In 2004 three researchers at Columbia built a website on which visitors could listen to forty-eight songs by unknown bands, rate them, and download the ones they liked. More than fourteen thousand people took part, each assigned at random to one of nine worlds. In one world, each visitor saw only the songs; in the other eight, visitors also saw how many times each song had already been downloaded by the people before them in that world. The eight social worlds started identical and were sealed from one another.
The results in the eight social worlds diverged. A song called "Lockdown," by a band called 52metro, finished first in one world and fortieth in another. In the social worlds, the gap between the most and least downloaded songs was far wider than in the world without social information. Which songs ended up at the top was also far less predictable. Quality still counted for something: the songs that visitors in the independent world rated highest rarely did badly in any social world, and the ones rated lowest rarely did well. But for the songs in between, which were most of them, an early run of downloads that happened to land on one song rather than another was enough to carry it to the top of its world and leave an equally good song near the bottom of the next.
Nobody who looked at the final chart in a single world could have told which part of a song's position was the song and which part was the order in which the first few hundred visitors happened to arrive. Chapter 3 was about that limit. But the experiment shows something else as well: the early downloads, however they came, were real. In the world where "Lockdown" led, it kept leading, because every new visitor saw it leading. An accident at the start had become a position that kept the song on top from then on.
The Mechanism Has a Name
The sociologist Robert Merton described the pattern among scientists in 1968 and borrowed its name from the Gospel of Matthew (unto every one that hath shall be given). A discovery made jointly by a famous scientist and an unknown one is credited to the famous one. The credit brings the famous scientist more resources and more attention; those produce more discoveries, and the discoveries bring more credit. Merton called it the Matthew effect. The general term is cumulative advantage: a process in which an early difference in outcomes, from whatever cause, increases the probability of later favorable outcomes.
Stated as a mechanism, in the manner chapter 1 asked for, cumulative advantage runs through four channels, each of which can be observed without knowing how the story ends. The first is resources: an early success brings money, and money buys the next attempt (a second fund, a second product, a larger lab). The second is attention. An early success makes the actor visible, and visibility lowers the cost of every later transaction; customers call instead of being called. The third, and I think the most important of the four and the hardest to see, is access. An early success changes the set of opportunities the actor is offered. Better deals come to better-known investors, better students to better-known professors, and harder problems to the engineer whose last project worked. The actor's later record is then drawn from a better pool, and a better pool produces better outcomes with no change in the actor at all. The fourth channel runs through the actor's rivals: those who lose early often stop trying, and the field thins.
None of the four requires the early success to have been earned; all four require it to have happened. That is the sense in which luck becomes an asset: the position that an early outcome creates passes the two tests of chapter 2. It improves expected results through the four channels, and a rival cannot easily acquire it, because the only way to get it is to have had the early outcome.
Luck Given at Random
The music experiment showed cumulative advantage inside an artificial market. Two later studies looked for it in real ones, using the cleanest method for separating the position from the person: giving the early success away at random, or finding a setting where something close to chance gave it away.
Arnout van de Rijt and his colleagues, in 2014, chose four online settings where success is visible and accumulates. They donated to randomly selected projects on a crowdfunding site, gave randomly selected editors on a collaborative encyclopedia a status award, rated randomly selected product reviews favorably, and signed randomly selected petitions. In every setting the recipients of the random early success went on to attract more subsequent success than the matched controls who received nothing. Although the gifts were small and the recipients did not know they were arbitrary, they worked. The experiment also found a limit: a larger initial boost did not produce a proportionately larger later gain. The second helping did less than the first.
The second study used a natural experiment in science funding, made possible because the Dutch research council awards early-career grants on a ranked score with a fixed cutoff. Thomas Bol, Mathijs de Vaan and van de Rijt compared applicants who scored just above the line with those who scored just below it, two groups whose proposals were, by construction, almost indistinguishable in quality. Over the following eight years the winners accumulated more than twice as much research funding as the near-miss losers. Part of the gap came from evaluators, who looked more favorably on later applications from people with a grant on their record. Part came from the applicants, since a substantial share of the near-miss losers stopped applying for the next grant at all.
Because the experimenters assigned the early boost in the first study, and a cutoff assigned it among nearly equal proposals in the second, the later advantage cannot have been a hidden signal of ability in either.
Venture capital, where I think confusing the position with the person is more expensive than in any other business, shows all four channels at once.
