Phil Rosenzweig on Leadership, Decisions, and Behavioral Economics
What this covers
Phil Rosenzweig, a strategy professor at IMD, joins Russ Roberts to examine how behavioral economics research translates—and often mistranslates—into real-world leadership. The conversation ranges across psychology, business, and strategy, treating the central tension between what controlled experiments reveal about human decision-making and what actually happens when leaders must act in competitive settings. Rosenzweig argues that the findings are often valid in their original laboratory context but mislead when applied wholesale to situations where outcomes can be shaped, where winning requires surpassing rivals, and where inaction carries its own heavy cost.
The discussion unpacks several contested distinctions. One is the word "overconfidence," which Rosenzweig breaks into three separate phenomena—overprecision, overestimation, and overplacement—each with different evidence and different implications. Another is the Winner's Curse, which applies cleanly to common value auctions but not to private value auctions where your ability to extract future value differs from your competitors' and elevated bids reflect rational adaptation. A third separates decisions you can shape from those you cannot: most research deliberately studies uncontrollable outcomes like weather or election results to get clean data, but real leadership is largely about improving things within your power. Throughout, Rosenzweig emphasizes that what looks like bias in hindsight—elevated confidence, selective transparency, willingness to act despite uncertain odds—may in competitive contexts be essential rather than excessive. The practical upshot is reconsidering when behavioral findings illuminate and when they obscure.
Rosenzweig argues that behavioral economics findings about decision errors like overconfidence and the winner's curse, while valid in controlled experiments, mislead when applied to real-world leadership decisions where actors can shape outcomes, face competition, and must act to win.
- Lab decisions involve uncontrollable, uncared-about variables; real leadership decisions can be actively influenced
- Winning requires acting and surpassing rivals, so elevated confidence can be essential rather than a bias
- The word 'overconfidence' conflates three distinct phenomena (overprecision, overestimation, overplacement) with very different evidence
Elevated confidence is necessary for winning in competitive settings, not a cognitive error to eliminate.
- Because winning decisions require doing better than others and require mobilizing followers, a good leader must selectively convey information—conveying confidence that the team can achieve the unprecedented—since brutal transparency about poor odds makes failure more likely; what looks like deception by one definition may be one of the highest traits of leadership.
“if you are brutally honest with everybody and you tell them, 'Gosh, the odds are against us; this has never been done before. I'm not sure we can do it,' you very likely won't do it.”
- In competitive settings—business, sports, military, politics—the one who wins probably had a level of confidence that surpassed others, so a somewhat elevated level of confidence is not merely useful but probably essential, and the main danger is the Type II error of failing to act.
“the one who wins probably had a level of confidence that surpassed others. And in that sense, I would say that somewhat elevated level of confidence is not only useful--it's probably essential.”
- When you must actually do something rather than merely predict an outcome someone else controls, holding a level of confidence higher than your past achievements can lead you to perform better—so labeling it 'overconfidence' is misleading because if it improves your outcome it is not truly excessive.
“when it comes to things that we actually have to do, where we're not just predicting something that somebody else will do, but we have to do it, to have a level of confidence that is somewhat elevated, somewhat optimistic, can, for many people, actually lead for them to do better.”
- If the lesson drawn from the decision literature is that the way to avoid errors like confirmation bias and overconfidence is to not take action, that is unhelpful and even harmful, because the real danger in competitive settings is the Type II error of failing to act (an error of omission); only by acting can you win and can you take steps to improve your chances.
“if what we have learned from a lot of this decision literature is, [?] the way to avoid errors is not to do stuff, then now you've got another set of problems.”
- The winner's curse is not a cognitive bias or individual estimation error but an outcome of a bidding process: in a distribution of bids, even if everyone bids conservatively on average, the highest bidder—who is wildest on the high side—wins the auction and therefore probably overpaid (and conversely, the lowest contractor bid wins but probably won't make money).
“it's not a cognitive bias. It's not where any individual has made an error in estimation... It's actually the outcome of a process.”
Overconfidence research conflates three distinct constructs and misleads when generalized beyond controlled experimental settings.
- Behavioral economics findings from controlled experiments are valid for certain decisions but mislead when generalized to real-world leadership settings, because in real life you can change the available options, improve them, and face a competitive dimension—none of which exist in experiments where you simply select among fixed options.
