Thread by @DellAnnaLuca

THE LINDY PRINCIPLE, GENERALIZED

The Lindy Principle, from Taleb's "Antifragile", uses age to infer information about life expectancy.

Here, I propose an intuitive justification and I explore practical uses other than estimating life expectancy.

(a thread, 1/N)

2/ For a person, every year of life reduces its conditional life expectancy. A 70 years-old is expected to live 14.4 more years, and a 71yo is only expected to live 13.7 more years.

Conversely, the longer a book is on the NYT bestseller list, the longer it's expected to stay.

3/ In Antifragile, building on Mandelbrot, Taleb writes the Lindy Principle as:

"For the perishable, every additional day in its life translates into a shorter additional life expectancy.

For the nonperishable, every additional day may imply a longer life expectancy."

4/ What justifies the Lindy Principle? *

The older something is,
→ the more conditions it must have been fit for,
→ and thus the broader range of possible futures it is fit for,
→ and thus the longer it is likely to survive,

(in absence of bounds such as senescence).

5/ (*) In "Antifragile", Taleb also presents a statistical argument for the Lindy Principle. Here, I just attempt to give readers an intuitive justification.

6/ For non-perishables such as objects and ideas, the main determinant of life expectancy is the hazard rate (the chances of dying/disappearing at age X).

7/ When we observe an object's life, we can use Lindy to estimate its life expectancy or its hazard rate.

Eg, we can estimate a book's life expectancy on the bestsellers' list or we can estimate its chances of dropping off next week.

Of course, the two are negatively correlated

8/ That said, we can reason the following.

The older something is,
→ the more conditions it must have been fit for,
→ and thus the broader range of possible futures it is fit for,
→ and thus the lower its hazard rate.

9/ It follows that our estimate of an entity's hazard rate decreases as time passes by without it breaking/disappearing.

10/ The first key word is "an entity's."

A book staying months on the NYT bestsellers' list doesn't mean that all books on it are less likely to drop off next week. It just means that that book in particular is less likely to disappear.

11/ The second key word is "our estimate".

The book's hazard rate doesn't decrease over time. Its hazard rate is probably constant.

Instead, it's *our estimate* that 📉. The longer the book survives, the more reasons we have to lower our estimates of the hazard rate.

12/ One more thing: in addition to hazards, people also die for senility.

Hence, the Lindy principle doesn't apply to people.
An 80yo person cannot survive another 80 years.

Similarly, it doesn't apply to perishables; or it applies in a reduced form.

13 / (More precisely: estimations made by the Lindy Principle are influenced by intrinsic limits such as senescence or perishability.)

14/ Lindy is not just about time but also applies to other dimensions: space, cultures, uses, conditions, etc.

A few examples over the next few tweets.

15/ Continuing the example of the NYT bestseller, a book that is sold in one country only might be successful because it's a great book or bc it talks about something very relevant to that country

Once it's translated & does well in another country, the odds it's great increase.

16/ In general, the more geographically-widespread something is,
→ the more conditions it must have been fit for,
→ thus the broader range of conditions it is fit for,
→ thus the lower the estimate of its hazard rate upon entering a new geography.

17/ The same works across cultures, use conditions, and most dimensions, I suppose. (Remember the limitation of "estimates made by the Lindy Principle are subordinate to intrinsic limits such as senescence or perishability".)

18/ For example, bicycles are Lindier than cars.
→ not only they are expected to be around for longer,
→ but they can also be used in a wider range of conditions (off-road, in absence of fuel) and can be built/repaired by more people with less specialized tooling.

19/ The Lindy principle can often be used to estimate not only life expectancy, but usefulness / relevance / maintainability across a wider range of conditions / use cases / skills etc.
(again, reminder: it is probabilistic, not deterministic)

20/ Before closing this thread, two more remarks.

First, the Lindy Principle estimates an entity's hazard rate, not whether it is good or bad. You can't say, "it's Lindy therefore it's good". Mosquitoes are Lindy.

21/ Second, something Lindy is not immortal. It might disappear tomorrow.

Lindy doesn't mean it's impossible for it to happen; it just says we have reasons to believe it's less likely than it would be if it hadn't been around for so long.

22/ The Lindy Principle doesn't tell you how long something will survive.

It helps you estimate its hazard's rate or life expectancy – both of which are probabilistic.

23/ More information about the Lindy Principle in @nntaleb's Antifragile (whose reading I really recommend).
It's where I first read about the Principle.

Here, just some thoughts on the process behind it and how it can be applied to more use cases.

24/ BTW, it's called the Lindy Effect, not Principle.
Ops, apologies (where's the edit button?)

25/ A clarification on the tweet below.

Lindy applies to perishables, but only when distant from natural bounds such as senility.

Eg: a baby's survival during the first hours / days after his birth make us increase our estimate of his life expectancy

https://twitter.com/DellAnnaLuca/status/1341977203773501441?s=20