Hacker News Reader

> It really bothers me that people write news articles about the impact of this "research".

News articles about science, even good science are often terrible. The most common problems I see are oversimplifying, overgeneralizing and overstating conclusions. Many readers then further generalize and simplify what they read in the article.

I just saw this recently with an article about a study that concluded (and I'm oversimplifying here) that NSAIDs are a better first-line treatment for arthritis pain than opioids. By the time a layperson was referencing it in conversation, they had concluded that it's never a good idea to use opioids to treat any pain.

I'd love to see your model. Criticizing a model is relatively easy, coming up with something that describes the reality (in this case, Pareto distribution of wealth) yet better is much harder.

> As opposed to what? What kind of distribution might you expect?

For one thing, they might have expected the result not to be dominated by luck.

> It bears literally no relationship to reality, where random massive doublings of wealth are the exception, not the rule.

I don't think this will help to control the luck as opposed to talent. I suspect the problem is in the "leverage". Having more wealth makes it easier to accrue more wealth in the future, however, the same is not true for talent. Having more talent doesn't help you get much more talent.

So eventually, regardless how big your talent can be, the accrued wealth will dominate the end of the distribution. The rate at which the wealth is accrued doesn't matter, it will only reduce social mobility.

Addendum: Interestingly, the model actually shows what we would expect for low to middle end of income distribution, that talent there is more important than wealth. Which is in line with my explanation.

I don't think that's how it works. A model doesn't become a good model simply because it's hard to or nobody has bothered to make a better one.

> A model doesn't become a good model simply because

I agree, and

> nobody has bothered to make a better one

So bother! I have made 4 comments here and all of them ask the same question. What would you improve and how do you think it would affect the outcome?

Sometimes the simple model is (almost) as good as any in answering a simple question. For example, some people do not accept climate models (and probably never will), ostensibly for reasons that they are not good. I see the opposite - the models have gotten much better and much more realistic in the past 60 years. Yet, the same basic outcome (Earth will warm roughly 3 degrees by doubling CO2) was foretold by very crude models that even don't have the circulation.

R/economics went over this a bit too. There are other issues with the model that and to make it very poor:

* - fixed rate of return (doubled or halved) make luck play an oversized role. One of talents greatest contributions is knowing how to let your winners run and cut your losses. The setup is close to a random walk by definition.

- risk reduction is hugely important. High earners especially when speaking of finance tend to know how to reduce risk. Only fools always bet half they capital in every decision.

* - talent would understand what events are more likely to succeed than just every event having the same probability.

- talent and preparation generate events. The more you prepare the more chances you'll have available. Most successful people have failed multiple times but keep trying.

- no learning. One of talents great benefits is learning from mistakes. And when combined with increased number of chances that talent provides, this is a critical feedback cycle the model misses.

- competition. Businesses aren't competing against the impersonal universe, but against other individuals. The trial events should have occurred in that context. If they really wanted to keep this silly 2D thing, it would have been better to have two agents roll against each other when meeting, taking talent into consideration.

Basically the model made everybody similar and then had an exponential random function. With such small tails in the talent pool, there is no other decision it could have reached. I'm shocked this is being upheld as valid research.

* Edit: The response by @edmccard below says that a couple of these might also be luck based (those related to evaluating events), and that is a really valid point. We know that stock pickers are mostly luck for example. It would seem plausible this also extends to evaluating what job to enter or what company to start.

Still not possible to easily type asterisks on HN without them getting interpreted as italics

>...One of talents greatest contributions is knowing how to let your winners run and cut your losses...

>...talent would understand what events are more likely to succeed...

If we already know that these things are the result of talent and not luck, and if it's a foregone conclusion that the model gives an "oversized role" to luck, then I guess these kinds of investigation are doomed from the start.

My status on this is that the model could be a good model, but that hasn't been established. You cannot reason from "the outcomes look like reality" to "the parameters and mechanisms in the model are the same as the parameters and mechanisms in reality." That is backwards.

One thing you can say is, "IF reality's parameters and mechanisms are like this model's, you would expect the outcome that we in fact observe in reality." This is evidence, if not particularly strong evidence. There are infinitely many models that lead to the similar outcomes that we observe in reality, and an even larger infinitely many models that don't.

Their model is if you have a lucky event you have P(talent) chance of doubling your capital and if you hit an unlucky event your capital is halved. Any step in time where you neither experience luck or unluck you are left unchanged. It seems wrong on face for several reasons:

- You need to be "lucky" in order to ever increase your wealth, which makes the "finding" that luck is more important than talent completely obvious from the way the model was designed

- Capital only changes by multiplication and division, so the fact that this model "recreated" the power distribution we see in wealth in the world is obvious from the setup. It was designed to be a power distribution and couldn't be anything else

- Talent has no effect on your life during unlucky or neutral periods in their model. This obviously fails to conform to the real world.

tldr; they designed a power-distributed model where luck was clearly more important than talent, and "discovered" that it showed a power distribution where luck is more important than talent.

How would you change the model to be more realistic and how you think would your changes affect the outcome?

This paper reminds me of an article I read yesterday about how we are no longer in the information age, but rather the "reputation" age. Goes to show, even science that might seem reputable to the average layperson can be complete garbage.

The model shows that a mix of luck and merit can lead to highly unequal outcomes in which success is only tenuously linked to merit. There are other models yielding similar results, as for example outlined in Robert Frank's enjoyable book Success and Luck: Good Fortune and the Myth of Meritocracy. I don't find the mechanism entirely implausible, but that's a longer discussion.

These models are an important counterpoint to naive "Economism" [1] models (where income = marginal product, thus just and justified and necessary for efficiency), and thus relevant for an informed policy discussion.

[1] See James Kwak's book of that title.