“when managers, when leaders are making decisions in real-world settings, those kinds of controls don't exist. You are not just making, you are not selecting options that are presented in front of you; you can actually change the options.”
- There is robust evidence that people are over-precise much of the time (a stated 99% certainty isn't really 99%), but evidence for overestimation and overplacement is far weaker—people frequently underestimate their abilities, and overplacement findings can be manipulated by the questions and survey design used.
“regarding overprecision--yes, there is very robust evidence that most people are over-precise much of the time... The evidence on estimation and placement is nowhere near as powerful.”
- The single word 'overconfidence' has been used to mean three very different things—overprecision (being too certain in stated ranges), overestimation (thinking you can do more than you can), and overplacement (thinking you rank higher than others)—and researchers improperly use evidence of one to claim another.
“this one word has been used to mean three very different things... Overestimation, overprecision, and overplacement... we kind of go back and forth among them using evidence of one to claim another.”
- The fundamental problem is not the basic research but generalizing a lesson from a contrived experimental setting to a very different real-world setting without recognizing the difference; the appropriate stance is 'Yes, But'—the findings make sense for these circumstances but should not be applied to those.
“The problem is when you take a lesson from a contrived experimental setting and then you then generalize that to a setting that is very, very different. Be careful.”
People make reasonable judgments based on limited information, not suffer from a universal bias toward overplacement.
- Rather than being uniformly overconfident or underconfident, people are myopic: they see clearly what is close to them (their own abilities) and have less information about everyone else, so they make reasonable but skewed inferences—judging themselves above average where they are skilled (driving) and below average where they know experts exist (drawing).
“the word that I like better... He says people are myopic. What you do is you see very clearly what's close to you, which is yourself.”
- Whether someone is 'above average' depends entirely on the reference set: a right-fielder with a better-than-average arm relative to all baseball players need not have a better-than-average arm for a right-fielder, since right field selects for strong throwers.
“They have a better than average arm compared to most outfielders; that's why they are in right field. But it doesn't follow that they have a better-than-average arm for a right-fielder.”
- If humans had a consistent propensity for overplacement, they would rate themselves above average at both driving and drawing, but in surveys most people rate themselves below average at drawing—demonstrating the survey-design dependence of overplacement rather than a universal bias.
“the word that I like better... He says people are myopic. What you do is you see very clearly what's close to you, which is yourself.”
- In everyday speech, 'overconfidence' is applied retrospectively only when something goes badly—ex ante nobody thinks they are overconfident, and if things go well people say they had an appropriate level of confidence—making the term an unreliable post-hoc attribution.
“We tend to apply it retrospectively when something has gone badly. So, 'Gee, this didn't work out. Well, I guess I was overconfident.' Ex ante, nobody thinks they are overconfident.”
Entrepreneurship and new business formation reflect rational adaptation and learning, not evidence of cognitive bias.
- It is implausible that founding new businesses reflects a harmful cognitive bias, because no one in society laments that we start too many new businesses; on the contrary, the entrepreneurial climate is widely regarded as a strength other countries want to emulate, so the 'failures prove bias' argument fails even at the level of face validity.
“Nobody ever says 'A great problem in American society is we start too many new businesses.' Quite the contrary.”
- Contrary to the romantic ideal of entrepreneurs as wild-eyed dreamers who take big leaps despite skepticism, success comes not so much from taking risk as from managing it—distinguishing technical risk (within your power and your value-add), market risk (external and uncontrollable, so don't depend on consumers buying what they never have), and financial risk (how fast you burn resources, with a crossover point requiring caution early and bold commitment when the product succeeds).
“It's not so much taking risk as dealing with risk.”
- Successful ventures often succeed not at their original product or market because they adapt and pivot using an effectual approach—starting with available means and resources and combining them toward an interesting end—rather than a causal approach that fixes the end and executes a business plan, and a key ingredient of success is abandoning the original plan quickly.
“A causal approach is when you have the end in mind and you know what to do to get to the end. An effectual approach says let's start with our means, let's start with our resources, and let's see how best to combine them for an interesting end.”
- The well-documented statistic that half of new businesses are gone after 5 years and 80% after 7 does not imply that starting them was a mistake or that founders suffered a cognitive error like neglecting base rates, because many founders limit their losses to what they were willing to lose (affordable loss), learn, and would do it again.
