Common Knowledge and the Super Bowl Ad
Everyone knowing that everyone knows is the favourite explanation for why a Super Bowl slot costs what it does. The evidence underneath it is one unpublished regression, two field studies that measure something else, and a dispute its own authors have not settled.
Answer to the question
Do Super Bowl ads work by creating common knowledge?
Partly, and nobody has shown it. The common knowledge account rests on a 1998 working paper covering 119 brands on 1988 and 1989 network television, where brands selling publicly consumed goods paid about 20 per cent more per thousand households. Its author called the result more suggestive than definitive, and said it might be a beer-drinker effect.
Key takeaways
- The regression usually credited to Rational Ritual is a January 1998 working paper on 119 brands and 357 monthly campaign observations, never peer reviewed, with no Super Bowl data in it.
- Its outcome is what advertisers paid, not what buyers did: social brands paid about 20 per cent more per thousand households once more than forty demographic categories were held still.
- Thirteen of the 18 social brands were beers, and the author wrote that the result might be "simply describing a beer drinker effect".
- The one peer-reviewed paper to test a consumption mechanism on Super Bowl beer and soda sales, led by Wesley Hartmann in 2018, rejected it and landed on learned association with sport.
The strongest evidence that a Super Bowl advertisement works by manufacturing common knowledge is a regression on 119 brands of American network television from 1988 and 1989. It never measured a sale. Its author called the result more suggestive than definitive, and wrote in print that it might be describing beer drinkers rather than anything social. We think the idea deserves to be taken seriously. We also think nobody has settled it, and that saying so is the finding rather than a failure to reach one.
Start with the sentence, not the slogan
Steven Pinker's When Everyone Knows That Everyone Knows..., published by Allen Lane in 2025, puts the claim at page 18. "The special role of the Super Bowl ad as an instant common-knowledge creator was explained by the political scientist Michael Chwe in his 2001 book Rational Ritual." Page numbers here are from that British paperback, because the American edition is set differently.
Then the book does something trade books rarely bother with. Across pages 18 and 19 it reports the method. Chwe used regression, a technique for holding other differences still while one of them is allowed to vary, to control for audience size and for the ways brands differ from one another. Companies selling publicly consumed products, cars, beer, sodas, films, clothing and shoes, were readier to advertise, readier to buy popular shows rather than niche ones, and willing to pay more per viewer than companies selling things used in private, such as batteries, motor oil and breakfast cereal.
Crediting a named researcher for an empirical result and stating his method is more than most popular accounts of common knowledge manage. It also carries a complication that travels badly. Rational Ritual is a monograph and contains no new experiment of its own. The regression lives in a separate working paper of January 1998, circulated from the economics department at the University of Chicago, and a working paper is a draft a researcher puts into circulation before any journal has vetted it. Whoever quotes the finding inherits that status whether they mention it or not.
Notice where the book narrows the term
Common knowledge in the strict sense is a formal object. Robert Aumann proved theorems about it in 1976, and the object those proofs concern is endless: I know it, you know it, I know that you know it, you know that I know that you know it, with no final step. Very little in ordinary life obviously contains a regress of that shape.
At pages 56 and 57 the book notices the problem and changes its subject. It adopts common p-belief, meaning highly confident common belief rather than certainty, and declares that from there on, in any real-life context, common knowledge will stand for that weaker thing. The move is honest, it is stated plainly, and it has to travel with every later citation of the book. After page 57 this is not a book about the object Aumann's theorems are about.
A second rule follows from the author's own concessions. Chapters 2 and 3 establish common knowledge as a formal object; chapter 4 asks whether human minds hold it. At page 44 Pinker writes that the proofs "depend on idealizations of reasoners which are less than fully realistic" and that the findings "may not literally apply to particular situations in everyday life". At page 102 he adds that the games "can't be taken too literally". We therefore take chapter 4 for anything empirical and neither of the two chapters before it, which is the book's own instruction read as an instruction.
Hear both sides at full strength
The book's own strongest version of the thesis is not the weakened one. "Two people really do need to think an infinite vortex of 'I know that he knows that I know that he knows…' thoughts to be sure they will coordinate their plans. Two or three, or any finite number, aren't enough." Set that beside pages 56 and 57 and the two pull in opposite directions. Both are the book's words. We are not going to settle the tension by quoting whichever half suits the paragraph we are writing.
Writing in Science in October 2025, Jeremy Goodman and Chaz Firestone reviewed the book generously and then disagreed with it twice. They call the book excellent, say it "brings psychology into the fold", and credit it with anticipating and rebutting numerous objections.
Their first argument is that common knowledge in the endless sense may be unreachable. Two neighbours have weather apps accurate to within a degree. At 34 degrees neither can rule out that the other's app reads 33, and the same gap reopens at every temperature, so even at 60 degrees the pair lack common knowledge that it is above freezing. The philosophical version of that argument is Harvey Lederman's, in Mind in 2018.
The second argument is sharper and does not depend on the first. Coordination may not need common knowledge at all. "If Alice's and Bob's apps each read 60°, they will surely meet." Their verdict is that such demonstrations "fall short of establishing its central thesis: that common knowledge is the key to coordination."