Consider a two-partner firm whose first fund, through a combination of judgment and the particular year in which it was raised, contains one company that returns the whole fund several times over. (This firm is hypothetical, assembled from how the industry works rather than from any firm's record.) The second fund is larger and raised faster, because limited partners, the investors in the fund, allocate on records. The partners are now invited to see companies before other investors do, because founders who can choose their investors choose the ones with a famous hit. When the firm's name is on a round, the company finds its next round easier to raise and so is more likely to succeed, and its success adds to the firm's record. Rival firms that missed the hit lose partners and sometimes fold. All four channels are now running, and the partners' second fund is drawn from a better pool of companies than their first.
How much of a venture firm's apparent skill is that better pool, and how much is what the firm does once a deal is in hand? Morten Sorensen, in a 2007 paper in the Journal of Finance, set out to measure it. His estimate, from a structural model of how investors and companies match, is that about two thirds of the persistence in top-quartile venture returns comes from better firms getting access to better companies, and about one third from the firms' influence on the companies they fund. The split depends on the model's assumptions, so I would not repeat it as a law. Still, if the estimate is even roughly right, most of what a limited partner is paying for when it pays for a top firm's record is the firm's position. The position is a different asset from the partners' judgment. The position may attach to the franchise rather than to the people, in which case it need not survive partner departures, successions and increases in fund size, precisely the occasions on which a limited partner decides whether to reinvest.
The founder evidence has the same shape: Paul Gompers and his colleagues found, in a 2010 study, that founders with a prior venture-backed success were about 30 percent likely to succeed again, against 18 percent for first-time founders. So an investor who backs the serial founder is making a better bet. But skill is only one of the channels the paper identifies. A founder with a prior success attracts better co-founders, better early employees, and better investors, and is given more time by all of them. Part of the second success comes through the position the first one created, whether or not the first one reflected the founder's skill.
Three Questions, Not One
People usually ask about an early success as if it raised one question, when it actually raises three, with different answers.
The first is origin: did the actor earn the early success? Chapter 3 argued that in most consequential domains the record cannot answer this. Even where it could, the answer would not settle the other two.
The second is current effect: does the position created by the early success improve expected results now? This one can be answered, the same way chapter 2 tests any alpha asset, by stating the mechanism and checking it against observations that do not depend on the outcome. Whether the venture firm sees deals before its rivals can be checked by counting the inbound. Does the professor's name on a grant application raise its score? The Dutch study measured it. Whether the salesperson from chapter 1 has a territory full of customers whose equipment is due for replacement is in the service records. These are facts about the present.
The third is portability: does the advantage belong to the person or to the position? The cleanest evidence is Boris Groysberg's study of star securities analysts, the ones ranked at the top of their sector by institutional investors. When a star moved to a different firm, their ranked performance usually fell and often stayed down for years. The fall was smaller when they moved with their team and when the new firm had capabilities comparable to the old one. Much of what had looked like the analyst's skill was the firm's research platform, its relationships with the companies they covered, and the colleagues who fed them. The moves were not random, so the study does not give a universal penalty for changing employers. It does show that what travels is a question with an empirical answer. I suspect the answer is often less than the star believes, and less than the firm that hires them is paying for.
The same three questions apply to the salesperson with the inherited territory. Whether they earned it is unknowable and, for most purposes, beside the point. Whether the territory is an advantage now is a matter of record. Their next employer should ask whether the territory goes with them before matching their number. So should they, before they assume that the number is theirs.
My Own Case
In 1995 Steven Eppinger and I published a textbook on product design and development, which engineering and business schools adopted in numbers that surprised us. Neither of us could have predicted those numbers from its contents, a clear organization of what good product development teams already did. What followed ran through the four channels. The book made me visible to people I had never met. Executive education programs asked for me because their participants had read it, and companies asked me to consult. Doctoral students chose to work with me, and the ones who chose me were, on average, stronger than the ones I would have been able to recruit without the book. Each of those brought the next project, and the next project was drawn from a better pool than the one before.
I can state the mechanism and point to observations that do not depend on the outcome: the adoption numbers, the invitations, the composition of the students who came. What I cannot do is separate the book's success from the fact that it appeared when it did, in a field that was about to grow, at a school with a name that opened doors. A different book of equal quality, published three years later or from a less visible address, might have sold a tenth as many copies, and I would have been a different professor. I believe the book was good, though I do not know whether its reception was a measure of how good. I do know that most of what the reception brought me was position rather than skill. The position was real, and I used it.