“it's true that a lot of these businesses don't persist--they don't survive. But many of them find ways to limit their costs and essentially only lose as much as they were willing to. And then they can learn from that and start again.”
Research studying uncontrollable outcomes misses how real decisions involve shaping outcomes through action and improvement.
- Much decision research deliberately studies evaluations of things people cannot shape—weather, sports outcomes, elections, prices—because that yields clean objective data, but real life is mostly about doing things we can actively improve, so judgments about uncontrollable phenomena are a minority of what we actually do.
“There's a reason why scientists like to ask you to choose from a set of options that you cannot change, with certain parameters that you cannot alter, or make an estimate of something that you are not able to influence.”
- The winner's curse lesson applies to common value auctions (where the item is worth the same to all), but in private value auctions—like an oil tract whose value depends on your future extraction capabilities over a 10-20 year life—you may rationally pay more, and in competition you must bid somewhat beyond your present capabilities (trusting your historic rate of improvement) or you will never win.
“if I'm not willing to bid somewhere beyond my present capabilities, I'll never win the tract. So you almost have to bid beyond today's capabilities.”
- Deliberate practice is extraordinarily powerful for developing expertise only in sequential activities of short duration with concrete, rapid feedback (shooting baskets, hitting golf balls, surgery); it does not apply to big strategic, military, political, or business decisions where feedback is slow and confounded by intervening events, making the 'try, learn, try again' model illusory for such decisions.
“every one of those examples is an example of a sequential activity, typically not that long in duration for which you could get concrete feedback and try again. That doesn't apply very well, in, let's say, the business world to decisions where you don't get rapid feedback.”
- The intellectual arc runs from a post-war economic thesis assuming humans are fundamentally rational, to a behavioral antithesis showing people make predictable errors that diverge from rational theory, toward a synthesis that recognizes there is not one kind of decision but many—some shapeable, some not, some competitive, some repeated—so making the best decision requires understanding what kind of decision you are making.
“the thesis that has been so strong in at least post-War economics has been to assume that humans are fundamentally rational... The antithesis... has been to point out that actually people make some fairly predictable errors... now... move from the antithesis to a synthesis.”
Individual confidence should match the type of mental state that produces better performance for that specific person.
- Shaw's claim that all progress depends on the unreasonable man who adapts the world to himself is only partly right, because the greatest catastrophes are also due to unreasonable people who inflict grandiose schemes on others, and much success comes instead from being nimble and adapting.
“You could also say that the greatest catastrophes are due to unreasonable men and women who inflict their grandiose ideas and hard-brained schemes, and other kinds of destructions, on fellow people.”
- Obama's '50-50' framing of the bin Laden intelligence reflected the principle that refining a probability estimate from 45% to 60% is not time well spent; the better use of time is improving the performance and execution of the mission rather than achieving more precise odds.
“that's an example of somebody who says, 'Let's use our time not to assess the odds but to improve the performance that we are going to deliver.'”
- There is no single mental state that produces good results; you should understand whether you are the kind of person who performs better when optimistic about your prospects or the kind who performs better when motivated by fear of failure, because what matters is how your mindset translates into action.
“it's not having a certain thing in your mind that leads to results. It's how what's in your mind either does or does not translate into your actions.”
- During Apollo 13, Gene Kranz refused to discuss the slim odds of a safe return with his team, insisting 'the crew is coming back,' because conveying that the odds didn't matter—while relentlessly and systematically working to reduce the odds of failure—was what made a successful outcome possible.
“he says, 'I think they may have. In any event, I wouldn't have given them because I said to my team: The crew is coming back.'”
Leaders face asymmetric judgment that rewards persistence over adaptability, obscuring whether success came from steadfastness or luck.
- Leaders are judged asymmetrically: one who persists and fails still gets credit for persistence, while one who changes course and wins is often dismissed as merely lucky rather than adaptable—reflecting society's expectation that leaders be steadfast.
“If a leader persists, persists, persists, and always fails, they don't get credit for succeeding but they get credit for persistence. But if you change, change, change and win, they usually say, 'You were lucky.'”
- As careers progress, poor performers are selected out while continuing performers receive repeated reinforcement that they are good, so some apparently good decision-makers may merely be lucky survivors who have come to believe their own success narrative.
“as people's careers go forward, some people don't do well; they are selected out. And those who continue to do well, they keep getting reinforcement that says, 'You're good. You're good. You're very good.'”