Human mortality database download: the harmonized life tables, delivered as rows

The orientation layer reads like this - both-sex life expectancy at birth, one value per country-year, exactly as it lands:

series : Both-sex life expectancy at birth (e0)   sheet : hmd_summary_ex_0_65_80.xlsx / Both-sex e0
year   : 2024                                     Belgium : 82.28       U.S.A. : 79.29

year   : 2023     Belgium : 82.05   U.S.A. : 78.76   Canada : 81.75   Chile : 80.59

Read what the numbers say before reading what they imply. The United States sits 2.99 years below Belgium at e(0) in 2024; Canada splits them at 81.75; Chile closes the row at 80.59. Underneath this summary runs the full machinery: ten columns per life-table row - m(x), q(x), a(x), l(x), d(x), L(x), T(x), e(x), with Year and Age keys - where a synthetic cohort of l(0) = 100,000 births marches from first birthday to last survivor.

• q(x) - probability of death between ages x and x+n. The raw material of valuation and projection; pricing teams start here and graduate outward.
• m(x) - central death rate, deaths over person-years exposed. The observed quantity q(x) is derived from, and the honest check when a graduation drifts too far from experience.
• l(x) - survivors at exact age x out of a radix of 100,000 (92,741 at age 65 in the documented example). Survival curves for annuity blocks read straight off this column.
• L(x) and T(x) - person-years lived in the interval and remaining above it. Reserving arithmetic lives here; T(x) above age 85 stands near 1.1 million in the worked row.
• e(x) - life expectancy at exact age x, 19.9 years at 65 in the documented example. The number that ends every table and starts every headline.

Age itself deserves respect: single-year resolution to 109 plus an open 110+ interval, laid out in six standard combinations (1x1, 1x5, 1x10, 5x1, 5x5, 5x10). Deaths additionally arrive as Lexis triangles of age x birth cohort x calendar year, which is the shape longevity models actually want and almost nothing else publishes ready-made.

Geography - 41 populations across North America, Europe, Asia-Pacific and Chile, and the splits matter more than the count. East and West Germany stay separately observable; France divides civilian from total; the UK breaks into England & Wales, Scotland, Northern Ireland and a UK total; New Zealand separates Maori, non-Maori and total. Inequality questions answerable nowhere else are answerable here.

Temporal - depth varies by design. Some series begin in the 18th or 19th century, giving mortality-improvement models two centuries of runway; leading countries extend to 2024 or later, with USA figures revised through 2024. Weekly STMF counts run from as early as 1990 (Finland) into 2026.

API, files, or your warehouse. Daily, weekly, or hourly.

Request a sample first - name the populations, the tables and the age grids - and the extract arrives shaped identically to the standing feed, so anything prototyped on it survives into production unchanged. Versioned releases carry DOIs, which keeps citation-grade work auditable after the fact; the method is written up under citation-grade research. See how the slice behaves at machine speed in the life & health insurance data APIs.

Three workflows dominate among insurance users:

Valuation calibration. Population life tables set the experience baseline; statutory work still needs published bases. Teams benchmark select-and-ultimate layouts pulled from the SOA Research Institute and Mortality Tables - 3,038 rate tables tracing to the 1941 CSO across the US plus roughly 55 other nations - against the HMD period table for the matching country and years, on identical single-year grids.

Cross-country longevity research. Because split series are published rather than estimated, trend studies can isolate East versus West Germany or Maori versus non-Maori New Zealand without extra modeling. Where a market falls outside the 41 populations, the World Bank Life Expectancy and Mortality Indicators extend the comparison to 200+ economies over 1960-2024 - roughly 17,500 observations on the flagship life-expectancy series alone.

Excess-mortality monitoring. STMF weekly counts feed survivorship and claims-assumption reviews between annual cycles, with UK ONS Health and Social Care Statistics providing an independent weekly second read for England and Wales into 2026.

Four boundaries define the record, and named companions fill each:

Stacked deliberately, the four fill every gap the core leaves open - which is the standard Datadory configuration for this slice.

Ranked by how directly the life tables answer their day job:

1. Actuaries & valuation teams get complete ten-column tables on standard grids - assumption setting without a reconciliation project bolted on.
2. Investors & quants price pension risk, annuity blocks and longevity swaps off two centuries of trend plus a weekly pulse; see investors & quants.
3. Data scientists & modelers train forecasting models on centuries of consistent observations - holdout decades most datasets simply lack; see data scientists.
4. Journalists, academics & students cite the demographic reference itself, with versioned releases behind every figure; see journalists, academics & students.
5. Market researchers & consultants read population structure and survival for aging-economy sizing across developed markets; see market researchers.

Because the hard part was never the first extract - it is the tenth. Country archives whose vintage dates differ mid-analysis. Age layouts that change shape between populations. Cohort tables keyed differently from period ones. Weekly counts that arrive on another calendar entirely. Each is survivable once; none is fun to re-solve in every new notebook.

Datadory normalizes before delivery: one population key across every series, six age layouts typed as explicit choices, cohort and period tables distinguished at the column level, and the weekly STMF stream joined to the annual spine on a documented calendar. Suppressed or lagging vintages surface explicitly rather than being discovered mid-model.

Name the countries, tables and age grids when you request a sample and it arrives already cut to that scope - the production feed follows the same shape, so anything prototyped on the sample survives delivery intact.

Start with the Human Mortality Database (HMD) dataset page for the full ten-field dictionary and to request a sample cut to your populations. For the head-to-heads, read Human Mortality Database vs Medical Expenditure Panel Survey and NAIC Insurance Data vs World Bank Life Expectancy and Mortality Indicators. For the whole pool, the Life & Health Insurance Data Guide covers all 21 records in the slice, and the life & health insurance hub holds the pooled view with the scorecard side by side. The definition underneath it all lives in the glossary entry for life table.