Then there is portability. Some of what the book brought traveled with me (the students, the method, the relationships with the companies), and some stayed with the address. When I left the university's administration and the daily company of its colleagues for a company I had co-founded, the invitations did not stop, but they changed in kind. Some of what had looked like my reputation turned out to have been the institution's. I had been warned, by Groysberg's analysts among others, and the warning was accurate.
A board, an investor, or a hiring manager who cannot know whether a record was earned can still learn whether the position the record created is an asset now, and can find out whether it travels. Pay for position when the position comes with the deal; pay for the person only when there is evidence that the person is what you are buying. The limited partner paying premium terms for a franchise's deal flow may be making a sound investment. It is a different investment from paying for the partners' judgment, with a different risk.
And what should the actor do with an early success? Others will read it as skill, and so will the actor. Chapter 3 gave the reasons that reading is mostly unwarranted. The actor's advantage is the position the success created, and positions decay. The deals keep coming only while the record is fresh, the recruits keep choosing the professor only while students keep hearing the name, and the territory keeps yielding only while the customers keep needing the equipment. So the useful question is what to build while the position lasts that will remain after it fades. The venture firm converts a hit into a platform of relationships and a brand that outlast the partners who made the hit. The professor converts a book into students who carry the method. The salesperson converts an inherited territory into expertise in the customers' problems, which is theirs.
That conversion is the subject of the next chapter. The mechanisms by which advantage accumulates, which I call flywheels, work the same way whatever produced the first push. A customer base grows from the customers a firm already has, whatever brought them; learning accumulates with volume, whatever produced the volume; and a brand strengthens with every satisfied customer, whether the first customers came from merit or from a famous name on the label. Although in most cases nobody knows the answer to the origin question, for the flywheels it makes no difference.
Notes
More than fourteen thousand people took part Matthew J. Salganik, Peter Sheridan Dodds, and Duncan J. Watts, "Experimental Study of Inequality and Unpredictability in an Artificial Cultural Market," Science 311, no. 5762 (2006): 854–856. The experiment used 14,341 participants, 48 songs, one independent world and eight social-influence worlds; the authors report that social influence increased both the inequality and the unpredictability of success, with the best songs rarely doing poorly and the worst rarely doing well.
Merton called it the Matthew effect Robert K. Merton, "The Matthew Effect in Science," Science 159, no. 3810 (1968): 56–63.
giving the early success away at random Arnout van de Rijt, Soong Moon Kang, Michael Restivo, and Akshay Patil, "Field Experiments of Success-Breeds-Success Dynamics," Proceedings of the National Academy of Sciences 111, no. 19 (2014): 6934–6939. The four settings were a crowdfunding site, a collaborative encyclopedia, a product-review site, and a petition site; initial successes were randomly assigned, recipients attracted more subsequent success than controls, and larger initial endowments did not yield proportionately larger gains.
more than twice as much research funding Thomas Bol, Mathijs de Vaan, and Arnout van de Rijt, "The Matthew Effect in Science Funding," Proceedings of the National Academy of Sciences 115, no. 19 (2018): 4887–4890. The study compares early-career applicants to the Netherlands Organisation for Scientific Research just above and just below the funding threshold, over the following eight years, and attributes the gap to both evaluators' later judgments and the losers' reduced application rates.
about two thirds of the persistence in top-quartile venture returns Morten Sorensen, "How Smart Is Smart Money? A Two-Sided Matching Model of Venture Capital," Journal of Finance 62, no. 6 (2007): 2725–2762. The allocation between sorting and influence is a structural estimate and depends on the matching model's assumptions.
about 30 percent likely to succeed again, against 18 percent Paul Gompers, Anna Kovner, Josh Lerner, and David Scharfstein, "Performance Persistence in Entrepreneurship," Journal of Financial Economics 96, no. 1 (2010): 18–32.
Boris Groysberg's study of star securities analysts Boris Groysberg, Linda-Eling Lee, and Ashish Nanda, "Can They Take It With Them? The Portability of Star Knowledge Workers' Performance," Management Science 54, no. 7 (2008): 1213–1230. Analysts ranked by Institutional Investor; performance generally declined after a move, less so when the analyst moved with colleagues or to a firm with comparable capabilities. Moves were not randomly assigned.