Age-Period-Cohort Model
Also known as: APC Model, Age-Period-Cohort Analysis, Holford APC Model
The age-period-cohort (APC) model decomposes variation in a vital rate — mortality, incidence, fertility — into three temporal dimensions: the age of individuals, the calendar period of observation, and the birth cohort to which they belong. It is the standard framework for asking whether a trend reflects how risk changes with age, contemporaneous period influences affecting all ages at once, or generational effects carried by successive cohorts. Its defining technical challenge is that cohort equals period minus age, an exact linear dependence that makes the three sets of linear effects unidentifiable without further assumptions; Holford's 1983 formulation clarified exactly which quantities can and cannot be estimated.
Key highlights
- Provides a principled framework for separating age, period, and cohort sources of variation in vital rates.
- Estimable functions (curvatures and second differences) yield conclusions invariant to the arbitrary identifying constraint.
- Fits cleanly within the generalized linear model machinery, giving likelihood-based inference and standard errors.
- Reveals generational (cohort) signatures — such as a smoking or famine cohort — that pure age-period analyses miss.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use APC modeling when you have rates cross-classified by age and calendar period over a span long enough to distinguish generations, and you want to attribute trends to age, period, or cohort influences — for example separating a period intervention from a generational risk. It is the appropriate tool for studying disease incidence, mortality, and fertility trends across the Lexis surface. Assumptions: the chosen functional form (usually log-linear additive effects) is adequate, the three effects act additively on the log scale, and exposures and events are well measured. Do NOT interpret the individual age, period, or cohort linear slopes as identified — they are not — and do NOT impose an arbitrary constraint (such as setting two period effects equal) and read the resulting trends as real, since the identification problem makes such trends artefacts of the constraint.
Strengths & limitations
- Provides a principled framework for separating age, period, and cohort sources of variation in vital rates.
- Estimable functions (curvatures and second differences) yield conclusions invariant to the arbitrary identifying constraint.
- Fits cleanly within the generalized linear model machinery, giving likelihood-based inference and standard errors.
- Reveals generational (cohort) signatures — such as a smoking or famine cohort — that pure age-period analyses miss.
- The exact age = period − cohort identity makes the individual linear trends fundamentally non-identifiable; no amount of data can resolve this.
- Conclusions about linear trends require external assumptions, and different reasonable assumptions can give opposite stories.
- Results can be sensitive to the width and boundaries of age and period groupings, which determine the cohort definition.
- The additive log-linear form may be too rigid when effects interact, requiring more complex and harder-to-identify models.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
Why can't the age, period, and cohort effects all be estimated?
Because cohort is exactly period minus age, the three predictors are perfectly linearly dependent: you can add a trend to one effect and subtract it from another and obtain identical fitted rates. This means the individual linear components are not identifiable from the data, no matter how large the sample. Only nonlinear features — curvatures and deviations from trend — are estimable.
What is an estimable function in APC analysis?
An estimable function is any combination of the parameters whose value is the same regardless of which identifying constraint is imposed. Holford showed the second differences (curvatures) of the age, period, and cohort effects are estimable, so statements about acceleration, deceleration, and turning points are valid while statements about overall slopes are not.
Is the intrinsic estimator a solution to the identification problem?
The intrinsic estimator picks a particular solution using the structure of the design matrix, and it produces a unique, reproducible answer. However, critics note that it still embeds an implicit assumption to break the linear dependence, so its estimated trends carry that assumption rather than resolving the fundamental non-identifiability. It should be used with awareness of this caveat.
Sources
- 1.Holford, T. R. (1983). The estimation of age, period and cohort effects for vital rates. Biometrics, 39(2), 311–324.
- 2.Preston, S. H., Heuveline, P., & Guillot, M. (2001). Demography: Measuring and Modeling Population Processes. Blackwell.ISBN 9781557864512
You have read it. What now?
Cite this page
ScholarGate. (2026, June 22). Age-Period-Cohort Model. ScholarGate. https://scholargate.app/demography/age-period-cohort-model