How the number is calculated

Step one: the life table

Everything starts from the Social Security Administration’s 2023 period life table, as used in the 2026 Trustees Report. It gives q(x) — the probability that a person at exact age x dies within the next year — separately for men and women, from age 0 to 119. It is U.S. government work and in the public domain.

Remaining life expectancy is computed from that table in the standard way: carry a hypothetical cohort forward year by year, applying q(x) at each age, and total the person-years lived. Our implementation reproduces the SSA’s own published figures to two decimal places — 75.79 for men at birth, 81.06 for women, 18.12 for men at 65.

If you decline to give a sex, the model averages the two tables, which is the conventional unisex approach.

Step two: re-basing onto your country

The SSA table describes the United States. To use it anywhere else, we solve for the single multiplier k on the force of mortality that makes the table reproduce your country’s published life expectancy at birth. Japan comes out around k = 0.63; Nigeria around k = 4.8.

Hazard ratios multiply the force of mortality, not the probability of death directly. The two are related by q = 1 − e^(−m), so the model converts, scales, and converts back. Scaling q directly would produce probabilities above 1 at old ages.

This is a real simplification. A country’s mortality does not differ from the U.S. by one flat multiplier at every age — high-mortality countries are usually much worse in infancy specifically. So the country adjustment is most accurate in the middle of life, which is where most people using this are.

Step three: your answers

Each answer carries an all-cause mortality hazard ratio taken from large cohort studies and meta-analyses. They are expressed relative to the population mix rather than to the healthiest possible group, so that someone answering “average” to everything lands close to the life table rather than well below it.

How the factors combine

Multiplying ten hazard ratios together assumes they are independent, and they are obviously not — the person sleeping four hours a night is disproportionately likely to be the same person who is stressed, isolated and sedentary. Naive multiplication would hand a thirty-year-old a hazard ratio near 40 and kill them off in the model within a decade.

So the log-hazards are summed and then passed through a curve that is exactly linear up to a threshold and saturates beyond it. The threshold sits just above the largest single factor in the table, which means any one risk on its own passes through untouched and moves the clock by precisely the amount the literature says it should. Compression only begins once risks genuinely stack. The result is bounded between roughly 0.14 and 7.4.

Does it agree with anything?

The strongest external check available is Li et al., Circulation 2018, which followed 123,219 people and found that women with five low-risk lifestyle factors lived about 14.0 years longer at age 50 than women with none.

Running the same comparison through this model gives 12.7 years. That is a 1.3-year discrepancy on a 14-year effect, without any tuning against that study — the calibration falls out of the hazard ratios themselves.

What this model cannot do

• It cannot predict your death. It produces a population average for people who answered the way you did. The survival band on your result — the 9-in-10, 1-in-2 and 1-in-10 ages — is the honest shape of the answer. The single date is a headline.
• It does not know your medical history beyond one coarse question, and it cannot see medication, genetics beyond a family-history proxy, or anything a doctor would actually test.
• It cannot see accidents. Roughly a third of deaths under 40 are external causes, and no lifestyle questionnaire predicts those.
• Hazard ratios come from observational studies, which show association, not proof of cause. They are adjusted for confounders, imperfectly. Smoking is about as close to settled causation as epidemiology gets; the evidence on stress and social connection is real but softer.
• It assumes today’s mortality lasts forever. A period life table applies current death rates to your whole remaining life. If medicine keeps improving, this understates how long you will live. Cohort projections generally run a few years higher.
• Fun mode adds a small deterministic wobble of up to about ±1.75 years, so the number feels less like a spreadsheet. Serious mode does not.

Privacy

The entire calculation runs in your browser. There is no account, no database, and no request carrying your answers to any server. Your result is encoded into the share link after the #, and browsers never transmit that part of a URL — so it does not appear in server logs, ours or anyone else’s. See the privacy policy for the full detail.

Corrections

If you think a hazard ratio is wrong, or you have a better source, please say so. The whole point of publishing the method is that it can be argued with.