How long you'll live

How accurate are life expectancy calculators?

By GyroscopeUpdated November 14, 20263 sources

They’re good for groups and rough for one person. A careful calculator gets the average right for people like you, but people like you still spread across 15 years or more. Trust one that gives a range and prints its formula, and walk away from any that gives you a date.

Why do two calculators give you ten different years?

Put the same answers into a few online calculators and you can get results a decade apart. That makes people think they’re all random. They differ for plain reasons. Some start from an old life table, or another country’s. Some ask only your age and sex. Others ask 30 questions and add or take away a year for each answer, which counts the same thing twice, since people who exercise also tend to smoke less and eat better.

How a good one works

It has three steps, and you should be able to see all of them.

1. Start from the national life table for your age and sex. Ours is the 2024 US table, and it already knows that a 65-year-old woman reaches about 86 on average.
2. Adjust by how much your habits changed risk in a big study. Ours uses the five habits from a study of 123,219 US adults followed for up to 34 years (Li 2018). The researchers applied each group’s risk to the US life table, and we do the same.
3. Show a range. The study’s own uncertainty goes at both ends, rounded outward to whole years.

The simple version

A calculator is like a flight’s scheduled arrival time. Across thousands of flights it’s close to right on average. Your own flight still lands early or late. An app that told you the exact second your wheels would touch down would be inventing it, and that’s what a calculator with a date is doing.

How accurate can they get?

Doctors use the best-tested versions of these tools to judge how long older patients may live. In 2012, researchers reviewed 16 of them in JAMA, all built for adults around 60 and over. Thirteen scored a C statistic of 0.70 or higher, and none reached 0.90. Only two had been checked by researchers who didn’t build them.

A C statistic is easier than it sounds. Take two people, one who died during the study and one who didn’t. A score of 0.70 means the tool gave the higher risk to the right person 70 times in 100. That’s good for sorting people into groups and nowhere near good enough to name a year for one person.

The life table shows the same thing from the other side. Of American women who are 50 today, the middle half will live to somewhere between 77 and 92. No questionnaire can squeeze that spread down to one age, because a lot of what decides it, from a car crash to a lucky gene, isn’t in any form.

The red flags

• It gives a date, a countdown or a cause of death. No published model can do this.
• It gives one number with no range.
• It hides how the number is worked out.
• It adds a year for every good answer, which double counts habits that travel together.
• It wants your contact details first.

Our calculator gives a range, asks for no contact details, and prints its formula under the result.