4 ms·

This is a fun read, however- "Life insurers can predict when you'll die with about 98% accuracy." This conclusion isn't supported by the linked document. The

This is a fun read, however- "Life insurers can predict when you'll die with about 98% accuracy." This conclusion isn't supported by the linked document. The document instead is talking about expected vs actual deaths among demographic groups as a whole, not individual people. And that expected vs actual is just history + trends. This doesn't mean that insurance can say that Joe Blow is going to die in June of 2027 with "98% accuracy", obviously.

PaulDavisThe1st 1y ago
Put a little differently, they can predict that of your cohort (defined somehow), after June 2027, only X percent of you will still be alive. Will you be one of them? Click here to find out!
- etothepii 1y ago
  For mortality tables age and sex are pretty much sufficient to get to 98% accuracy.
  - adgjlsfhk1 1y ago
    You definitely (at least in the US) need income (or an income proxy like zip code). Mississippi has a life expectancy ~10 years less than Massachusettes.
    - etothepii 1y ago
      This typically has surprisingly little impact on force of mortality. Men under 40 essentially have incredibly low mortality. Once you exclude cars and suicide it drops basically to zero. Even things that are very very bad for you, like being very obese (BMI>40), are factors that scale a mortality that is massively dominated by age and sex. Even 3x mortality at 25 rounds to zero. At 60 background become 1% per year and a 3x increase would make a huge difference. I would wager $50 that there are no contiguous collection of zip codes in the USA where the average 40 year old has a higher life expectancy than the average 30 year old in any other contiguous set of zip codes assuming there are over 1000 people of that age in both.
astrange 1y ago
> This doesn't mean that insurance can say that Joe Blow is going to die in June of 2027 with "98% accuracy", obviously. Pretty easy to predict if you're willing to make it happen.
- _9ptr 1y ago
  Life insurance is just another word for bounty
crazygringo 1y ago
Yeah that was a bizarre line in the article. Not to mention it's meaningless because it doesn't say within what time interval. But even if you assume a year (i.e. predict your age of death) it's obviously false. Life insurers are very much not predicting the year an individual will die and getting it right 98% off the time. That would be absurd.
ks2048 1y ago
I can predict pretty well "Will person X die this year" by saying "No". Yeah, this number seems meaningless without more context.
- Qem 1y ago
  Unless the preson in question is over 102 years. IIRC that is roughly the age when mortality rates on average surpass 50% per year. Up until tha age you can celebrate your birthday and reasonably expect is more likely to still be alive by the next than not.
paulpauper 1y ago
They can predict it in the sense most people will die within some specified window in which the insurer makes a profit. This is why its so profitable for the insurer. They have a very wide window where it's profitable and the vast majority of people, 98%, fall within this window. .
popupeyecare 1y ago
The link even has the source being ChatGPT - it hallucinated!

PaulDavisThe1st 1y ago

Put a little differently, they can predict that of your cohort (defined somehow), after June 2027, only X percent of you will still be alive. Will you be one of them? Click here to find out!

etothepii 1y ago
For mortality tables age and sex are pretty much sufficient to get to 98% accuracy.
- adgjlsfhk1 1y ago
  You definitely (at least in the US) need income (or an income proxy like zip code). Mississippi has a life expectancy ~10 years less than Massachusettes.
  - etothepii 1y ago
    This typically has surprisingly little impact on force of mortality. Men under 40 essentially have incredibly low mortality. Once you exclude cars and suicide it drops basically to zero. Even things that are very very bad for you, like being very obese (BMI>40), are factors that scale a mortality that is massively dominated by age and sex. Even 3x mortality at 25 rounds to zero. At 60 background become 1% per year and a 3x increase would make a huge difference. I would wager $50 that there are no contiguous collection of zip codes in the USA where the average 40 year old has a higher life expectancy than the average 30 year old in any other contiguous set of zip codes assuming there are over 1000 people of that age in both.

astrange 1y ago

> This doesn't mean that insurance can say that Joe Blow is going to die in June of 2027 with "98% accuracy", obviously. Pretty easy to predict if you're willing to make it happen.

_9ptr 1y ago
Life insurance is just another word for bounty

crazygringo 1y ago

Yeah that was a bizarre line in the article. Not to mention it's meaningless because it doesn't say within what time interval. But even if you assume a year (i.e. predict your age of death) it's obviously false. Life insurers are very much not predicting the year an individual will die and getting it right 98% off the time. That would be absurd.

ks2048 1y ago

I can predict pretty well "Will person X die this year" by saying "No". Yeah, this number seems meaningless without more context.

Qem 1y ago
Unless the preson in question is over 102 years. IIRC that is roughly the age when mortality rates on average surpass 50% per year. Up until tha age you can celebrate your birthday and reasonably expect is more likely to still be alive by the next than not.

paulpauper 1y ago

They can predict it in the sense most people will die within some specified window in which the insurer makes a profit. This is why its so profitable for the insurer. They have a very wide window where it's profitable and the vast majority of people, 98%, fall within this window. .

popupeyecare 1y ago

The link even has the source being ChatGPT - it hallucinated!