Life Expectancy Decomposition
- Introduced by
- Eduardo E. Arriaga; John H. Pollard
- Year
- 1984
- Type
- Demographic decomposition pipeline for differences in a summary measure
- Family
- Demography / life-table decomposition
What does it do?
Life-expectancy decomposition answers a question that a single number cannot: when life expectancy rises over time, or differs between two populations, exactly which ages and which causes of death are responsible?
The family of methods takes two life tables and splits their gap in e0 (or ex at any age) into additive contributions from mortality differences in each age interval, with the contributions summing exactly to the total gap. Eduardo Arriaga's 1984 stepwise discrete method became the field standard because it is exact, intuitive, and easy to extend to a cause-of-death breakdown, separating a 'direct' effect of changed survival within an interval from an 'indirect plus interaction' effect that the change propagates to later ages. John Pollard's continuous formulation expresses the same decomposition as an integral of age-specific mortality differences weighted by their leverage on life expectancy, providing the theoretical underpinning and a cross-check. This page treats the general decomposition pipeline; the dedicated Arriaga and Pollard pages cover each estimator in depth.
The intuition behind the method
Life expectancy at birth is an average of years lived, and any average difference can be traced back to where the underlying rates changed. Picture two survival curves, one for each population. The area between them is the difference in life expectancy. Decomposition slices that area horizontally by age: a fall in mortality between ages 60 and 65 widens the gap right there, but it also means more people survive to 65 and go on to enjoy whatever mortality conditions prevail afterward — a knock-on effect at older ages. Arriaga's method bookkeeps both pieces so that, summed over all ages, they reconstruct the total difference exactly. Add cause-of-death information and each age slice can be further divided among heart disease, cancer, injuries, and so on. The payoff is a story with arithmetic behind it: 'the two-year gain came mostly from reduced cardiovascular mortality in the 65-80 range,' rather than a bare headline number.
History
Demographers long sought to attribute changes in life expectancy to particular ages and causes. Early discrete attempts were superseded by Eduardo Arriaga's 1984 Demography paper, which gave an exact stepwise method separating direct from indirect-and-interaction effects and extending cleanly to causes of death; it remains the most widely used decomposition in population health. In parallel, John Pollard developed continuous decompositions, culminating in his 1988 Demography paper that unified earlier approaches and expressed the decomposition as an integral of weighted mortality differences, reconciling discrete and continuous traditions. Later work by Andreev, Shkolnikov, Vaupel and others generalized these ideas, but the Arriaga and Pollard formulations remain the canonical references and are cross-linked from the dedicated method pages.
When to use it
Use life-expectancy decomposition whenever you have two life tables and want to explain a difference or trend in life expectancy in terms of specific ages and, with cause data, specific causes of death. It is the standard tool for narrating mortality change over time, gaps between sexes or socioeconomic groups, cross-national differences, and the demographic impact of shocks such as epidemics or opioid crises. It requires comparable life tables on matched age intervals; the cause split additionally requires cause-of-death distributions for each population. It is descriptive, not causal — it attributes the difference arithmetically to where rates differ, not to why they differ. For decomposing healthy or disability-free life expectancy rather than total life expectancy, use the healthy-life-expectancy decomposition, which extends this logic to add a disability dimension.
Limitations and considerations
- Purely descriptive: it shows where mortality differs, not the behavioral, social, or biological causes of that difference.
- Results depend on the comparability of the two life tables; mismatched age groups, radices, or open intervals distort contributions.
- The cause split assumes cause-of-death coding is consistent and complete across both populations, which is often violated over time or across countries.
- Decomposition is not unique — direct/indirect ordering and discrete versus continuous choices can shift how effects are apportioned at the margins.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Exact and additive: the age (and cause) contributions sum precisely to the total life-expectancy difference, giving a complete accounting.
- Highly interpretable, turning a single summary gap into a clear age-by-cause story for policy and surveillance.
- Arriaga's discrete form is easy to compute from standard abridged life-table columns and extends naturally to a cause-of-death breakdown.
- Pollard's continuous formulation provides theoretical grounding and a cross-check, and clarifies that contributions reflect both rate change and demographic weight.
Applications
- Explaining trends in national life expectancy by identifying the ages and causes driving gains or stagnation.
- Quantifying the contribution of specific causes (cardiovascular disease, cancer, injuries, drug overdoses) to mortality change.
- Decomposing sex, racial, or socioeconomic gaps in life expectancy into age- and cause-specific components.
- Comparing life-expectancy differences across countries to locate where one population's mortality advantage arises.
- Assessing the demographic impact of mortality shocks such as pandemics by decomposing the drop and recovery in e0.
How it works
How it works
The pipeline starts from two comparable life tables — two time points for the same population, or two populations at one time — built on the same age intervals, typically with an abridged life table. The target is the difference in life expectancy at the chosen starting age, most often e0. The decomposition guarantees that the per-age contributions sum exactly to this total difference, so aligning the radix, age groups, and open interval of both tables is the essential precondition. Any discrepancy in table construction leaks into the contributions and breaks the additivity that makes the method interpretable.
