Lexis Diagram

Lexis diagram is a two-dimensional graphical and analytical coordinate system in demography, actuarial science, and longitudinal epidemiology that plots calendar time (period, P) along the horizontal abscissa against chronological age (A) along the vertical ordinate, so that each individual's life course traces a continuous diagonal lifeline inclined at a 45° angle (slope = 1 year of age per 1 year of calendar time). Introduced independently in the 1860s and 1870s by German statisticians Gustav Zeuner (1869), Karl Friedrich Hermann Knapp (1868), and Wilhelm Lexis (1875)—after whom the diagram is named following his influential monograph Einleitung in die Theorie der Bevölkerungsstatistik—the Lexis plane provides the geometric foundation in biostatistics for defining birth cohorts, calculating person-years at risk in a cohort study, constructing actuarial life tables, and disentangling Age–Period–Cohort (APC) effects.

Lifelines, Events, Censoring, and Birth Cohorts

When identical linear scales (such as 1 year per unit length) are used on both the horizontal calendar-time axis (t or P) and the vertical age axis (a or A), every person born at calendar date C = t₀ enters the diagram at the baseline point (t₀, 0) on the horizontal axis and moves upward and to the right along the straight line:

• Life Course Trajectory (45° Lifeline): An individual's lifeline continues diagonally until it terminates at an event point (t_d, a_d) (marked by a solid dot ● representing death or disease onset) or exits observation via right-censoring (emigration or study end, used in the Kaplan–Meier estimator and proportional hazards model).
• Exact Person-Time Integration: The total person-years at risk (PT) contributed by a dynamic cohort inside any geometric region Ω of the Lexis plane equals the sum of the vertical (or horizontal) projections of all lifeline segments traversing Ω.

Elementary Lexis Polygons and Demographic Aggregation Shapes

Partitioning the Lexis grid by integer calendar years (t, t + 1) and integer single-year ages (a, a + 1) creates four fundamental geometric shapes used to tabulate vital statistics registries:

Biological Meaning of Age, Period, and Cohort Dimensions

When analyzing long-term disease incidence or mortality trends (such as tuberculosis, gastric carcinoma, or cardiovascular mortality), rates vary across all three temporal dimensions of the Lexis diagram:

1. Age Effects (A): Reflect intrinsic biological aging, immunological senescence, developmental stages, or cumulative somatic mutation across an individual's lifespan, independent of the historical era in which the person lived.
2. Period Effects (P): Reflect contemporaneous historical, environmental, diagnostic, or medical shocks that simultaneously affect all age groups living in a given calendar year (such as the 1918 H1N1 influenza pandemic, the introduction of penicillin in the 1940s, clean-air legislation, or a change in ICD diagnostic coding).
3. Cohort (Generation) Effects (C = P − A): Reflect early-life exposures, nutritional imprinting, or generational behavioral habits shared by individuals born in the same era that accompany them diagonally across the Lexis plane as they age (for example, Helicobacter pylori acquisition in childhood driving generational declines in stomach cancer, or generational cigarette-smoking uptake).

The Linear Collinearity Identifiability Problem

In a generalized linear Poisson regression modeling the log-incidence rate ln(λ_ap) in age group a and calendar period p:

where c = p − a denotes the birth cohort index, the three temporal predictors satisfy the exact linear dependency Period − Age − Cohort = 0. Because the design matrix X is rank-deficient by 1, adding an arbitrary linear drift v to the age slope, subtracting v from the period slope, and adding v to the cohort slope leaves every predicted cell rate λ_ap unchanged:

As shown by Theodore Holford (1983) and allied biostatistical theory, while the individual linear slopes of Age, Period, and Cohort are non-identifiable without an external biological constraint, the second differences (curvatures / nonlinear deviations from linearity)—analogous to the second-difference operator in Farr's laws—and the sum of the period and cohort slopes (net drift) are uniquely estimable.

Modern computational demography extends the classic line diagram into a Lexis contour surface, plotting mortality or disease incidence intensity h(a, t) as a color-coded heat map or topographic elevation surface over the (P, A) plane (Vaupel et al., 1998). In occupational cohort studies and nested case–control studies, splitting each participant's follow-up trajectory across a multi-scale Lexis grid (tracking attained age, calendar period, and time since first occupational exposure simultaneously) allows Poisson regression and time-dependent proportional hazards models to separate latency effects from biological aging.

Why did Wade Hampton Frost use cohort diagonals rather than cross-sectional age curves to study tuberculosis?

In a classic 1939 paper, Wade Hampton Frost (co-creator of the Reed–Frost model) examined Massachusetts tuberculosis mortality from 1880 to 1930. In cross-sectional period curves (reading vertically up a Lexis column), tuberculosis mortality appeared to peak in old age (~60–70 years), suggesting late-life biological vulnerability. However, when Frost plotted mortality along 45° birth-cohort diagonals, every single generation exhibited its true biological peak in young adulthood (age 20–29); older adults in 1930 had higher rates simply because they belonged to earlier nineteenth-century birth cohorts that had experienced much higher childhood infection rates.

Why must a Lexis square be split into two triangles to isolate a single birth cohort?

Individuals who are age a during calendar year t belong to two different birth cohorts: those who have already celebrated their a-th birthday in year t were born in calendar year C = t − a (lower Lexis triangle), whereas those who have not yet reached their (a + 1)-th birthday in year t were born in the preceding calendar year C = t − a − 1 (upper Lexis triangle).

1. Lexis, W. (1875). Einleitung in die Theorie der Bevölkerungsstatistik [Introduction to the Theory of Population Statistics]. Karl J. Trübner, Strasbourg. https://doi.org/10.1515/9783111476193
2. Frost, W. H. (1939). "The age selection of mortality from tuberculosis in successive decades." American Journal of Hygiene, 30(Section A), 91–96. https://doi.org/10.1093/oxfordjournals.aje.a118570
3. Holford, T. R. (1983). "The estimation of age, period and cohort effects for vital rates." Biometrics, 39(2), 311–324. https://doi.org/10.2307/2531004
4. Keiding, N. (1990). "Statistical inference in the Lexis diagram." Philosophical Transactions of the Royal Society of London. Series A, 332(1627), 487–509. https://doi.org/10.1098/rsta.1990.0128
5. Vandeschrick, C. (2001). "The Lexis diagram, a misnomer." Demographic Research, 4, 97–124. https://doi.org/10.4054/DemRes.2001.4.3