Attention economy

The attention economy is the allocation of human attention as a scarce resource across competing demands — and, in its current form, the industry whose dominant business model is to capture attention in order to sell advertising against it. The concept predates the internet (Herbert Simon’s 1971 observation that “a wealth of information creates a poverty of attention”), but the contemporary version is defined by algorithmic short-form video and engagement-optimized feeds, whose success is measured in time-on-device rather than user welfare.

The revealed-preference puzzle

Matthew Yglesias frames the modern attention economy as a revealed-preference trap: by what people actually do, video “absolutely felicity-moggs” every prior medium, yet the result “doesn’t make for a very appealing vignette about modern life.”¹ The dominant use of the internet’s abundance is algorithmic short-form video (Reels, TikTok, YouTube Shorts), and the largest technology companies are, in practice, vendors of streaming video and the devices it plays on. Yglesias’s provocation: the 75-year “rolling moral panic” about television was essentially correct — television displaced exactly the communal, social-capital-building time that Putnam documented in Bowling Alone, and the smartphone generalizes that substitution. See trust-and-social-capital for the Putnam cohort evidence and abundance-agenda for the supply-side frame this critiques.

Scitovsky supplied an earlier account of this tension. His creative-consumption framework distinguishes immediate comfort from stimulation that requires cultivated skill; saving time and effort does not by itself teach people how to enjoy the resources recovered. Habituation and low entry costs can make comfortable activity self-reinforcing while its stimulation value declines. This is a mechanism to test against the evidence below, not a substitute for it.²³

The empirical record: two experiments and a panel null

Two large randomized experiments give the puzzle quantitative teeth; a five-wave panel study complicates the reading.

Deactivation improves emotional state. Allcott, Gentzkow, and co-authors ran two large randomized experiments before the 2020 U.S. election in which participants deactivated Facebook or Instagram for six weeks.⁴ Deactivating Facebook produced a 0.060 standard-deviation improvement on an index of happiness, depression, and anxiety (relative to controls who deactivated for only the first week); deactivating Instagram produced a 0.041 SD improvement. The effects are statistically significant (p < 0.01). Exploratory cuts: the Facebook effect concentrates in users over 35, the Instagram effect in women under 25. People who leave feel better — yet they largely return.

The collective trap. Bursztyn, Handel, Jimenez, and Roth (“When Product Markets Become Collective Traps”) separate individual consumer surplus from product-market surplus by eliciting both willingness-to-accept (WTA) to deactivate and willingness-to-pay (WTP) to deactivate the entire market.⁵ On the standard individual measure, users derive large positive surplus — averaging $55 (TikTok) and $47 (Instagram), with 92% / 86% of users positive. But accounting for the externality that non-use imposes (being the only one off the network is costly), the picture inverts: a large share of users experience negative welfare from the product’s existence, and users would on average pay $24 (TikTok) / $6 (Instagram) to have the whole product market shut down. This is the product-market trap: it is individually rational to use a product you would collectively pay to abolish — the same structure as an arms race or a positional good.

The panel null. Kushlev, Moon, Motyl, Fast, and Schroeder (Nature Human Behaviour, 2026) ran the observational counterpart: five waves of the Understanding America Study panel (N = 1,966 at wave 1, mean age around 40), self-reported use frequency of ten social technologies measured every three months, analyzed with Bayesian and frequentist random-intercept cross-lagged panel models (RI-CLPM).⁶ Within person, nothing predicts anything: no technology’s use — texting and video calling included — preceded a change in life satisfaction, with posterior probabilities of exceeding the ±0.10 region-of-practical-equivalence bound at or near zero for every use-to-satisfaction coefficient. The reverse direction is nearly as quiet: increases in life satisfaction preceded modest increases in video calling (FaceTime; β = .093 in women) but no scrolling behavior. Between people, the familiar pattern does appear — heavier texters are more satisfied (r = 0.134), heavier YouTube (r = −0.107) and TikTok (r = −0.106) users less so — but between-person comparison is exactly what the RI-CLPM design strips out, so those associations read as selection rather than dose. The peer review file shows the claim being trimmed in public: the submitted title asserted a one-sided null (“Social Technology Use Is Not Associated with Greater Subsequent Life Satisfaction”), and reviewers — after an earlier review round at Nature Communications — forced a neutral title, a ±0.10 smallest-effect-size threshold, and conclusions restricted to multiple-comparison-corrected results, with one reviewer warning that “superficial readers, journalists and LLMs” would misread the abstract as a flat no-effect verdict.

