Inverse Projection
- Introduced by
- Ronald Lee; E. A. Wrigley and R. S. Schofield; generalized by Jim Oeppen
- Year
- 1981
- Type
- Demographic reconstruction from aggregate flows
- Family
- Aggregative reconstruction
What does it do?
Inverse projection, and its more flexible successor generalized inverse projection, reconstructs the demographic history of a population from the outside in.
Where conventional cohort-component projection runs a known population forward using assumed rates, inverse projection runs the logic backward: starting from a population of known size and age structure at one date, and given annual totals of births and deaths, it infers the population sizes, age distributions, mortality levels, life expectancies and fertility rates that must have prevailed in earlier years. The technique was devised by Ronald Lee and applied by Wrigley and Schofield to their English aggregative series, allowing three centuries of population history to be recovered without any direct census before 1801. Jim Oeppen's generalization relaxed restrictive assumptions about migration and closed populations. The method is the bridge that turns raw counts of vital events into a fully articulated demographic regime.
The intuition behind the method
Picture a reservoir whose water level you cannot see for most of its history, but whose inflow (births) and outflow (deaths) you have metered every year, and whose level you happen to know at one recent moment. By stepping backward year by year—adding back the deaths that have since occurred, removing the births that had not yet happened—you can reconstruct the level at every earlier date. Inverse projection does this for a population, but with the added subtlety that people are not interchangeable litres: they age, and their chance of dying depends on that age. So the method borrows a family of model life tables, picking the mortality level each year that makes the total deaths come out right, then ages the survivors backward through the table. Anchored to the one known population, the whole demographic past unfolds—sizes, age pyramids, life expectancies—consistent with the metered flows of births and deaths.
History
Ronald Lee introduced inverse projection in the 1970s as a way to extract mortality and fertility from aggregate vital series. Wrigley and Schofield adopted and adapted it—their variant termed back projection—as the analytical engine of The Population History of England, 1541-1871 (1981), reconstructing the nation's demography from parish counts anchored to nineteenth-century censuses. The closed-population assumption and the practice of working only backward drew criticism, prompting Jim Oeppen's generalized inverse projection, which incorporated net migration and allowed boundary conditions at both ends, placing the method on firmer mathematical footing. Subsequent refinements and the spread of model life tables made inverse projection a standard instrument wherever vital flows survive but censuses do not.
When to use it
Use inverse projection when you possess long, reliable annual series of total births and deaths and at least one trustworthy population benchmark, but lack the regular censuses needed for direct demographic measurement. It is the standard tool for converting aggregative parish-register series into full demographic reconstructions—population size, age structure, life expectancy and fertility—across the pre-census era. It suits national or large regional populations where migration is modest or can be estimated, and where a model life-table family plausibly captures the mortality pattern. It is inappropriate for small, highly migratory populations, or where the birth and death series are too short, too incomplete, or too poorly dated to support reverse cohort accounting.
Limitations and considerations
- Depends on the assumed model life-table family, which imposes a mortality shape
- Sensitive to errors and under-registration in the input birth and death series
- Classic back projection assumes a closed population, mishandling migration
- Requires an accurate anchoring population; benchmark error propagates throughout
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Recovers full demographic regimes from aggregate flows without regular censuses
- Internally consistent: outputs satisfy birth, death and population accounting identities
- Generalized versions accommodate net migration and two-sided boundary conditions
- Turns inexpensive aggregative series into rich age-structured reconstructions
Applications
- Reconstructing English population size and structure from 1541 to 1871
- Estimating pre-census life expectancy and its long-run movements
- Deriving fertility trends from birth series and reconstructed populations
- Testing Malthusian versus nuptiality-driven models of population regulation
- Reconstructing demographic history of other countries with long vital but no census data
How it works
How it works
The method cannot estimate a full age schedule of mortality from a single total of deaths, so it borrows shape from a family of model life tables—Coale-Demeny or Princeton regional patterns—indexed by a single level parameter. Choosing the appropriate family encodes prior knowledge of the population's mortality structure; the level parameter is then the unknown the procedure solves for in each period, converting one aggregate death count into a complete set of age-specific survival probabilities.