Pinker's own group has conceded part of the first point. A 2026 paper in the Proceedings of the National Academy of Sciences led by Julian De Freitas opens by granting that common knowledge is theoretically required for coordination, "But a finite mind cannot represent infinitely embedded propositions, and even a few are cognitively taxing."
Ask what the experiments put in people's heads
The empirical core of chapter 4 is a study led by Kyle Thomas and published in 2014, with Pinker as its fourth author. Around 4,000 people recruited online played a version of the stag hunt, a game in which the best outcome needs both players to take the risk. Work alone for a sure dollar, or try to make hot dogs together for slightly more, which pays nothing if your partner plays safe. Information arrived four ways: privately, told that your partner knew too, told that your partner knew you knew, or announced over a loudspeaker so that the whole room heard it.
More people tried to cooperate when the price was announced publicly than in any private or shared condition, at every payoff level. Being told the partner knew beat being told nothing in three of the four payoff conditions. Being told one rung higher than that made no reliable difference at all.
The result that cuts against a tidy reading is inside the same paper. The authors report that their findings were inconsistent with a strict rational-actor account, because the share of people who tried to work together barely moved across payoff conditions even as the relative value of cooperating rose tenfold. People responded to publicness and largely ignored the money. A 2019 review from the same group proposes that nobody represents nested propositions at all, and that people instead register "a distinctive cognitive state corresponding to the sense that something is public, unignorable, or 'out there.'"
Pointing the other way, a preprint from 2025 by Thomas Bolander and colleagues, not yet peer reviewed, ran 802 people through a coordination game built so that the useful fact could never become common knowledge. Players behaved as though it had, at shallow depths, and lost money for it.
Read the 1998 regression before quoting it
Here is what the working paper actually contains, which is the part that almost never travels with the claim.
1. The sample is 119 brands advertised on the three major American networks in October 1988, February 1989 and July 1989, giving 357 monthly campaign observations, of which 18 brands were classed as social goods and 101 were not. Thirteen of the 18 are beers.
2. The outcome is not sales and not persuasion. It is the average cost per thousand households and the average audience size, taken from Nielsen's estimates of what a slot cost, which are figures reported by the networks rather than records of what changed hands.
3. Before controls, social brands averaged about ten dollars twenty per thousand households against about five dollars forty, and audiences of 7.1 million households against 5.6 million.
4. With more than forty demographic categories held still, the premium for a social good falls to roughly a dollar twenty per thousand, about 20 per cent. Adding a measure of four-week reach barely moves it.
5. No Super Bowl data appear in the regressions anywhere. The Super Bowl is the illustration, alongside Apple's 1984 advertisement, the Discover card and Master Lock.
The author's own caveats are more severe than anything a critic has said since. He calls the result "at this point more suggestive than definitive because of data limitations", notes that his social-good sample is "dominated by male oriented products and beer in particular", and states that "there remains the possibility that instead of describing a social good effect, we are simply describing a beer drinker effect." He concedes that social goods are defined "in a very ad hoc way", and he names a rival explanation without dismissing it: a person may buy a thing because others know about it, rather than because he expects others to buy it too.
The author of the only regression behind the Super Bowl claim wrote that it might be a beer-drinker effect. Thirteen of his eighteen social brands were beers.
Weigh the Super Bowl studies for what they measure
Two peer-reviewed papers have since put real Super Bowl data to work, and neither tests common knowledge.
A 2017 study led by Seth Stephens-Davidowitz used the fact that advertising slots sell out months before anyone knows which teams will qualify, which makes local viewership a lucky accident. Across 70 films advertised between 2004 and 2014, home cities of qualifying teams saw about 20 per cent more households watching, and that extra audience raised release-week searches and opening-weekend takings, worth roughly 21 per cent more tickets at an average Super Bowl audience. Matched films that were not advertised showed no such pattern. The design measures the return on extra reach, and because the lever is simply more local people watching, it cannot separate common knowledge from ordinary extra exposure.
A 2018 paper led by Wesley Hartmann went further and tested a mechanism on beer and soda sales. Pre-game sales do reveal which brands people buy to drink during the game, but that had nothing to do with whether a brand ever advertised in the Super Bowl, which the authors say rejects the theory that the advertising works as a complement to consumption. The effects they did find sat in weeks containing later sporting events, and basketball viewership data confirmed the prediction. That is a learned-association account rather than a coordination one, and it is the only mechanism anyone has tested on Super Bowl sales.
### What this evidence cannot settle
No experiment has manipulated publicness in advertising. We searched for one and did not find it. The two field studies vary how many people in a city saw the advertisement, which is not the same variable. The 1998 regression is unpublished, is about what advertisers paid rather than what buyers did, and its author flags the beer problem himself. The coordination experiments are mostly from one research group, with Pinker among the authors, and we found no independent replication of the butcher-and-baker study. Measuring advertising returns is brutally hard in general: across 25 large field experiments, Randall Lewis and Justin Rao found the typical range of plausible returns on investment wider than 100 percentage points.