Reconciling the designs. The experiments and the panel answer different questions. Deactivation measures the extensive margin — removing a heavy default during six unusual weeks around an election — while the panel measures ordinary use frequency at three-month lags, where within-person use barely moves (the authors name this temporal stability as their main detection limit). The collective-trap model predicts the individual-level null: if the welfare damage is a market-level externality, individual dose-response can be flat while the product’s existence remains a net negative. What the panel undercuts is the simpler between-person reading in both directions — “screens are fine” (users report equal or higher satisfaction) and “screens are poison” (heavy video users are less satisfied) — since neither survives the within-person comparison.

Policy levers and their shape

Yglesias draws the policy implication explicitly: not prohibition (“we’re not going to un-invent the smartphone”), but rebalancing the tax burden — taxing digital advertising revenue and other parts of the “compulsive video industry” while shifting the burden off activities that build social capital (his example: making it “cheaper and easier and more lucrative” to bowl, dine out, or exercise than to run a “glowing screen, home alone” business).⁷ The logic mirrors Pigouvian taxation of cigarettes: if the externality is real (the collective-trap evidence says it is), the tax nudges the margin without banning the behavior. The Bursztyn et al. framing sharpens why individual willpower fails — the trap is structural, so the remedy is structural (price, default, or market-shape), not exhortation.

Open questions

• Effect durability. Do the deactivation gains persist past six weeks, or does hedonic adaptation / substitution to other screens erode them?
• Generalizability. Both NBER studies are U.S. samples around the 2020 election; does the collective-trap result hold for non-student, non-U.S. populations?
• Subscription vs. ad models. Does shifting platforms from advertising to subscription (as Yglesias suggests) actually reduce compulsive use, or merely re-price it?
• Regulation of the externality. If the harm is a collective externality rather than individual overuse, does that argue for market-level intervention (ad taxes, design mandates) over screen-time tools aimed at individuals?
• Intensive vs. extensive margin. Ordinary-frequency panels show within-person nulls while deactivation experiments show gains — is the welfare-relevant margin how much one uses, or whether one (and everyone else) is on the network at all?

Related pages

• creative-consumption — skill, novelty, and habituation as a mechanism behind the revealed-preference puzzle
• trust-and-social-capital — the social-capital stock the attention economy draws down
• abundance-agenda — the supply-side celebration this critiques on the demand side
• infinite-game-cooperation — the repeated-interaction substrate that communal time feeds
• moral-economy-price-tag — another case of a market mechanism colliding with a non-market good
• bullshit-jobs — a rival account of modern work/consumption malaise
• post-literacy-debate — the civilizational-decline framing of the same screen ecosystem, and the education-system counter-evidence
• llm-context-engineering — the machine instance of attention scarcity: the transformer’s finite attention budget, spent token by token

Sources

• Matthew Yglesias 2026 — The real reason nobody feels good about technological progress
• Hunt Allcott, Matthew Gentzkow, et al. 2026 — The Effect of Deactivating Facebook and Instagram on Users’ Emotional State
• Leonardo Bursztyn, Benjamin R. Handel, Rafael Jimenez, Christopher Roth 2024 — When Product Markets Become Collective Traps: The Case of Social Media
• Kushlev, Moon, Motyl, Fast, Schroeder 2026 — Social technology use and life satisfaction in a five-wave panel study of US adults — Peer Review File
• Marina Bianchi 2003 — A questioning economist: Tibor Scitovsky’s attempt to bring joy into economics
• Maurizio Pugno 2014 — Scitovsky, Behavioural Economics, and Beyond

Footnotes

1. Matthew Yglesias 2026 — The real reason nobody feels good about technological progress ↩

2. Marina Bianchi 2003 — A questioning economist: Tibor Scitovsky’s attempt to bring joy into economics ↩

3. Maurizio Pugno 2014 — Scitovsky, Behavioural Economics, and Beyond ↩

4. Hunt Allcott, Matthew Gentzkow, et al. 2026 — The Effect of Deactivating Facebook and Instagram on Users’ Emotional State ↩

5. Leonardo Bursztyn, Benjamin R. Handel, Rafael Jimenez, Christopher Roth 2024 — When Product Markets Become Collective Traps: The Case of Social Media ↩

6. Kushlev, Moon, Motyl, Fast, Schroeder 2026 — Social technology use and life satisfaction in a five-wave panel study of US adults — Peer Review File ↩

7. Matthew Yglesias 2026 — The real reason nobody feels good about technological progress ↩