposure of a group-X member to group Y. It measures potential contact across groups rather than insulation within a group.
Formula
{}_{x}P^{*}_{y} = \sum_{i=1}^{n} \left(\frac{x_i}{X}\right)\left(\frac{y_i}{t_i}\right)In the two-group case the isolation and interaction indices sum to one: a group-X member's unit is either X or Y, so their own-group exposure plus their other-group exposure exhausts the unit. High isolation therefore directly implies low cross-group interaction.
Formula
{}_{x}P^{*}_{x} + {}_{x}P^{*}_{y} = 1\quad (\text{two-group case})Because raw isolation depends on the group's overall size, a normalized version (the eta-squared or correlation-ratio form) rescales it by subtracting the group's citywide share and dividing by its complement. This removes the mechanical floor set by group size and isolates the distributional component, making isolation comparable across areas where the group has different sizes.
Formula
\eta^2 = \frac{{}_{x}P^{*}_{x} - X/T}{1 - X/T}\quad(\text{normalized isolation})Common pitfalls
- Comparing raw isolation across groups or cities of different sizes without using the size-adjusted (eta-squared) version.
- Interpreting low isolation of a small minority as integration when its small size mechanically lowers the index.
- Confusing isolation (own-group exposure) with dissimilarity (evenness); the two can diverge sharply.
- Ignoring that isolation rises with both segregation and minority population growth, conflating the two trends.
Frequently asked questions
How does the isolation index differ from the index of dissimilarity?
Dissimilarity measures evenness — how unequally two groups are spread across units — and is independent of group size. Isolation measures exposure — the own-group composition a typical member encounters — and depends on group size as well as distribution. A small minority can be highly uneven (high dissimilarity) yet have low isolation simply because there are too few of its members to fill any unit. The two capture distinct dimensions of segregation and should be reported together.
What is the interaction index?
The interaction index is the exposure counterpart that measures the average proportion of a focal group member's unit made up of another group — that is, potential cross-group contact. In the two-group case isolation and interaction sum to one. Interaction indices are useful when the question is about intergroup contact rather than within-group insulation.
Why does isolation depend on group size?
Because the index averages the actual own-group share in members' units, a group that is only a few percent of the total population cannot, even under perfect evenness, make up a large share of any unit. Its isolation is therefore mechanically low. The normalized eta-squared form removes this size floor so that the purely distributional component can be compared across groups and places.
Related methods
Sources
Bell, W. (1954). A probability model for the measurement of ecological segregation. Social Forces, 32(4), 357–364.
Open DOI referenceCopied.Massey, D. S., & Denton, N. A. (1988). The dimensions of residential segregation. Social Forces, 67(2), 281–315.
Open DOI referenceCopied.