If the claim is indeed correct, then it makes sense for a company whose annual profits are billions of dollars to pay its CEO a salary and bonuses of “just” a few hundred million dollars. If the claim is incorrect, then the company should instead hire one of the many other equally qualified potential CEOs, and pay them a salary of just a few hundred thousand dollars. If a company seems to be doing poorly, should the board of directors fire the CEO, or will a new CEO facing the same problems probably do just as poorly? If you are an ambitious young rising executive who wants to make a difference, in which business sector or country are you most likely to succeed? (Hint: we shall see that you’d do better to become the CEO of a perfume company than of a public utility company, and to become a CEO in the USA than in Europe or Japan.)

Every year, thousands of business CEOs do or don’t die, or do or don’t get fired, or do or don’t get sick and hospitalized for varying numbers of days, or do or don’t get distracted by the deaths of relatives of various degrees of closeness, or do or don’t transfer to other companies. In recent decades, social scientists have made much progress in analysing those natural experiments in the historical, business and sports spheres, and in one study also in the religious sphere.

Natural experiments have always been the sole or principal method of answering questions in numerous fields that are acknowledged to be sciences, such as astronomy, geology, field biology and epidemiology. Astronomers learn about stars only by observing and comparing them, never by experimentally turning them on and off. Geologists learn about glaciers also by observing and comparing them, not by experimentally melting and freezing them. Field biologists, who are usually not permitted to exterminate bird populations, have nevertheless succeeded in gaining much understanding of them by observations and comparisons; that’s how I’ve conducted all of my research on New Guinea birds for the last 45 years.

A famous natural experiment in epidemiology established that the dreaded disease cholera was water‑borne. During a London cholera epidemic in 1854, the physician John Snow observed that new cases were concentrated among people obtaining their water from one particular public pump on Broad Street. He correctly inferred that that pump’s water was infected, and he convinced the aldermen to remove the pump’s handle as a manipulative experiment. Lo and behold, the local epidemic promptly waned, confirming that John Snow’s natural experiment had yielded the correct conclusion. Hence natural experiments have had a long and honourable history, before the recent explosion of their use in the social sciences.

From this manipulative experiment, let’s now return to another natural experiment. It addresses two important questions: by how much CEOs affect company earnings, and what else affects company earnings. If we were omnipotent manipulators of people, we could test the importance of CEOs by experimentally transferring them between companies.

Suppose that CEOs affect company earnings, and that there are some good CEOs and some mediocre CEOs—just as there are good as well as mediocre violinists and football players. You’d then expect that, when we randomly transfer CEOs, good CEOs might produce high earnings at company after company, while mediocre CEOs might produce mediocre earnings at company after company. However, it’s also possible that the business world quickly weed out mediocre CEOs, but that there are also very few great CEOs: perhaps the vast majority of CEOs are neither mediocre nor great but just competent. If that were true, we would find that our random transfers of CEOs would on the average have little effect on company earnings.

Of course, we can’t experimentally transfer CEOs. But nature does transfer CEOs: they often are recruited from one company to become the CEO at a different company. How can we use those transfers to measure the effect of CEOs on company earnings? That’s a tricky problem. Even if CEOs have an effect, company earnings surely also depend on at least three other factors (“independent variables”): the year, the company, and the industry.

Some years are years of economic slowdowns affecting most companies (like the worldwide Great Depression of the 1930s), whereas other years are boom years when many companies prosper. Some companies have better products, better locations, or better organization than do other companies; those differences between companies may persist for a long time, extending over the tenures of many CEOs. (For instance, Procter & Gamble, a leading American manufacturer of detergents and other household products, has been well managed and profitable for more than a century and a half.)

Still another variable affecting company earnings is the effect of the industry to which the company belongs: some industries are more profitable than other industries. (For example, public utility companies are tightly regulated by the government and can never make a killing, whereas computer start‑ups sometimes can make a killing.)

Statisticians have developed techniques for calculating how strong is the effect of each independent variable, even when there are many different variables. For instance, from the heights of thousands of people of different genders, ages, parents, and birth years and places, you won’t be surprised to know that statisticians can calculate the average height difference between men and women, the average growth rates of young people and shrinkage rates of very old people, and the effects of genetics and nutrition on height. A standard statistical method for teasing out such simultaneous effects of several variables is termed “analysis of variance” (abbreviated ANOVA), subdivided into sequential ANOVA versus simultaneous ANOVA. There have been numerous such ANOVA analyses of company earnings, carried out by different authors, using different databases and different statistical methods.

The studies encompass dozens of industries and hundreds of thousands of companies. But there is still moderate agreement about the results: most such analyses show that the smallest effect is of the individual year, which accounts on the average for only a few percent (e.g., between 1% and 6%) of the variation in company earnings. An analysis of American companies by Alison Mackey found that CEOs accounted for 29% of variation in earnings, and the company and the industry accounted for 8% and 6%, respectively. 

These analyses have several implications. One is that CEOs do affect company earnings. That’s hardly surprising, because companies do need someone to coordinate their activities and to make decisions. If there were no CEO effect, the CEO’s job could be done by tossing coins or rolling dice. That conclusion is consistent with economic theory, which suggests that if some CEO were really incompetent and consistently made bad decisions, that CEO would soon be fired or demoted; that if some CEO were really brilliant and consistently made brilliant decisions, that CEO would soon be promoted and widely imitated; and that most CEOs observed at any one time must be just competent. Hence while CEOs do affect company earnings, many CEOs don’t differ enormously in their competence, and company and industry differences also play large or even larger roles. All three factors account for much more variation in company earnings than do annual fluctuations in the economy.

In order to understand more about when CEOs are likely to have big effects on company earnings, we need to consider the concept of “managerial discretion”—that is, the extent to which a CEO is free to make choices that affect company performance. Some industries provide much more discretion to CEOs than do other industries. For example, industries such as computers, soap and perfume, games and toys, motion pictures, and clothing for young people are characterized by rapidly changing markets, many competitors, and high uncertainty. In such industries, CEOs who make good decisions can greatly increase their companies’ earnings, and CEOs who make bad decisions can greatly decrease their companies’ earnings. In contrast, industries such as public utilities, steel, shipbuilding, railways, and gas transmission are heavily regulated, capital‑intensive, and slow‑changing. CEOs of public utilities have little discretion, because government regulations may specify the price that they can charge and the rate of return that they can earn; if they want to raise prices, they may have to get permission from a public utilities commission.

In high‑discretion industries, CEOs’ decisions have large potential effects on their companies’ earnings, and hence the CEO effect on earnings is larger. In low‑discretion industries, CEOs’ decisions have small potential effects on earnings, and hence the CEO effect is smaller. That explains why an ambitious young executive who wants to make a difference would do better to become the CEO of a perfume company than of a public utility company. But the CEO in a high-discretion industry like a perfume company also has more power to make a disastrous decision that will get him/her fired than is the CEO of a low-discretion industry like a railroad. As for national differences in CEO discretion, boards of directors and governments in Europe and Japan impose more constraints on CEOs’ decisions than do their counterparts in the USA. That reduces CEOs’ discretion and their potential effect on company earnings (and also reduces their salaries and their risk of getting fired?) in Europe and Japan compared to the USA.

Excerpted from PROFITS, PROPHETS, COACHES, AND KINGS: (When) Do Leaders Matter? By Jared Diamond. Copyright © 2026 by Jared Diamond. Published by Mariner Books, an imprint of HarperCollins Publishers. All rights reserved. Reprinted with permission.