Formula
{}_nL_x = f(\text{level}, \text{model life table family})Given the survival ratios implied by the chosen model life table at a given level, each age group is projected to the next age and period: survivors equal the prior group scaled by the ratio of person-years lived. Run in reverse, this bookkeeping recovers how many people of each age must have existed earlier to produce the observed survivors. It is ordinary cohort-component accounting, inverted so that the unknown is the past rather than the future.
Formula
N_{x+1,t+1} = N_{x,t} \cdot \frac{{}_nL_{x+1}}{{}_nL_x}Two accounting identities anchor the system: the youngest age group in each year equals that year's births, and the net change in total population equals births minus deaths (plus net migration in generalized versions). These constraints tie the abstract survival arithmetic to the observed aggregate series, ensuring the reconstructed age structure is consistent with the actual recorded flows of entrances and exits.
Formula
N_{0,t} = B_t, \quad \sum_x \Delta N_{x,t} = B_t - D_tWithin each year the mortality level is not known in advance, so the procedure searches for the level whose implied age-specific death rates, applied to the reconstructed age structure, reproduce the observed total deaths exactly. This is an iterative solve—adjust the level, recompute deaths, compare to the target—repeated until the books balance. The fitted level then yields that year's complete life table, including life expectancy at birth.
Formula
\text{level}_t : \sum_x N_{x,t}\,{}_nm_x \overset{!}{=} D_tOnce each year's age structure and mortality level are pinned down, the demographic indicators of interest follow directly: total population, the age pyramid, expectation of life at birth, and—by relating births to the female population at risk—gross and net reproduction and total fertility. The output is a year-by-year demographic regime, the synthetic history that the aggregate birth and death series and the anchoring population jointly imply.
Formula
e_0(t), \text{TFR}(t), N(t) \leftarrow \{N_{x,t}\}The reconstruction must satisfy a boundary condition: at the anchoring date the reconstructed population must match the known census or estimate in size and structure. Back projection runs from this terminal census into the past; generalized inverse projection can fit boundary conditions at both ends and accommodate net migration, making the closure assumption explicit and testable rather than silently imposed.
Formula
N(T) = N_{\text{census}}(T) \text{ (boundary condition)}Common pitfalls
- Choosing a model life-table family inconsistent with the population's true age pattern of mortality
- Feeding in uncorrected vital series, so under-registration distorts the reconstruction
- Ignoring migration in periods when it materially affected the population
- Over-interpreting year-to-year estimates whose precision the method cannot support
Frequently asked questions
How does it differ from ordinary projection?
Standard cohort-component projection takes a known population and assumed future rates and computes the population forward. Inverse projection reverses the unknown: it takes observed aggregate births and deaths plus an anchoring population and infers the past populations, age structures and rates consistent with them. One predicts the future from rates; the other reconstructs the past from flows and a benchmark.
Why are model life tables necessary?
A single annual total of deaths cannot reveal how mortality was distributed across ages. Model life tables supply that age pattern from a family of empirically derived schedules indexed by one level parameter. The method then solves for the level each year that reproduces the observed deaths. Without this borrowed shape, converting aggregate deaths into age-specific survival would be underdetermined.
What did generalized inverse projection fix?
Classic back projection assumed a closed population and worked only from a terminal census backward, mishandling migration and depending on one boundary. Oeppen's generalized version lets net migration enter the accounting and permits boundary conditions at both ends of the series, improving stability and realism. It made the previously implicit closure assumptions explicit and allowed reconstructions for populations with significant migration.
Related methods
Sources
Wrigley, E. A., & Schofield, R. S. (1981). The Population History of England 1541-1871: A Reconstruction. Edward Arnold / Harvard University Press.
ISBN: 9780674690073Copied.Wrigley, E. A., Davies, R. S., Oeppen, J. E., & Schofield, R. S. (1997). English Population History from Family Reconstitution 1580-1837. Cambridge University Press.
ISBN: 9780521590150Copied.