Everyone knowing that everyone knows is a beautiful explanation for the price of a Super Bowl slot, and a beautiful explanation is the easiest kind to stop checking.
Common questions
Do Super Bowl ads work by creating common knowledge?
No, or at least nobody has shown it. Michael Chwe's 1998 working paper found that brands selling publicly consumed goods paid around 20 per cent more per thousand households, across 119 brands on 1988 and 1989 network television, with no Super Bowl data in the regressions. He called the finding more suggestive than definitive and noted that 13 of his 18 social brands were beers, so it might be a beer-drinker effect.
What does common knowledge actually mean here?
In the strict sense Robert Aumann formalised in 1976, it means a fact everyone knows, everyone knows everyone knows, and so on endlessly. Steven Pinker's 2025 book narrows this at pages 56 and 57 to common p-belief, highly confident common belief rather than certainty, and uses the weaker term for the rest of the book.
Do Super Bowl ads work?
For films, yes, by raising reach. A 2017 study led by Seth Stephens-Davidowitz found that cities with unexpectedly high viewership delivered roughly 21 per cent more opening-weekend tickets. Why they work is open: the only paper to test a consumption mechanism on Super Bowl beer and soda sales, led by Wesley Hartmann in 2018, rejected it and landed on learned association with sport instead.
Sources
Chwe, M. S.-Y. (1998). 'Believe the Hype: Solving Coordination Problems with Television Advertising.' Unpublished working paper, Department of Economics, University of Chicago, January 1998. Full text read at the author's own copy, chwe.net/michael/h.pdf. Working paper, not peer reviewed.
Chwe, M. S.-Y. (1998). 'Culture, Circles, and Commercials: Publicity, Common Knowledge, and Social Coordination.' Rationality and Society, 10(1), 47-75. doi:10.1177/104346398010001002. Crossref deposited abstract read; full text paywalled and not reached.
Chwe, M. S.-Y. (2001). Rational Ritual: Culture, Coordination, and Common Knowledge. Princeton University Press; paperback reissue 2013. doi:10.23943/princeton/9780691158280.001.0001. Publisher's abstract read at Crossref. A monograph, no new experiment.
Thomas, K. A., DeScioli, P., Haque, O. S., & Pinker, S. (2014). 'The psychology of coordination and common knowledge.' Journal of Personality and Social Psychology, 107(4), 657-676. doi:10.1037/a0037037. Full text read, author manuscript at Harvard DASH; author order checked at Crossref.
De Freitas, J., Thomas, K., DeScioli, P., & Pinker, S. (2019). 'Common knowledge, coordination, and strategic mentalizing in human social life.' PNAS, 116(28), 13751-13758. doi:10.1073/pnas.1905518116. Full text read at PMC6628641. A review of the group's own experiments; PNAS Contributed submission with named reviewers.
De Freitas, J., Huang, Y., & Pinker, S. (2026). 'Recursive mentalizing and public salience in the perception of common knowledge.' PNAS, 123(33), e2536294123. doi:10.1073/pnas.2536294123. Abstract read in full at Crossref; open access. Sample sizes not stated in the abstract and not used.
Goodman, J., & Firestone, C. (2025). 'Knowledge for two.' Science, 390(6770), 246. doi:10.1126/science.aeb1081. Full text read at the authors' copy, perception.jhu.edu. A book review, no data of its own.
Lederman, H. (2018). 'Uncommon Knowledge.' Mind, 127(508), 1069-1105. doi:10.1093/mind/fzw072. Bibliographic record read at Crossref; abstract not deposited and full text not reached, so characterised only as the Science review characterises it.
Bolander, T., Engelhardt, R., & Nicolet, T. S. (2025). 'The Curse of Shared Knowledge: Recursive Belief Reasoning in a Coordination Game with Imperfect Information.' arXiv:2008.08849v3, 11 November 2025. Abstract read in full at the arXiv listing. Preprint, no peer-reviewed version found.
Stephens-Davidowitz, S., Varian, H., & Smith, M. D. (2017). 'Super returns to Super Bowl ads?' Quantitative Marketing and Economics, 15(1), 1-28. doi:10.1007/s11129-016-9179-0. Full text read, author copy at cmu.edu.
Hartmann, W. R., & Klapper, D. (2018). 'Super Bowl Ads.' Marketing Science, 37(1), 78-96. doi:10.1287/mksc.2017.1055. Abstract read in full, Crossref deposited abstract.
Lewis, R. A., & Rao, J. M. (2015). 'The Unfavorable Economics of Measuring the Returns to Advertising.' Quarterly Journal of Economics, 130(4), 1941-1973. doi:10.1093/qje/qjv023. Abstract read in full at the OpenAlex record.
Pinker, S. (2025). When Everyone Knows That Everyone Knows... Common Knowledge and the Science of Harmony, Hypocrisy and Outrage. Allen Lane, first British printing, ISBN 978-0-241-61883-7. Read through in the vault; all page numbers are this edition.
Aumann, R. J. (1976). 'Agreeing to Disagree.' The Annals of Statistics, 4(6), 1236-1239. Named only as the origin of the formal object the seed book narrows away from at pages 56 to 57.