Formula
\Delta e_0 = e_0^{(2)} - e_0^{(1)} = \sum_{x} {}_{n}\Delta_xArriaga's direct effect captures the years of life expectancy gained or lost because of changed mortality within an age interval itself, holding the number entering that interval fixed at the first population's level. It compares the average person-years lived per survivor in the interval between the two tables, scaled by the share of the synthetic cohort that reaches the interval. Intuitively, it asks: if survival within this band improved, how many extra years did people who were already there accumulate, before accounting for the fact that more of them now move on to later ages?
Formula
DE_x = \frac{l_x^{(1)}}{l_0^{(1)}}\left(\frac{{}_{n}L_x^{(2)}}{l_x^{(2)}} - \frac{{}_{n}L_x^{(1)}}{l_x^{(1)}}\right)The indirect-plus-interaction effect captures the downstream consequence: because mortality fell in this interval, more people survive into all subsequent intervals and live out the person-years available there. Arriaga combines this with the interaction term so that, added to the direct effect, the age-specific contribution is exact and the whole set sums to the total difference. This separation is what distinguishes a genuine within-age improvement from improvements that merely deliver more survivors to later ages, and it is the core of why the method is both additive and interpretable.
Formula
IE_x = \frac{T_{x+n}^{(2)}}{l_0^{(1)}}\left(\frac{l_x^{(1)}}{l_x^{(2)}} - \frac{l_{x+n}^{(1)}}{l_{x+n}^{(2)}}\right),\qquad {}_{n}\Delta_x = DE_x + IE_xOnce each age interval has a total contribution, it can be partitioned across causes of death. The standard step allocates the age contribution in proportion to how much each cause's mortality changed between the two populations, using the cause-specific shares of the all-cause rate in each table. The cause contributions within an age sum to the age total, and summing across ages gives each cause's overall contribution to the life-expectancy difference. This yields the familiar age-by-cause matrix that drives substantive interpretation in population health.
Formula
{}_{n}\Delta_x^{(i)} = {}_{n}\Delta_x\,\frac{{}_{n}R_x^{i(2)}\,{}_{n}m_x^{(2)} - {}_{n}R_x^{i(1)}\,{}_{n}m_x^{(1)}}{{}_{n}m_x^{(2)} - {}_{n}m_x^{(1)}}Pollard's continuous decomposition expresses the same quantity as an integral over age of the difference in the force of mortality, weighted by a function that measures how strongly mortality at each age levers life expectancy. It makes explicit that contributions depend both on how much mortality changed and on the demographic weight of that age, and it serves as a theoretical justification and numerical cross-check for the discrete Arriaga result. When age intervals are fine, the discrete contributions converge to the continuous integral, confirming that the two formulations describe one underlying decomposition.
Formula
\Delta e_0 = \int_0^{\infty} w(x)\,\bigl[\mu^{(1)}(x) - \mu^{(2)}(x)\bigr]\,dxCommon pitfalls
- Decomposing life tables built on different age groupings or open intervals, which breaks additivity and produces misleading contributions.
- Interpreting an age-cause contribution causally, as if reducing that cause would have produced the observed gain in some counterfactual sense.
- Ignoring changes in cause-of-death classification (e.g., ICD revisions) that create artificial cause contributions across time.
- Confusing decomposition of a difference between two populations with decomposition of a change over time when the two require different table pairings.
Frequently asked questions
How does this general page relate to the Arriaga and Pollard method pages?
This page describes the life-expectancy-decomposition family and the shared pipeline: take two comparable life tables, attribute the e0 difference to age intervals, and optionally split by cause. The dedicated Arriaga decomposition page details the discrete stepwise estimator with its direct and indirect-plus-interaction terms, and the Pollard decomposition page details the continuous integral formulation. Use this page to understand the goal and workflow, and the specific pages when you need the exact estimator you intend to implement or cite.
Is life-expectancy decomposition a causal method?
No. It is an exact arithmetic attribution of an observed difference in life expectancy to the ages and causes where mortality rates differ. It tells you where the gap comes from, not why mortality differs there. A cause contribution of two years to cardiovascular disease does not imply that an intervention on that cause would yield two years, because it ignores competing risks, behavioral responses, and confounding. Treat the output as a descriptive decomposition that informs, but does not replace, causal analysis.
What data do I need, and what can go wrong with the cause split?
You need two comparable abridged life tables built on matched age intervals, plus cause-specific death rates for each population if you want the cause breakdown. The most common problem is inconsistent cause-of-death coding across the two populations or time points — for example ICD revisions or shifts in certification practice — which can create spurious cause contributions even when total mortality change is real. Harmonizing cause categories and checking for coding breaks before decomposing is essential, as both Arriaga and downstream users emphasize.
Related methods
Sources
Arriaga, E. E. (1984). Measuring and explaining the change in life expectancies. Demography, 21(1), 83-96.
Open DOI referenceCopied.Pollard, J. H. (1988). On the decomposition of changes in expectation of life and differentials in life expectancy. Demography, 25(2), 265-276.
Open DOI referenceCopied.