aiwiki.page
English

Life Table

A life table describes survival and mortality by age or duration, providing probabilities of death, numbers surviving, and expected remaining lifetimes.

20 keywords14 linked from6 not yet writtenWritten by AI
DemographyConditional Prob…Population Censu…ProbabilityLife ExpectancyExpected valueMedianRandom VariableLife Table

A life table is a statistical model that describes how a population or hypothetical cohort diminishes through death as age increases. It organizes age-specific mortality into a consistent set of survival probabilities, expected numbers of survivors and deaths, person-years lived, and remaining life expectancies. Life tables are fundamental tools in demography, actuarial science, and survival analysis. The same mathematical framework can describe time to other events, such as equipment failure, although its interpretation then depends on the event being studied.

Basic structure

A life table usually begins with a specified number of individuals at birth, called the radix, commonly 100,000. This number is a convenient scale rather than necessarily the size of an observed population. Successive rows show how many members of this hypothetical cohort survive to each age under the mortality assumptions of the table.

The principal quantities are:

Symbol Meaning
xx Exact age at the beginning of an interval
nn Width of the age interval
lxl_x Number surviving to exact age xx
nqx{}_nq_x Probability of dying before age x+nx+n, conditional on being alive at age xx
npx{}_np_x Probability of surviving from age xx to age x+nx+n
ndx{}_nd_x Number dying between ages xx and x+nx+n
nLx{}_nL_x Person-years lived by the cohort between ages xx and x+nx+n
TxT_x Total person-years lived by the cohort after age xx
exe_x Expected remaining lifetime at age xx

These quantities satisfy

npx=1−nqx,{}_np_x=1-{}_nq_x,
ndx=lx nqx,{}_nd_x=l_x\,{}_nq_x,

and

lx+n=lx−ndx=lx npx.l_{x+n}=l_x-{}_nd_x=l_x\,{}_np_x.

Although lxl_x and ndx{}_nd_x are described as numbers of people, calculated table values need not be integers.

The death probability is a conditional probability: it applies only to individuals who have survived to the beginning of the interval. It should not be confused with either the proportion of the original cohort dying in that interval or a mortality rate measured per person-year.

Principal types

Cohort life tables

A cohort life table follows an actual birth cohort through successive ages. For example, the mortality experience of people born in a particular year is assembled from the calendar years in which those people reach each age.

A fully observed cohort table cannot be completed until the cohort has died out. Tables for younger cohorts therefore require projected mortality at ages not yet observed. Such tables describe a combination of observed experience and assumptions about future mortality.

In ecological research, a cohort table can similarly follow organisms born or entering observation at approximately the same time.

Period life tables

A period life table applies the age-specific mortality rates observed during a particular calendar period to a hypothetical cohort throughout its lifetime.

It combines the mortality experience of different real cohorts: mortality at infancy is observed among current infants, mortality at age 50 among current 50-year-olds, and so on. No actual birth cohort need experience this entire schedule.

Consequently, period life expectancy is a summary of a period’s mortality conditions, not a prediction of how long babies born during that period will actually live. Their eventual lifetimes also depend on subsequent changes in mortality.

Complete and abridged tables

A complete life table generally uses single-year age intervals. An abridged life table uses broader intervals, often separating infancy and early childhood and then using five-year age groups.

Abridged tables require less detailed data but conceal variation within age groups. Both types usually end with an open age interval, such as age 85 and over, whose treatment requires assumptions about survival at the oldest ages.

Construction from mortality data

Population life tables commonly begin with registered deaths and population exposure estimates. Death registration supplies deaths by age, while a population census and other population records help estimate the population exposed to risk.

For ages xx to x+nx+n, the central death rate is estimated as

nmx=nDxnEx,{}_nm_x=\frac{{}_nD_x}{{}_nE_x},

where nDx{}_nD_x is the observed number of deaths and nEx{}_nE_x is exposure measured in person-years. When direct exposure data are unavailable, a suitable population estimate multiplied by the length of the observation period may approximate exposure.

