Multi-State Demography

Core Idea

Multi-state demography extends classical demographic methods — which track transitions between two states (alive and dead, present and absent) — to model populations where individuals can occupy and move between multiple states simultaneously. Examples include marital status (single, married, divorced, widowed), health status (healthy, disabled, recovered), labor force status (employed, unemployed, not in labor force), and region of residence. The framework, developed primarily by Andrei Rogers in the 1970s and formalized through increment-decrement life tables and multi-state population projections, uses age-specific transition rates between states to compute state-specific life expectancies, prevalence, and population composition. Multi-state models are the foundation of health expectancy calculations (how many years a person can expect to live in good health versus disability) and have become essential tools for analyzing aging populations, migration flows, and the dynamics of family formation and dissolution.

Explainer

Classical demography developed powerful tools for analyzing populations moving through a single transition: from alive to dead (life tables), from childless to parent (fertility analysis), from one region to another (migration analysis). But real populations are more complex than these binary transitions suggest. A person is simultaneously in a marital state, a health state, a labor force state, and a geographic state — and all of these change over the life course in ways that interact with each other and with mortality. Multi-state demography provides the mathematical framework for modeling these parallel, interacting processes.

The intellectual foundation was laid by Andrei Rogers in the 1970s, who extended matrix population models to handle multiple interacting states. The key innovation is the increment-decrement life table, which generalizes the standard life table by allowing individuals to move between multiple living states as well as into the absorbing state of death. In a standard life table, the only transition is alive-to-dead. In a multi-state table with three health states (healthy, disabled, dead), there are five possible transitions at each age: healthy-to-disabled, disabled-to-healthy, healthy-to-dead, disabled-to-dead, and staying in the current state. The age-specific rates of all these transitions are organized into a transition matrix, and a synthetic cohort is followed through these matrices from birth (or some starting age) to extinction. The output is a set of state-specific life expectancies: the expected number of years spent healthy, the expected number of years spent disabled, and total life expectancy as their sum.

The most prominent application of multi-state demography is the calculation of health expectancy — the number of years a person can expect to live in good health. The World Health Organization's Healthy Life Expectancy (HALE) indicator, published for every country, is derived from multi-state methods. As populations age and life expectancy increases, the question of whether additional years are spent in good health or in disability becomes a central policy concern. Multi-state models reveal that the answer varies dramatically across countries, socioeconomic groups, and sexes. Women typically live longer than men but spend more years in disability — a finding that only multi-state analysis can quantify precisely. Countries with similar total life expectancies can have very different health expectancies, reflecting differences in chronic disease patterns, healthcare quality, and social support systems.

Beyond health, multi-state demography has been applied to marital dynamics (calculating expected years spent married, divorced, or widowed), labor force participation (expected working life), migration (multi-regional population projections), and long-term care planning (estimating the probability and duration of needing institutional care). In each application, the framework's strength is its ability to capture the dynamics of reversible transitions — people recover from disability, remarry after divorce, re-enter the labor force after retirement — that simpler models cannot represent. Multi-state population projections, which forecast not just total population but its distribution across health, marital, educational, and geographic states, have become increasingly important for planning pension systems, healthcare infrastructure, and social services in aging societies.

The practical limitations of multi-state models center on data requirements and the Markov assumption. Estimating age-specific transition rates between multiple states requires either longitudinal panel data (following individuals over time) or repeated cross-sectional surveys with careful indirect estimation. The Markov assumption — that transition rates depend only on current state and age, not on history or duration — simplifies the mathematics but can be violated in important ways. Duration dependence in unemployment (the longer you are unemployed, the harder it is to find work) and health history effects (prior disability increases future disability risk) are well-documented violations. Semi-Markov extensions and microsimulation models address these limitations at the cost of greater complexity and data demands.