Skip to content
Back to search results

Keyfitz Entropy

Introduced by
Nathan Keyfitz
Year
1977
Type
Elasticity of life expectancy to proportional mortality change / lifespan dispersion measure
Family
Life-table summary measures
Font size

What does it do?

Keyfitz's entropy, usually written H, is a dimensionless summary of a life table that measures how sensitive life expectancy is to a proportional change in mortality, and equivalently how unequal the distribution of ages at death is.

Introduced by Nathan Keyfitz, it is the elasticity of life expectancy at birth with respect to the force of mortality: an H near one means deaths are spread across all ages so that reducing mortality everywhere lengthens life proportionally, while an H near zero means deaths are concentrated near the maximum lifespan so further mortality reductions yield little gain. It bridges the demography of survival and the broader study of lifespan inequality.

The intuition behind the method

Imagine pressing down on the entire mortality schedule by the same proportion at every age. How much does life expectancy stretch in response? If deaths are scattered evenly across ages, lowering mortality everywhere adds life at many ages and life expectancy responds strongly — high entropy. If almost everyone already survives to a narrow band of old age, there is little early mortality left to remove, so the same proportional push barely lengthens life — low entropy. Keyfitz's H captures exactly this responsiveness. The same number also measures how spread out ages at death are: a rectangular survival curve (everyone dying at nearly the same old age) has low entropy, while an exponential survival curve (constant hazard, deaths at all ages) has entropy equal to one.

History

Nathan Keyfitz introduced the entropy of the life table in his 1977 textbook Applied Mathematical Demography, showing that it equals the elasticity of life expectancy to proportional mortality change. The measure was independently connected to ideas of demographic stability and evolutionary fitness by Lloyd Demetrius, and its links to lifespan inequality were developed by James Vaupel and colleagues, who related H to life disparity e†. As survival curves rectangularized across the 20th century, entropy became a standard diagnostic of how mortality reductions translate into longevity gains and of the convergence between life expectancy and lifespan equality.

When to use it

Use Keyfitz entropy when you want a scale-free measure of how concentrated or dispersed ages at death are, or how responsive life expectancy is to proportional mortality change — for example to characterize where a population sits on the path from high-mortality (high H) to low-mortality rectangularized (low H) regimes. It is useful for comparing lifespan inequality across populations on a common normalized scale. Assumptions: a complete and accurate life table, and that proportional mortality change is the relevant counterfactual for the elasticity interpretation. Do NOT treat H as a measure of the absolute spread of ages at death (use life disparity e† or the standard deviation for that), and do NOT compare H across populations without recognizing that it is a ratio, so equal H can arise from very different absolute lifespan distributions.

Limitations and considerations

  • Being a ratio (life disparity over life expectancy), it can mask large differences in the absolute spread of ages at death.
  • The elasticity interpretation assumes a uniform proportional change in mortality, which rarely matches real interventions.
  • It is sensitive to mortality at young ages, so populations differing mainly in old-age mortality may show similar H.
  • Estimates depend on accurate life-table construction, especially the open-ended oldest age interval.

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

Strengths

  • Dimensionless and bounded, so it places populations on a common normalized scale of lifespan dispersion.
  • Carries a precise interpretation as the elasticity of life expectancy to proportional mortality change.
  • Computable directly from standard life-table columns, requiring no extra data.
  • Links survival demography to the wider literature on lifespan inequality and the rectangularization of survival curves.

Applications

  • Characterizing where populations lie on the spectrum from high-mortality dispersed to low-mortality rectangularized survival.
  • Quantifying lifespan inequality on a normalized scale for cross-national comparison.
  • Assessing how much further mortality reduction can extend life expectancy in low-mortality populations.
  • Studying the historical rectangularization of survival curves and the compression of mortality.
  • Connecting demographic survival patterns to evolutionary and ecological theories of life history via Demetrius's work.

How it works

How it works

1

Start from the life-table survival function ℓ(x), the probability of surviving from birth to age x, whose integral over all ages equals life expectancy at birth e₀. This curve is the object whose shape — rectangular versus spread out — Keyfitz entropy summarizes.

Formula
\ell(x) = \frac{l_x}{l_0},\qquad e_0 = \int_0^{\infty} \ell(x)\, dx
2

Keyfitz entropy is defined as the negative integral of ℓ(x)·ln ℓ(x) divided by life expectancy e₀. This is a survival-weighted average that grows when the survival curve declines gradually across ages (deaths dispersed) and shrinks when it stays high then drops sharply (deaths concentrated), giving H its meaning as a measure of lifespan dispersion.

Formula
H = \frac{-\int_0^{\infty} \ell(x) \ln \ell(x)\, dx}{\int_0^{\infty} \ell(x)\, dx}
3

Equivalently, H equals minus the elasticity of life expectancy at birth with respect to a proportional change in the force of mortality: it is the percentage gain in e₀ from a one-percent uniform reduction in mortality. It can also be written as the average of remaining life expectancy weighted by deaths, divided by e₀, linking the dispersion and sensitivity views.

Formula
H = -\frac{d \ln e_0}{d \ln \mu},\qquad H = \frac{\int_0^{\infty} \ell(x) e(x)\, \mu(x)\, dx}{e_0}
4

In practice H is estimated from a discrete life table as the sum over ages of deaths times remaining life expectancy, normalized by life expectancy, which is closely approximated by e†/e₀ — the average years of life lost at death (life disparity) divided by life expectancy. This makes H computable directly from standard life-table columns.

Formula
\hat{H} = \frac{\sum_x d_x\, e_x}{\sum_x d_x\, x + e_0}\ \approx\ \frac{e^{\dagger}}{e_0}

Common pitfalls

  • Interpreting H as the absolute dispersion of ages at death rather than as a normalized, scale-free ratio.
  • Comparing H across populations without noting that identical values can come from different lifespan distributions.
  • Mishandling the open age interval in the life table, which distorts the entropy integral.
  • Confusing Keyfitz entropy with information-theoretic Shannon entropy of the death distribution, which is a related but distinct quantity.

Frequently asked questions

What does it mean when Keyfitz entropy equals one?

An entropy of one corresponds to an exponential survival curve with a constant force of mortality at all ages, where a proportional reduction in mortality produces an exactly proportional gain in life expectancy. Values below one indicate that deaths are increasingly concentrated near older ages, so uniform mortality reduction yields proportionally smaller longevity gains.

How is Keyfitz entropy related to life disparity e†?

Keyfitz entropy is closely approximated by the ratio e†/e₀, where e† is life disparity — the average remaining life expectancy lost at the ages where people die. Thus entropy is essentially life disparity normalized by life expectancy, which is why it serves both as a sensitivity elasticity and as a relative measure of lifespan inequality.

Does a falling entropy mean people are living longer?

Not directly; entropy measures the shape of the survival curve, not its level. Falling entropy means ages at death are becoming more concentrated (rectangularization), which historically accompanied rising life expectancy as early deaths were eliminated. But life expectancy and entropy are distinct: two populations can share a life expectancy yet differ in entropy.

Related methods

Sources

  1. Keyfitz, N. (1977). Applied Mathematical Demography. John Wiley & Sons, New York.

    ISBN: 9780471473503
    Copied.
  2. Demetrius, L. (1979). Relations between demographic parameters. Demography, 16(2), 329–338.

    Open DOI reference
    Copied.