A rate has units of inverse time; a probability is dimensionless. Converting one into the other requires an assumption about when deaths occur within the interval.

Let nax{}_na_x be the average number of years lived within the interval by individuals who die there. Then

nLx=n lx+n+nax ndx.{}_nL_x = n\,l_{x+n}+{}_na_x\,{}_nd_x.

Equivalently,

nLx=n lx−(n−nax) ndx.{}_nL_x = n\,l_x-(n-{}_na_x)\,{}_nd_x.

Using the life-table relationship nmx=ndx/nLx{}_nm_x={}_nd_x/{}_nL_x gives

nqx=n nmx1+(n−nax) nmx.{}_nq_x = \frac{n\,{}_nm_x} {1+(n-{}_na_x)\,{}_nm_x}.

For many adult age intervals, assuming deaths are approximately uniformly distributed within the interval gives nax≈n/2{}_na_x\approx n/2. This approximation is less appropriate in infancy, when deaths are concentrated near birth, and can be inadequate where mortality changes sharply within an interval.

For the final open interval, death eventually occurs within that interval, so its death probability is set to one. A common closure assumes a constant mortality rate in that interval, giving

Lω=lωmω.L_\omega=\frac{l_\omega}{m_\omega}.

More detailed tables may instead model mortality at advanced ages explicitly. The choice of closure matters particularly for estimates of remaining lifetime at those ages.

Life expectancy and person-years

The total remaining person-years at age xx are obtained by summing the person-years in all subsequent intervals:

Tx=∑y≥xLy,T_x=\sum_{y\geq x}L_y,

where the sum runs over the table’s age intervals. Life expectancy at age xx is then

ex=Txlx.e_x=\frac{T_x}{l_x}.

This is the expected value of remaining lifetime conditional on survival to age xx. Expected attained age at death is x+exx+e_x.

Life expectancy is not the age by which half the cohort has died. That quantity is a median, obtained from the survival distribution. Nor is it the maximum possible lifespan.

Life expectancy at an older age also need not equal life expectancy at birth minus the age already attained. Surviving to an older age removes the possibility of deaths that could have occurred earlier.

Numerical example

Suppose a hypothetical cohort has 100,000 survivors at age 50, with a probability of death over the next five years of 0.02. Then

5d50=100,000(0.02)=2,000,{}_5d_{50}=100{,}000(0.02)=2{,}000,

and

l55=98,000.l_{55}=98{,}000.

If deaths are assumed to occur, on average, halfway through the interval,

5L50=5(98,000)+2.5(2,000)=495,000{}_5L_{50} = 5(98{,}000)+2.5(2{,}000) = 495{,}000

person-years.

If the table gives T50=3,200,000T_{50}=3{,}200{,}000 remaining person-years, then

e50=3,200,000100,000=32 years.e_{50} = \frac{3{,}200{,}000}{100{,}000} = 32\text{ years}.

This is a hypothetical illustration, not an estimate for a particular population.

Continuous formulation

Let XX be a nonnegative random variable representing age at death. Its survival function is

S(x)=Pr⁡(X>x).S(x)=\Pr(X>x).

If l0l_0 is the radix, the continuous counterpart of the survivor column is

l(x)=l0S(x).l(x)=l_0S(x).

Where a probability density f(x)f(x) exists, the instantaneous mortality rate, or force of mortality, is

μ(x)=f(x)S(x).\mu(x)=\frac{f(x)}{S(x)}.

This is the hazard function of age at death. For an absolutely continuous lifetime distribution,

S(x)=exp⁡(−∫0xμ(u) du).S(x) = \exp\left(-\int_0^x\mu(u)\,du\right).

Conditional survival over an interval is therefore

npx=S(x+n)S(x)=exp⁡(−∫xx+nμ(u) du).{}_np_x = \frac{S(x+n)}{S(x)} = \exp\left(-\int_x^{x+n}\mu(u)\,du\right).

If the hazard is constant at μ\mu over the interval,

nqx=1−e−μn.{}_nq_x=1-e^{-\mu n}.

Thus multiplying an annual mortality rate by an interval’s length is generally only an approximation to the probability of death.

