Friends, the problem of statistics is that even if they worked (they don't), whatever is derived statistically ONLY applies to groups never to single individuals. You are not hiring a subcontinent, but a single individual. Variance determines the max.

Oleg Okunev, my mentor/teacher from college, and fellow flâneur, used to say that “even if a paper or a doctor says that potatoes are good for you, it still doesn’t say anything about what the potato that I ate yesterday did to me.”

The problem isn’t the statistics per se. It’s trying to predict a complex individual-level outcome from a single, noisy proxy such as IQ. Even a statistically significant group level association may have poor individual predictive power.

The problem is that when you hire a single individual from a certain subcontinent you will, in fact, end up bringing in the entire subcontinent.

This is one of my eternal aggravations with medical advice. There is asymmetry in the use of data. They tell you to adhere to treatments which are claimed to marginally benefit large numbers of pts but if your results deviate they say it’s something else, unknown that caused it.

“You are not hiring a subcontinent, but a single individual.” - @nntaleb Great way to phrase it

Statistics applies to an individual exactly to the degree that the individual can be ergodized

There is a social version of the same statistical error: success is individualized, failure is collectivized. We praise the winner as an individual, then use the loser as evidence about his group. Same fallacy, different direction.

yes but it matters if you let an entire group lets say 1 million into your country.

... because the properties of the group are an emerging phenomenon, I guess

why cant we mathematically go from group to individual? My guess: If pop regression is known perfectly, Var(Y_indicidual | X_individual) ~ (1-R^2)*Var(Y) + measurement noise from Y and X, hence R^2 must be large (+measurement noise small) to have gained any insight?

IQ is a c** measure that predicts c**. If you want kids to be successful, free them from compulsory schooling. Coercive learning steals intelligence!

Most of life is working with unknown individuals or individuals you know very little. There’s some good books on operating under uncertainty

Sir Taleb - what books would you suggest to a 21 year old who’s trying to understand the nuances of risks. Thanks. Big fan of skin in the game.

This is good. A criticism I have with what you're saying with IQ is you're arguing about using only IQ. Take a dozen weak signals and you'll get a decent one. A lot of that variance comes from other signals that you're not using

“There was this retard Jeffrey” Vintage taleb 😂

I used to find it amusing how few statisticians truly understand this, let alone everyone else who uses statistics...

We’ve been hiring subcontinents

This is so accurate. One of the key challenges i find is to convince stakeholders that doing a very small sample analysis doesn’t yield anything about the population.

Thanks for the video. I enjoy hearing your explanations vocally.

Exactly! Statistics is a tool to see a kind of trend/tendency about a group like a society, not about individuals. Making a group data from individuals' history is possible, though. It can show trend/tendency about the individual, but not about themselves at particluar moment.

May I suggest you check what this 2018 paper actually claimed before deciding the authors are fools. Scatterplot below Figure 3 is not from the paper... I think a serious commenter would have made this clear to the viewers.

So true... yet people love extrapolating totally meaningless data

statistics is very hard because you have to understand your own assumptions. kinda higher-order-thinking most don't have.

Absolutely!! Bro. Statisticians commit same blunder while posting stats or findings to fit things in their narrative while ignoring individual variations.

He's right. Western institutions use group statistics to justify prejudice against entire populations.

Wait ... you're saying Lindy doesn't apply and I'm not good for 208 ??? 🤯🤯🤯🤯🤯

No one is the average. Important reminder.

Anscombe cuartet`??

false reversion to the mean

Just looking at that distribution should be reason enough not to draw any statistical conclusions

I've argued with game developers for years about this. In the ARPG genre for ex they go "the item drop chance is 10%"... yea if I had thousands of tries or more. Most players play for half an hour a day, that 10% is frequently experienced as 0.0001% by people.

Well said as expected of the legend

Average of a group is a hiring superstition. You’re betting on one person’s variance, not a country’s mean.

Ótima explicação

why are lower MCAT scores acceptable for medical school acceptance (for certain groups?)