"The Drunkard's Walk"

I’d like to praise “The Drunkard’s Walk,” a book I just finished listening to on CD, written by Leonard Mlodinow, a professor at Caltech. It’s in one of the many books I’ve read in the general arena of pop academia, all of which basically have the same theme — “you don’t know what you think you know.” It’s like on huge Venn Diagram where each circle is a book, so they all overlap, many citing the same experiments, though each one focuses on a different subject. There are the ones that focus on happiness, the one that focuses on crowds, one on split-second decisions, many on economics, etc.

This one is the math one. The theme is: “you don’t know what you think you know about probability.” Our brains are bad at calculating the probability that something will happen.

I started out skeptical, as Mlodinow starts with some very elementary pop academic concepts, like his discussion about how film executives get fired even when they’ve only been in their jobs for five years, and does anyone really know how much their actions actually have an impact on whether a movie actually does well. It’s a concept I’ve heard in other similar books, namely “The Wisdom of Crowds” — that CEOs’ actions don’t always affect a company’s results in the way you think they can.

The book eventually takes us on a journey through the history of the field of probability. For each discovery, he explains the concept and uses various examples to help us understand it. Eventually, the book starts to point some concepts that I’m sure are understood only by a tiny fraction of the population, and which even very smart people don’t know about or completely ignore, and yet are crucial to hugely important situations.

For instance, Mlodinow points that that people do not have a good handle on conditional probability — meaning the probability of something happening IF something else has already happened. To take one example, in the OJ Simpson trial, OJ’s lawyer Alan Dershowitz says that while four million women in the US are battered every year by their male partners, only one in 2,500 is ultimately murdered by her partner, so OJ likely didn’t kill Nicole. Sound convincing? Yes, but it’s irrelevant. The crucial question is this: what percentage of the women battered by their husbands who end up murdered are murdered by their husbands? The answer is 90 percent.

I also enjoyed how Mlodinow pointed out errors in judgment in the financial industry. People think the market can be beaten, and certain people get lauded for doing so — like one fund manager named Bill Miller who had a fund that beat the S&P in 15 straight years. But Mlodinow points that if you assume that everyone has a 50-50 shot at beating the market, and then crunch the numbers, you find out that the chances are good that some fund manager would beat the market in 15 straight years. So it’s no big deal that this guy did it, and it’s still tough to know whether he’s a genius or just lucky. And he’s probably just lucky. I spoke to a Wharton business professor about this recently, and he said that no business professor believes that the market can be beaten. Millions of people are spending gobs of money and time trying to do it. For many, many of these people, it’s their entire job. And yet the experts in the field who have looked at the evidence are all certain that anyone who beats the market is basically just lucky. And there’s no evidence to think otherwise! It all sounds like a complete scam.

By the end of the book, when it revisits the question of firing film executives, I began to recognize the power of that example. Five years of work at a studio is only five data points. Is that enough data to judge someone? Enough to separate luck from expertise, in a field where it’s possible that no one really knows anything about what will succeed and what will fail? Probably not. When anyone at work makes a judgment based on their evaluation of another person’s results, do they ever have enough data? It’s enough to make you doubt the fairness of any workplace situation. And that’s pretty scary.

Comments

[Jacivi]

A nice little pzzule. The answer 16/22 (or 72.73%) is correct. Assume you have 100 people, 40 men and 60 women. That means 16 of the men (40% of 40) and 6 of the women (10% of 60) are obese, for a total of 22 obese people. If you randomly select one obese person, the odds are 16/22 that they are male.