I don't.
I think people overestimate the possibility of getting predictions for an individual that is useful for planning. With regard to inheritance and supposed "good genes," in the Framingham study there was only a 7% correlation between parent and child lifespan; the correlation with great-aunt Vera would surely be negligible. And the effect of the factors we can know about is small compared to the range of individual variation.
Let's take a 75% range (i.e. 12.5% and 87.5% percentiles) as a reasonable "planning" number. 75% of 65-year old males would live to these ages. My source was the Social Security life tables--other tables would give greater or lesser life expectancies but the general picture would be the same.
![Image]()
Look at how wide that range is. A male could anywhere from 6 to 27 years in retirement, a factor of more than four. And falling outside that range, living less than 6 or more than 27 years has a 25% chance, that's no black swan.
For females:
![Image]()
The range of reasonably likely retirement lengths is 9 to 29 years.
Superimposing the charts:
![Image]()
Note that the difference between male and female is one of the larger predictable differences. Women live an average of about three years more; in the SSA table, life expectancy at age 65 is 16.95 years for men, 19.75 years for women. Certainly three years of extra life is important, but in context, you're shifting a twenty-year range by three years.
Any other factor you can find will be similar, probably smaller. Adding or subtracting a several years of life expectancy is critically important for insurance companies, for the Social Security administration, for setting public policy. But it can't be used for individual retirement planning.
The big exception, of course, occurs when you have personally received a medical diagnosis that you personally have a serious condition, one that limits life expectancy, and that actually has known statistics. But the life expectancy tools I've seen only take into account vague factors like smoking, weight, and fitness. They don't have little checkboxes for multiple sclerosis, atrial fibrillation, or diabetes. The insurance companies probably have the numbers that could be used for that, but they don't seem to share them.
By the way I saw something the other day about an "AI powered" longevity calculator. Yeah right. The best current AI isn't going to be able to overcome the inherent uncertainty--well, inherent as of 2024--of human life.
I think people overestimate the possibility of getting predictions for an individual that is useful for planning. With regard to inheritance and supposed "good genes," in the Framingham study there was only a 7% correlation between parent and child lifespan; the correlation with great-aunt Vera would surely be negligible. And the effect of the factors we can know about is small compared to the range of individual variation.
Let's take a 75% range (i.e. 12.5% and 87.5% percentiles) as a reasonable "planning" number. 75% of 65-year old males would live to these ages. My source was the Social Security life tables--other tables would give greater or lesser life expectancies but the general picture would be the same.
Look at how wide that range is. A male could anywhere from 6 to 27 years in retirement, a factor of more than four. And falling outside that range, living less than 6 or more than 27 years has a 25% chance, that's no black swan.
For females:
The range of reasonably likely retirement lengths is 9 to 29 years.
Superimposing the charts:
Note that the difference between male and female is one of the larger predictable differences. Women live an average of about three years more; in the SSA table, life expectancy at age 65 is 16.95 years for men, 19.75 years for women. Certainly three years of extra life is important, but in context, you're shifting a twenty-year range by three years.
Any other factor you can find will be similar, probably smaller. Adding or subtracting a several years of life expectancy is critically important for insurance companies, for the Social Security administration, for setting public policy. But it can't be used for individual retirement planning.
The big exception, of course, occurs when you have personally received a medical diagnosis that you personally have a serious condition, one that limits life expectancy, and that actually has known statistics. But the life expectancy tools I've seen only take into account vague factors like smoking, weight, and fitness. They don't have little checkboxes for multiple sclerosis, atrial fibrillation, or diabetes. The insurance companies probably have the numbers that could be used for that, but they don't seem to share them.
By the way I saw something the other day about an "AI powered" longevity calculator. Yeah right. The best current AI isn't going to be able to overcome the inherent uncertainty--well, inherent as of 2024--of human life.
Statistics: Posted by nisiprius — Mon Dec 02, 2024 8:58 am — Replies 26 — Views 1535