Skip to content
Back to search results

Kitagawa Decomposition

Introduced by
Evelyn M. Kitagawa
Year
1955
Type
Arithmetic decomposition of a difference between two summary rates
Family
Standardization & decomposition
Font size

What does it do?

Kitagawa decomposition is a demographic technique that splits the difference between two summary rates — such as two crude death rates, birth rates, or prevalence figures — into the part attributable to differences in the underlying group-specific rates and the part attributable to differences in population composition.

Introduced by Evelyn Kitagawa in 1955, it answers whether a gap between two populations reflects genuinely different risks or merely a different age (or other) structure.

The intuition behind the method

A crude rate is a weighted average of group-specific rates, where the weights are the share of the population in each group. If country A has a higher crude death rate than country B, it could be because people of every age die at higher rates in A (a rate effect), or simply because A has an older population (a composition effect), or some mix of both. Kitagawa decomposition formalizes that intuition: it holds composition fixed to isolate the rate effect, holds rates fixed to isolate the composition effect, and uses the average of the two populations as the neutral reference so the two pieces add up exactly to the observed gap.

History

Evelyn M. Kitagawa, a demographer at the University of Chicago, published 'Components of a difference between two rates' in the Journal of the American Statistical Association in 1955. The paper gave standardization — long used informally by demographers and actuaries — a rigorous, exactly additive algebraic foundation. It became one of the most cited methodological papers in demography and the direct ancestor of the broader family of decomposition techniques, including Arriaga's life-expectancy decomposition and Das Gupta's multi-factor generalization.

When to use it

Use Kitagawa decomposition when you want to explain why two summary rates differ — across countries, time points, or subgroups — and need to separate the role of underlying risk from the role of population structure. It is the standard first tool when a crude rate comparison is potentially confounded by composition (most often age). It applies to any crude rate that can be written as a composition-weighted average of group-specific rates. It is restricted to comparing two populations across a single compositional dimension; for more than two populations or several interacting factors, the Das Gupta generalization is required.

Limitations and considerations

  • Handles a comparison of only two populations at a time and a single compositional factor; multi-population or multi-factor problems need Das Gupta's extension.
  • Assumes the crude rate is a simple composition-weighted average of group-specific rates; rates that are nonlinear functions of components do not decompose this cleanly.
  • The choice of compositional dimension (e.g., age vs. education) is the analyst's; different groupings yield different and non-comparable decompositions.
  • Provides a descriptive accounting identity, not a causal explanation — the components describe arithmetic, not mechanism.

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

Strengths

  • Exact additive identity: the two components sum to the observed difference with no leftover residual, making results easy to communicate.
  • Conceptually transparent — directly answers 'is the gap about risk or about structure?' in the units of the original rate.
  • Requires only group-specific rates and population shares, data that are routinely available from vital statistics and censuses.
  • Forms the conceptual basis for standardization and for more general multi-factor decompositions.

Applications

  • Explaining cross-country or cross-region differences in crude death rates as age-structure versus mortality-risk effects.
  • Decomposing changes in fertility into shifts in the age structure of women versus changes in age-specific fertility rates.
  • Separating compositional change (e.g., an ageing workforce) from behavioral change when comparing labor-force participation or disease prevalence over time.
  • Public-health analyses that ask whether a rising mortality rate reflects worsening health or simply a growing share of older residents.
  • Teaching the logic of standardization, for which the Kitagawa components are the formal justification.

How it works

How it works

1

Each crude rate R is expressed as the sum over groups i of the composition share c (the proportion of the population in group i) times the group-specific rate r. The quantity of interest is the gap R₂ − R₁ between population 2 and population 1, which mixes differences in both the c's and the r's.

Formula
R_2 - R_1 = \sum_i c_{2i} r_{2i} - \sum_i c_{1i} r_{1i}
2

The rate (or 'effect') component holds composition constant at the average of the two populations, (c₁ᵢ + c₂ᵢ)/2, and sums the differences in group-specific rates (r₂ᵢ − r₁ᵢ). It measures how much of the gap would remain if both populations had identical structure and differed only in their group-level risks.

Formula
\text{Rate} = \sum_i \frac{c_{1i}+c_{2i}}{2}\,(r_{2i}-r_{1i})
3

The composition (or 'structure') component holds rates constant at the average (r₁ᵢ + r₂ᵢ)/2 and sums the differences in composition shares (c₂ᵢ − c₁ᵢ). It measures how much of the gap is explained purely by differing population structure, such as age distribution.

Formula
\text{Composition} = \sum_i \frac{r_{1i}+r_{2i}}{2}\,(c_{2i}-c_{1i})
4

Using the symmetric average weights guarantees that the rate component and the composition component sum exactly to the observed crude-rate difference, with no residual interaction term. This exact additivity is what makes the decomposition interpretable and reproducible.

Formula
R_2 - R_1 = \text{Rate} + \text{Composition}

Common pitfalls

  • Interpreting the composition effect as causal: a large structure component shows arithmetic attribution, not that changing structure would change risk.
  • Decomposing rates defined on incompatible group boundaries between the two populations, which makes the c and r terms non-comparable.
  • Forgetting that switching which population is labelled 1 versus 2 flips the signs of both components; report the direction explicitly.
  • Using fine age groups with sparse counts, which makes group-specific rates noisy and inflates the apparent rate component.

Frequently asked questions

How is Kitagawa decomposition related to standardization?

Direct standardization removes composition differences by applying both populations' rates to a single standard structure; the standardized rate difference is essentially the Kitagawa rate component. Kitagawa decomposition makes the relationship explicit by also quantifying the composition component, so the two effects sum to the raw gap. Standardization reports one number; Kitagawa reports the full split.

Why are average weights used instead of one population's weights?

If you weight the rate component by population 1's composition and the composition component by population 1's rates, the two pieces leave an interaction residual. Using the symmetric average of the two populations for the held-constant factor makes the decomposition exactly additive with no residual, which is why Kitagawa's symmetric formulation became standard.

Can it decompose differences across more than two populations?

Not directly. The classic method is defined for two populations and one compositional factor. Prithwis Das Gupta (1993) generalized the approach to any number of populations and several interacting factors while preserving internal consistency, and that extension should be used for multi-population or multi-factor problems.

Related methods

Sources

  1. Kitagawa, E. M. (1955). Components of a difference between two rates. Journal of the American Statistical Association, 50(272), 1168–1194.

    Open DOI reference
    Copied.
  2. Preston, S. H., Heuveline, P., & Guillot, M. (2001). Demography: Measuring and Modeling Population Processes. Blackwell.

    ISBN: 9781557864512
    Copied.