Skip to content
Back to search results

Theil Segregation Index

Introduced by
Henri Theil & Anthony Finizza
Year
1971
Type
Entropy-based multigroup segregation index
Family
Segregation measurement
Font size

What does it do?

Theil's information index, denoted H, is an entropy-based measure of segregation that, unlike the two-group dissimilarity index, handles any number of groups at once.

It compares the diversity (entropy) found within each unit to the diversity of the whole population: segregation is high when units are internally homogeneous even though the overall population is diverse. Its defining virtue is exact decomposability across nested levels and across groups.

The intuition behind the method

Entropy measures how mixed a population is — it is highest when all groups are equally represented and zero when everyone is the same group. If the city as a whole is diverse but each neighborhood is internally uniform, then knowing someone's neighborhood tells you a lot about their group: that information gain is segregation. Theil's H is the population-weighted average of how much each unit falls short of the citywide diversity, expressed as a fraction of that diversity. Because entropy decomposes cleanly, H can be split into how much segregation occurs between districts versus within them — a property no other common index shares.

History

Henri Theil, an econometrician who applied information theory to economics, and Anthony Finizza introduced the entropy-based segregation index in their 1971 Journal of Mathematical Sociology note on measuring racial integration of schools, drawing on Theil's broader information-theoretic work on inequality. The index gained prominence as multigroup segregation became central with growing ethnic diversity. Sean Reardon and Glenn Firebaugh's 2002 Sociological Methodology paper systematically evaluated multigroup measures and established H's decomposition properties as decisive advantages, and the U.S. Census Bureau adopted it as a standard multigroup measure.

When to use it

Use Theil's H when segregation involves more than two groups, when you need to decompose segregation across nested geographic levels (region, district, neighborhood) or organizational levels, or when you want each group's contribution to total segregation. It is the preferred multigroup evenness measure and the natural choice for analyzing how segregation is distributed across scales. It is less intuitive than the dissimilarity index's relocation interpretation, does not capture exposure or spatial clustering, and like other aspatial indices ignores the arrangement of units. Small units and sparse cells bias it upward unless a bias correction is applied.

Limitations and considerations

  • Lacks the simple 'share that must relocate' interpretation of the dissimilarity index.
  • Measures evenness only — not exposure, clustering, or concentration.
  • Aspatial: ignores the relative location of units.
  • Biased upward with small units or sparse counts unless an explicit bias correction is used.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Handles any number of groups simultaneously, unlike the two-group dissimilarity index.
  • Exactly decomposable across nested spatial or organizational levels (between vs. within).
  • Decomposable across groups, isolating each group's contribution to total segregation.
  • Grounded in information theory with a clear interpretation as the diversity explained by between-unit differences.

Applications

  • Measuring multigroup racial and ethnic residential segregation in diverse metropolitan areas.
  • Decomposing school segregation into between-district and within-district components.
  • Analyzing occupational segregation across many groups simultaneously.
  • Studying how segregation is distributed across spatial scales (metro, district, tract).
  • Quantifying each group's contribution to total societal segregation.

How it works

How it works

1

First compute the entropy E of the entire population from the overall group shares π_g. Entropy is maximized when all G groups are equally common and is zero when the population is a single group; it quantifies the total diversity to be 'explained' by the spatial distribution.

Formula
E = \sum_{g=1}^{G} \pi_g \ln\!\frac{1}{\pi_g}
2

Compute the entropy E_i of each unit from its internal group shares π_ig. A perfectly mixed unit has entropy equal to the overall E; a unit containing only one group has zero entropy. The gap E − E_i measures how much less diverse the unit is than the population at large.

Formula
E_i = \sum_{g=1}^{G} \pi_{ig} \ln\!\frac{1}{\pi_{ig}}
3

Theil's H is the population-weighted average of each unit's relative entropy shortfall (E − E_i)/E. It ranges from 0, when every unit mirrors the overall composition, to 1, when every unit is internally homogeneous. H is exactly the proportion of total population diversity that is accounted for by between-unit differences.

Formula
H = \sum_{i=1}^{n} \frac{t_i}{T}\,\frac{E - E_i}{E}
4

Because entropy is additively decomposable, H itself decomposes: total segregation equals segregation between higher-level units (e.g., districts) plus a population-weighted sum of segregation within those units (e.g., neighborhoods within districts). It likewise decomposes across groups, allowing the contribution of each group to overall multigroup segregation to be isolated.

Formula
H = H_{\text{between}} + \sum_k w_k\, H_{\text{within},k}

Common pitfalls

  • Comparing H across studies that use different logarithm bases or normalizations without rescaling to a common convention.
  • Ignoring small-unit upward bias when comparing areas with different unit sizes; use bias-corrected estimators.
  • Interpreting H as exposure; a highly diverse area can still have low cross-group contact for a small group.
  • Decomposing across an arbitrary or non-nested unit hierarchy, which breaks the clean between/within accounting.

Frequently asked questions

Why use Theil's H instead of the dissimilarity index?

The dissimilarity index compares only two groups and cannot decompose segregation across levels. Theil's H handles any number of groups at once and, because entropy is additively decomposable, splits total segregation into between- and within-level components and into per-group contributions. When the setting is multigroup or multilevel, H provides analyses the dissimilarity index simply cannot.

What does an H value mean substantively?

H is the proportion of the population's total diversity (entropy) that is attributable to differences between units rather than within them. H = 0 means every unit has the same composition as the whole population (no segregation); H = 1 means every unit is internally homogeneous (complete segregation). It is a relative-diversity measure rather than a relocation count.

How does the decomposition property help?

Because H decomposes additively, you can ask how much of total segregation occurs between large units (e.g., school districts) versus within them (e.g., schools within districts), and how much each group contributes. This pinpoints the geographic scale and the groups driving segregation, guiding policy and theory in ways a single undecomposable number cannot.

Related methods

Sources

  1. Theil, H., & Finizza, A. J. (1971). A note on the measurement of racial integration of schools by means of informational concepts. Journal of Mathematical Sociology, 1(2), 187–193.

    Open DOI reference
    Copied.
  2. Reardon, S. F., & Firebaugh, G. (2002). Measures of multigroup segregation. Sociological Methodology, 32(1), 33–67.

    Open DOI reference
    Copied.