Provided the expectation is finite and S(x)>0S(x)>0, complete remaining life expectancy is

ex=1S(x)∫x∞S(u) du.e_x = \frac{1}{S(x)} \int_x^\infty S(u)\,du.

In actuarial terminology, this complete expectation of life includes fractional years. The curtate expectation of life counts only complete future years lived.

Extensions and applications

Insurance and pensions

Life tables supply survival and death probabilities for valuing life insurance, annuities, and pension liabilities. Financial calculations combine these probabilities with benefit amounts and discounting.

Tables may distinguish an insured population from the general population. Select tables allow mortality to depend on time since selection, such as entry into an insured group, while ultimate tables describe mortality after the selection effect is assumed to have dissipated. Mortality projections may also allow survival to vary with calendar time.

Population analysis

Life tables allow comparisons of mortality schedules without directly inheriting the observed age composition of each population. Two populations can have different crude death rates partly because one contains a larger share of older people; life tables instead organize mortality age by age.

They also provide survival inputs for population projections when combined with fertility and migration assumptions.

Survival analysis and censoring

In statistics, an actuarial or grouped survival table estimates survival across time intervals when some observations end before the event occurs. Such observations are censored.

A common grouped method adjusts the number at risk by assuming withdrawals occur approximately uniformly within each interval. The Kaplan–Meier estimator, by contrast, updates survival at observed event times rather than requiring fixed intervals.

These methods require appropriate assumptions about censoring. Ordinary estimates can be biased when leaving observation is related to unobserved event risk.

Multiple-decrement and multistate tables

A multiple-decrement table distinguishes several mutually exclusive ways of leaving an initial state, such as death from different causes or exit from employment for different reasons. It is closely related to competing-risks analysis.

A multistate life table allows transitions among several states, potentially including returns to earlier states. Examples include movement between employment states or between disability states, with death as an absorbing state. Such models may use Markov chains or other transition models.

Observed mortality from one cause cannot generally be removed while leaving all other outcomes unchanged without additional assumptions. Cause-deleted life tables are therefore model-based constructions rather than direct observations of what eliminating a cause would produce.

Historical development

John Graunt’s analysis of London’s bills of mortality in 1662 was an early contribution to numerical descriptions of survival. His work preceded modern age-specific death registration and relied on limited data and indirect reasoning.

In 1693, Edmond Halley published a mortality table based on records from Breslau, now Wrocław. He demonstrated its use in valuing life annuities, connecting mortality measurement with financial calculation.

Later developments in registration systems, censuses, probability, and actuarial mathematics made life tables more detailed and systematic. Modern statistical agencies routinely construct period tables, while insurers, pension institutions, and researchers develop tables for particular populations and purposes.

Limitations and interpretation

Life-table results depend on both data quality and modeling choices.

  • Incomplete records: Underregistered deaths, inaccurate population counts, and incorrect ages can distort mortality estimates.
  • Small populations: Sparse deaths create unstable age-specific rates. Combining years, smoothing rates, or using statistical models can improve stability but introduces additional assumptions.
  • Population heterogeneity: A table averages across individuals with different circumstances and risks. It does not determine any particular individual’s lifetime.
  • Changing mortality: Period tables hold a period’s mortality schedule fixed for calculation. They do not automatically incorporate future improvements, deterioration, or temporary shocks.
  • Interval assumptions: Converting death rates to probabilities depends on assumptions about the timing of deaths within age groups.
  • Oldest-age treatment: Age misreporting and limited observations complicate the estimation of mortality in the final age groups.
  • Statistical uncertainty: Published point estimates may be accompanied by confidence intervals, but uncertainty about future mortality also involves projection assumptions.
  • Causal interpretation: A difference between two life tables is descriptive. Attributing it to a particular exposure, policy, or intervention requires a separate causal inference framework.

The synthetic cohort in a standard period table is closed to migration and declines only through the specified mortality schedule. Migration can nevertheless affect the observed population data used to estimate that schedule, particularly when death and exposure records cover populations inconsistently.